One to one marketing management via customer complaint

doi:10.21203/rs.3.rs-2716680/v1

Download PDF

Research Article

One to one marketing management via customer complaint

https://doi.org/10.21203/rs.3.rs-2716680/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 05 May, 2023

Read the published version in Social Network Analysis and Mining →

You are reading this latest preprint version

For companies to retain customers and ensure effective management-level resolution, they need to anticipate customer churn and determine the root cause of complaints. To achieve this, analyzing personalized complaints from the customer's perspective is crucial. This research advocates for a multidisciplinary approach that combines language behavior, relevance feature extraction, feature weighting, and sentiment analysis to extract the underlying problem in real time. Applying this approach to the CFPB database sample yielded an accuracy rate of 82% and a system validity of 75%, which can help improve customer service and protect consumers in the financial and other service industries. By addressing individual customer issues that cause dissatisfaction, businesses can enhance customer satisfaction and retention levels. Thus, by analyzing complaints from a personalized standpoint, companies can identify the root cause of the problem, improve their services, and establish stronger customer relationships.

Root cause

Customer complaint

One to one marketing

Sentiment analysis

Customer retention

In recent years, product or service industry has undergone significant transformations as a result of technology and globalization-induced increases in competition and demand for more efficient and convenient services. For example, financial intermediaries, banks play a crucial role in converting savings into investments and efficiently managing loans. Stronger banks inspire greater customer confidence, which is essential for generating returns. Customer service is vital to retaining customers, as marketing principles are used to meet the needs of all customer groups. The SERVQUAL scale (Parasuraman et al., 1988) is a commonly employed instrument for measuring service quality. It consists of five factors: tangibles, reliability, responsiveness, assurance, and empathy. The tangible factors pertain to the precision, security, consistency, and speed of financial transactions, whereas the reliability factor relates to the timely delivery of services. Customer satisfaction hinges on responsiveness and attention to problem-solving, comprehensive service from employees, and prompt verification and correction of errors. Assurance is a feeling of safety in a transaction, competitive loan rates, interest rates on savings and time deposits, and pleasant employees. In addition, empathy is how employees seek to resolve customer issues.

To comprehend the actual issues behind customer complaints, it is crucial to have an individual understanding of each customer's unique problems, especially when dealing with a large number of customers. This personalized approach helps in developing specific strategies to address their concerns effectively. One-to-one marketing is a personalized approach to engaging with customers by recognizing that each customer is unique and requires a customized marketing strategy. On the other hand, customer complaints are essential for identifying the underlying problems that lead to customer dissatisfaction. By using customer complaints as an opportunity to gain insights into individual customers, companies can better understand their needs and preferences, which helps in creating personalized solutions to their problems. This approach enables companies to improve their customer service and build stronger relationships with their customers. One-to-one marketing management via customer complaints can also help businesses identify areas for improvement in their products or services and make necessary changes to better meet customer needs. In summary, one-to-one marketing management via customer complaints is a customer-centric strategy that prioritizes individual customer needs and can lead to improved customer satisfaction and loyalty (Aspinall et al., 2001).

Previous research relates to document for knowledge mining such as Griffiths et al. (2007) and Akash et al. (2013) utilized techniques such as principal component analysis to overcome challenges associated with high-dimensional, sparse, and redundant data. Similarly, Kaveh et al. (2016) applied a Latent Dirichlet allocation (LDA) probabilistic approach to the CFPB database, allowing for the clustering of consumer complaints into multiple categories beyond predefined annotations. Nadda et al. (2022) used mutual information to measure the correlation between keywords and classes. Most prior work has focused on frequency dimensions without considering language semantics (Griffiths et al., 2007; Akash et al., 2013).

There are several limitations to using model-based prediction to identify customer pain points. Firstly, the artificial model is constructed using supervised data or a fixed feature vector structure, resulting in a machine learning or statistical model that has limited flexibility. Secondly, only significant features are considered for prediction, and the number of features in the model is restricted, making it difficult to identify dissatisfaction pain points through narrative complaints. Ultimately, analyzing customer language and behavior may be necessary to identify individual pain points and resolve issues. Finally, the authors suggest a simple way to individualize and monitor customer churn in real-time to detect the root cause of complaints. Additionally, our reasoning method integrates financial event-oriented information and rough set theory to create purpose factors utilizing an enhanced semantic feature clustering concept. Our framework is efficient and effective in solving feature and time-consuming issues.

One to one marketing and customer retention

In the context of complaint management in the financial services sector, one-to-one marketing can be an effective way to resolve customer issues and improve retention. By offering personalized attention to customers who have lodged complaints, businesses can demonstrate their commitment to resolving issues and providing excellent service. For example, a bank that assigns a dedicated customer service representative to handle a customer complaint can create a sense of individualized attention and care that can help diffuse a potentially negative situation (Kotler, 2003). By taking the time to understand each customer's unique concerns and providing personalized solutions, businesses can not only resolve complaints but also build trust and loyalty with their clients (Ang & Buttle, 2006). In a highly regulated industry like finance, where compliance is essential, one-to-one marketing can be a powerful tool for maintaining customer satisfaction and retention while ensuring regulatory compliance.

Furthermore, in today's digital age, where customer feedback and reviews are widely available online, effective customer complaint management can also have a positive impact on a bank's online reputation and search engine rankings. Positive customer feedback and reviews can improve a bank's visibility and credibility, leading to more customers and a higher retention rate. In conclusion, customer complaint management plays a significant role in the financial industry, particularly in terms of customer retention. By effectively managing complaints, financial industry can improve customer satisfaction, foster loyalty, and ultimately increase customer retention. Previous research by Payne et al. (2001) revealed that only 23% of UK organizations' marketing budgets were allocated to customer retention. This is contrary to the findings of Aspinall et al. (2001), who discovered that 54% of businesses viewed customer retention as more important than customer acquisition.

The issue of customer retention in the financial industry can be supported by various theories. One widely used theory is the Service-Profit Chain theory, developed by Heskett et al. (1997), which posits that employee satisfaction, customer satisfaction, and financial performance are directly linked. According to the theory, when employees are satisfied and engaged, they provide superior customer service, resulting in greater customer satisfaction and loyalty. This, in turn, increases the company's revenue and profits. Another theory that can be applied to support customer retention in the financial industry is the Expectancy Theory of Motivation, developed by Vroom (1964), which suggests that customers are more likely to remain loyal if they expect the service unit to meet their needs and provide desired outcomes such as good service and competitive products. Another theory is the Relationship Marketing Theory developed by Grönroos (1990), which posits that successful companies focus on creating and maintaining long-term relationships with customers. In the context of customer retention, this theory suggests that banks should focus on building strong, long-term relationships with customers by providing high-quality products and services and addressing their needs and concerns in a timely and effective manner.

When previously repeat or subscription customers stop doing business with any firm, this is referred to as customer attrition. Maintaining customer retention is a crucial component of customer relationship management, as it is more cost-efficient to retain customers than to acquire new ones. Research has shown that the cost of acquiring a new customer can range from 5 to 25 times more expensive than retaining one (Aspinall et al., 2001, Kotler, 2003). The loss of recurring revenue is not the only impact of customer churn, as it can also result in deactivation costs, reduced brand value, missed cross-selling opportunities, and a potential ripple effect of other customers leaving. Industries like telecommunications are particularly vulnerable to customer churn, which can be mitigated through proactive measures such as anticipating customer behavior and engaging with them before they cancel or using targeted communication and incentives. The biggest challenge in preventing customer churn is accurately identifying at-risk customers and understanding the root causes for effective management.

The customer journey, a commonly used concept in direct marketing, involves identifying key experiences in the customer life cycle (Følstad et al., 2013, Dove et al., 2016). To understand customer interactions with a brand or service, marketers use the customer journey map (CJM), which includes touchpoints, paths, stages, and emotions such as satisfaction or dissatisfaction (Bougie et al., 2003). Dissatisfaction, which can occur when a customer receives service below their expectations, can lead to complaints being filed with the brand owner or government agency (Danet, 1978) and can also be a reason for customers to switch to other brands (Manrai & Manrai, 2007). Root-cause analysis is a method used to identify the underlying causes of customer complaints, in order to prevent them from recurring. The process of identifying root causes, on the other hand, can be complex, involving multiple factors and, in some cases, failure analysts identifying non-causal root causes. The detection of root causes in complaint messages is further complicated by barriers such as language, data cleaning, and text processing. Additionally, the amount of qualitative data has grown exponentially due to technology and communication (Feldman & Sanger, 2007, Akash et al., 2013). This study proposes a simple, semi-automatic approach to identify root causes based on data from the CFPB database that will be applicable to other industries.

Financial service and service gap

The financial services sector is essential for both consumers and businesses, offering various financial products and services, including banking, mortgage lending, credit cards, payment processing, tax management, accounting, and investment services. Retail banking offers individualized services, such as checking and savings accounts, loans, and credit cards. Commercial banking, on the other hand, provides tailored services to small businesses and corporations. Investment banking serves large corporations and governments, providing complex financial services and products. The banking industry generates profits through fees, commissions, and other means (Widyastuti & Armanto, 2013).

The financial services sector plays a vital role in people's lives and customer service is essential for successful interactions between consumers and businesses. In recent years, there has been a shift towards digital channels and online banking, but in-person interactions remain crucial for providing exceptional customer service. Customers seek banks that offer diverse levels of service for various banking activities, along with quality products. A strong commitment to customer service can enhance customer loyalty and attract new clients, making it a valuable investment for the financial services industry (Akhgari et al. ,2018, Omoregie et al., 2019)

The use of online customer service in the financial services industry has the advantage of being cost-effective, especially in epidermic crises (Chan et al., 2022), but it can also lead to significant pain points for customers. These include difficulties in protecting privacy, frustration and dissatisfaction with navigating menus or communicating with chat bots, a lack of personal touch, unequal access to support, and more. It is crucial for financial institutions to strive to bridge the gap in online customer service and provide all customers with the support they need in a timely and effective manner, regardless of their technological proficiency. By doing so, financial institutions can build trust and create positive customer experiences that lead to long-term customer satisfaction and loyalty.

Solving problems through computer menus in the banking sector has limitations as it may not fully address specific needs. Bot chat interactions lack the ability to understand human emotions and the nuances of a customer's issue (Ghamri, 2017, Crolic et al.,2022). When customers can describe their problem and experience using language, it provides a clearer picture of the issue and its impact. This allows the service unit to effectively resolve the problem immediately, leading to increased customer satisfaction and loyalty. When there are a lot of complaints, it's important to understand the root cause by analyzing the language. This keeps customers happy and keeps them using the services. The challenges of using a narrative customer compliance or journey pain point in this hypothetical root cause analyzer are not only to satisfy customer needs but also to maximize customer relationship management. Additionally, the crucial issue is a universal problem faced by consumers globally.

The study was conducted by using customer databases from the Consumer Financial Protection Bureau (CFPB). The CFPB was established in 2011 to protect the interests of consumers in financial markets. It receives complaints from consumers about various financial products and services and investigates these complaints. The complaint data is publicly available on the CFPB's website and is considered the largest public collection of consumer financial complaints. Previous studies have used the CFPB data to analyze customer complaints, including studies based on customers' demographics, government agency processing, and text mining (Litwin, 2015, Aggarwal & Zhai, 2012, Kaveh et al., 2016). Some studies, such as Salim et al. (2018), also investigated the effect of customer complaints on satisfaction and loyalty in other countries. For companies to come up with the best solution, they need to figure out why a product or service failed or made a mistake in the first place. Despite utilizing the CFPB database, this study finds that the preliminary results on American consumer issues are comparable to those globally. The same is true for financial service complaints in Thailand, including issues with invalid data and theft facilitated by employees. In 2021, 58,281 complaints were sent to the Financial Consumer Protection Center at the Bank of Thailand. These complaints can be broken down into different categories, such as product-based complaints, resolution time, miscommunication, and security control (Financial Consumer Protection Center, Bank of Thailand, 2021). Also, channels and interfaces for making complaints over the Internet are set up near the CFPB.

Customer compliant and root cause identification

Generally, but not exclusively, complaint behavior is one of the responses to perceived dissatisfaction in the post-purchase phase. The concept of "complaint behaviors" includes a broader terminology that also encompasses the notions of protest, communication (word of mouth), or recommendation to third parties, and even the notion of boycott (Crie, 2003). The UK business sector is estimated to incur an annual loss of £37 billion due to inadequate customer service and improper handling of complaints, as stated in a report by Ombudsman Services. Improper root cause analysis of complaints, which can stem from various factors such as product, service, or communication issues, can result in inadequate complaint management. (Crie, 2003).

Using narrative complaints to formulate a machine learning classification data structure can be challenging due to several barriers. Firstly, narrative documents tend to be unstructured, making it difficult to extract meaningful data and identify patterns. Machine learning algorithms require structured data to perform effective classification, which means that unstructured data must be transformed into a structured format before it can be used. This process of data cleaning and pre-processing can be time-consuming and may result in the loss of important information. Secondly, narrative documents may contain subjective information, which can be difficult to quantify and classify accurately. Machine learning algorithms may struggle to identify and classify subjective information, leading to inaccurate results. Finally, narrative documents may contain a large amount of irrelevant or redundant information, which can make it difficult to identify the key features that are relevant to the classification task (Blei et al., 2003, Coussement & Van den Poel, 2008). Addressing these barriers requires a combination of natural language processing techniques, expert knowledge, and careful data selection to ensure that the resulting classification model is accurate and effective.

Human language behavior

When analyzing customer complaint documents, businesses should consider key factors such as the focus point, repetition of pain points, extension, and document length. By identifying the primary issue and addressing it directly, businesses can alleviate the customer's primary concern and improve their satisfaction. Repeated pain points can help uncover underlying problems causing dissatisfaction, while the extension of the document provides a better understanding of the issue's full scope. The length of the document can also indicate the severity of the complaint and level of frustration expressed by the customer. By analyzing these factors, businesses can gain insight into customer concerns and take appropriate measures to improve overall customer satisfaction (Coussement & Van den Poel, 2008).

Feature selection

SERVQUAL and the customer journey were used in this study, and they were classified into five factors based on customer compliance: (1) communication (CO), (2) product (PD), (3) process (PR), (4) impact (IM), and (5) object (OB). All activities to transfer or exchange information to support an effective service activity are referred to as "communication." The factor covers bank employees' and customers' communication via face-to-face, telephone, or leaving mail messages or chatbots, and communication also includes network robustness, personality traits, service mind, etc. (Parasuraman et al., 1988, Salim et al., 2018). In addition, communication can be evaluated based on the time required to address reported customer problems and the response time after a customer has submitted a service request. Products refer to the price and quality of banking products, while "process" (bank operation) refers to the reliability of the trust in the company’s ability to perform service in a proper way, such as acting according to promises and declarations (Akbar & Parvez, 2009). Reliable service means the financial transactions must be audited, controlled, and updated without error. In addition, the system must be protected against theft and violation of personal data (Rumler & Waschiczek, 2010). Another two factors, "Impact" and "Object," are quite different from SERVQUAL due to the customers’ bank journey pain point and the object (customers’ family or employees or agents or banks) that’s related to the pain point (Andrews & Eade, 2013, Dhruv & Anne, 2020). The dissatisfaction caused by receiving service below customer expectations is called a "pain point" (Dhruv & Anne, 2020). This led to the concept of the factor "Impact" in this study. Examples of customer pain points are disputes, wasted time, identity theft, bankruptcy, and fraud. The final factor, "Object," means the person, employee, or entity involved in the incident in the "Customer Complaint Document." The authors used the purpose factors CO, PD, PR, IM, and OB to tackle the root causes of the problem and correct or prevent those problems from recurring (Corey & Friedman, 2022).

Rough set theory

Rough set theory, first introduced by Pawlak in 1982, is a formal mathematical framework used to analyze data and uncover structural relationships within imprecise and noisy information. It can be applied to relational databases and has been used in database mining and knowledge discovery (Pawlak, 2002). Under incomplete information, "rough set" is an approach for approximation decision making (Pessoa & Stephany, 2014). The theory is based on the logical properties of information systems and is related to fuzzy theory. It defines two disjoint classes of attributes: conditional (C) and decisional (D), and the system is called a decision table denoted by S = (U, C, D), where U is the relation function (decision rule) between C and D. In this approach, we use a feature selection process in which three experts select all words from the experience message that pass a 5% threshold and identify them as elements in the strong word set (C). STW = (stw₁, stw₂, …, stw_t) and weak word set as WTW = (wtw₁, wtw₂, …, wtw_m) are the resulting sets, and the relation function U = (u₁, u₂, …, u_h) is defined as u_i: Pr(wtw_i) < Pr(wttw_i | u_i (stw_j)) , = 1,2,.., t, with threshold. U is identified by posterior probability.

Consequently, the system was conducted by classifying strong words and weak words into subgroups based on the concept of financial service as: STW = (CO, PD, PR), WTW = (IM, OB), where U is ruling relation set of WTW ® STW.

Theory and model supporting

The SERVQUAL model is the main framework supporting this study, which is based on the customer's assessment of service quality. SERVQUAL is considered a multidimensional construct, with ten components, including reliability, responsiveness, competence, access, courtesy, communication, credibility, security, understanding/knowing the customer, and tangibles (Parasuraman et al., 1988). The model is built on the notion that the customer's assessment of service quality is crucial and is measured as a gap between what the customer expects from a class of service providers and their evaluations of the performance of a specific provider. This model is essential for service fault recovery management, as it highlights the key factors that contribute to customer satisfaction and helps identify areas for improvement.

Total Quality Management (TQM) aims to improve all aspects of a business, including processes, products, and services, through the involvement of all employees and a focus on data-driven decision making (Prajogo & Sohal, 2006). Root cause analysis is a key component of TQM, as it helps organizations identify the root causes of problems rather than just treating symptoms. By understanding the root causes, organizations can make meaningful and lasting improvements to their processes and systems to prevent similar problems from happening again in the future. This leads to increased customer satisfaction and improves overall business performance (Sila & Ebrahimpour, 2003).

Communication theory, first proposed by S. F. Scudder in 1980, states that all living beings engage in communication, though the methods may differ. Communication is crucial for self-expression as it enables individuals to share their thoughts, emotions, and information with others. From a psychological perspective, communication encompasses more than just the transfer of information from the sender to the receiver, it also includes the sender's thoughts and feelings and the receiver's reactions and emotions. Therefore, communication is not just the flow of information, but also the shared thoughts and emotions of the sender. It also includes the reactions and feelings of the receiver after he decodes the information. When a customer encounters a defect in the product or poor service, it can lead to customer dissatisfaction (Bougie et al., 2003). These incidents befall consumers, who are faced with the options of communicating complaints to influence the service delivery process, receiving compensation, or terminating the service exchange without having their service expectations met in a satisfactory manner (Singh, 1988).

Argumentation theory is relevant to communication theories that explain how to formulate complaints that communicate effectively with company employees and yield compensation for the consumer (Singh & Wilkes, 1996, Matusitz & Breen, 2011). Complaints can be drafted either hastily or carefully, and prior knowledge of complaint procedures can aid in effective complaint-making. However, a better understanding of communication theories can enhance the development of effective consumer complaints. Argumentation theory suggests that effective arguments should be based on thorough investigation, supported by well-designed sub-arguments, and presented concisely with only relevant and persuasive information. This principle was applied in the study by determining the weight of the text (word and sentence level) in complaint documents to identify the root cause (Corbett & Connors, 1999, Miller & Levine, 1996).

The Study Framework

The process of text processing, also known as "text mining," is conducted in multiple stages. The data is prepared by breaking the text into smaller units (tokenization), determining the frequency of word distribution, addressing acronyms and slang, returning words to their root form (stemming and lemmatization), removing common words, identifying patterns and word associations, and identifying key words. The text is then assigned to root-cause categories based on the frequencies and weights of the key words, and the results are validated with relevant stakeholders. The text mining framework for analyzing the root cause and performing Pareto analysis is illustrated in Figs. 1–3.

This study utilizes data from the CFPB database current as of 2021–2022. The database comprises 18 characteristics such as the submission date, the product, the issue relevant to the complaint, and the company that the complaint was directed to for a response. Additionally, the database includes a narrative provided by the consumer that describes the incident in question. The research specifically concentrates on analyzing these consumer complaint narratives, which were collected from 65536 narratives.

The text mining process begins by preprocessing the data and selecting 300 sample messages for testing analysis. These messages are then passed through a sentence segmentation process and analyzed for word and sentence weights using word and sentence analyzers. The results are then validated with relevant stakeholders. The example message, sentence, and word structure are meant to serve as a starting point for further investigation.

We proposed a rule for labeling each word in a sentence for further computation of weight, which was integrated from expert and rough set theory. The rule is as follows:

Rule 1: If Wordi is found in any of the sets CD, PD, or PR, then it is labeled to that set. (For example, if Wordi is found in CD, it is labeled to CD.)

Rule 2: If Wordi is found in any of the sets IM or OB and meets the condition uj 𝟄 U, then it is labeled to that set. (CD, PD, PR, IM, OB are disjoint sets).

The sequence of words in a sentence is then labeled with the corresponding group, as illustrated in Fig. 4.

The sentence is outlining a method for converting sentences into vectors for further examination. The process involves assigning each sentence a number, k, and each word within the sentence a feature that belongs to one of five groups. The sentence is defined as: St _k,m = (w_1,k,m, w _2,k,m,.., w _n,k,m); where n represents the last word in the sentence, k is the number assigned to the statement, and m is the message it belongs to.

Due to academic causal distance, the experimental data in this study is limited to five consecutive sentences per message (Monge et al., 1985; Lekham et al., 2020). Proximate causes refer to events that are closest to or directly responsible for producing an observed outcome (Monge & Kirste, 1980; Soto-Sanfie & Igartua, 2014). These causes differ from ultimate or distal causes, which are considered the underlying reasons for an occurrence. Additionally, due to the nature of human communication and communication theory (Walker et al., 2018), the topic sentence is considered the most significant sentence in a paragraph as it clearly defines the subject matter of the entire paragraph. This sentence is typically the first sentence of the paragraph, as it provides a summary of the sentences that follow (SEA; St. Louis Communication College). The authors have also assigned weighting values at both the word and sentence levels to aid in identifying root causes with the support of academic research. Previous studies, such as those conducted by Nadda et al. (2022), have employed mutual information (MI) to demonstrate the ability to classify individual keywords and measure the correlation between each keyword and class. The mutual information metric is represented by MI (w, c), where w represents a word and c represents a class set. It is calculated using Eq. 1.

$$MI(w,c)=\text{log}\frac{{f}_{A} .N}{{(f}_{A}{+f}_{C}{\left)\right(f}_{A}{+f}_{B})}$$

…

When ${f}_{A}$ is the number of documents in class c that contain word w, ${f}_{B}$ is the number of the documents not in class c that contain word w, ${f}_{C}$ is the number of the documents not in class c that do not contain word w. and N is the number of all documents. The MI (w, c) has a value in range [− log(N), log(N)]. Under the high feature dimension, this equation is pretty hard to understand, and it doesn't take into account how important location distance is in language. To solve the problem and save time, the authors use the following equations for the local (word) and global (sentence) weight levels:

${lw}_{m}$ $=(\frac{l}{l}, \frac{l-1}{l}, \frac{l-2}{l},\dots , \frac{1}{l})$ … (2)

Equation (2): ${lw}_{m}$ is a weight at word level in a sentence as m, and "l" is the sentence's word count.

${lst}_{j}$ $=(\frac{k}{k}, \frac{k-1}{k}, \frac{k-2}{k},\dots , \frac{1}{k})$ … (3)

Equation (3): ${lst}_{j}$ is a weight at sentence level and "k" is the sentence count ($1 \le k \le 5$).

Total weight of Gr_i𝟄D in a message = $=\sum _{i=1}^{no\_of\_sent}\sum _{j=1}^{no\_of\_word}{lw}_{ij}*{lst}_{i}$ … (4)

For example, i = 1 refers to the weight of CO in a message, while i = 5 refers to the total weight of OB in a message. D is the domain space. The root cause is identified by finding the maximum weight of each group (GR) in a message; GR = (CO, PD, PR, IM, OB); CO, PD, PR, IM, OB are mutually exclusive (Max_i𝟄GR).

Message Preprocessing, Word Selection And Computation

The experimental data used in the study was sourced from the Consumer Complaint Database (CFPB website) maintained by the Consumer Finance Protection Bureau (2021) in the United States. The database contains complaints about a variety of consumer financial products and services offered by banks and other financial institutions. The data is publicly accessible and does not require a license for use. Each complaint in the dataset includes attributes that can be used to uniquely identify it. The dataset used in the study was constructed by randomly selecting 65,536 records from the CFPB database that contain complaint messages between 2021 and 2022.

Word Extraction

The authors focused on analyzing the narratives of consumer complaints. To do this, they created the following subtask:

Subtask 1

Addressing acronyms, slang words, and misspelled words. The complaint messages often use slang words and acronyms, but they are written in various ways. For instance, some messages spell "knowledge" as "knowledg" or "know," and others use "pay" as an acronym for "payment."

Subtask 2

Text processing from the standpoint of information extraction, which entails extracting structured information from unstructured text. This process is applied in various applications, such as topic generation through co-occurrence clustering, knowledge extraction, and sentiment detection (Blei, Ng, & Jordan, 2003). The text preprocessing process may vary for different languages. For example, English text requires lemmatization, stemming, the removal of stop words, and punctuation (Jurafsky & Martin, 2000).

After the preprocessing, each sentence is represented as a word vector, denoted as St_i = (w_i,j ; j = 1, 2, …, last_word), where St_i is the i^− th sentence in a message and w_i,j is the j-th word in the i^− th sentence (the data format will look like Fig. 3).

Subtask 3

The dataset will be analyzed to identify the most frequent words. Only those words that appear more than 500 times will be retained for further analysis.

Subtask 4

Selecting relevance words and labeling by predefined rule set.

Subtask 5

Calculate the total weight of a message using Eq. 4, and determine the root cause of the message based on the criteria of Maxi𝟄GR.

Monitoring Churn To Improve Customer Retention

Sentiment and magnitude analysis play a crucial role in retaining customers in today's competitive business environment. By evaluating the sentiment, or the direction of feelings, expressed by customers, businesses can gain valuable insights into their level of satisfaction or dissatisfaction. In the same way, magnitude analysis can help a marketer figure out how strong these feelings are, whether they are good, bad, or neutral. Using these methods, businesses can identify potential "at-risk" customers and take proactive steps to improve their experience. This type of customer management is crucial in ensuring customer loyalty and preventing customer churn. The data from sentiment and magnitude analysis can be visualized to help understand patterns and trends. This makes it easier to figure out which groups of customers may need more attention (Radice, 2018; Luo et al., 2013).

Overall benefit, by incorporating sentiment and magnitude analysis into their customer management strategy, businesses can effectively retain customers and enhance their overall satisfaction. This kind of analysis tells businesses a lot about their customers' needs and preferences, which helps them engage and keep customers in a more personal way.

Words selection and construction knowledge set

The study analyzed 65,536 compliant messages that went through subtasks 1 to 3. Results showed that complaints had a minimum of 5 words and a maximum of 358 words per document, with an average of 185 words per document. After removing noise words, 8,535 words remained out of a total of 28,820. From these, 269 words with a frequency greater than 500 and relevance to this study evident were retained to identify to be strong and weak words. Based on three experts justify 54 words were select from 269 words and classify to be member in set CD, PD, and PR, while the remaining 215 words were take The relationship analysis is based on conditional probability (Posterior Bayes), and wi is considered a weak word if P (wi| CD, PD, or PR and ui exists) > P (wi| CD, PD, or PR and ui does not exist). An example of this can be seen in the sentence "I chatted with a representative after logging into my bank account regarding my options to ensure I protect my credit and bring my account current," where the probability of "representative" being relevant when "chat" is present and there is one (P (representative| chat, relevant) = 0.234) is higher than when "chat" is present and there is not one (P(representative| chat, not relevant) = 0.001). Thus, "representative" is considered a weak word in this context. This sample demonstrates how weak words are selected in the study. Another example is P (friend|chat, relevant) = 0.003; P(friend| chat, not relevant) = 0.035, which indicates that "friend" is not a weak word as the change of "friend" when predefined by chat and relevant is lower than when not relevant. By using this mechanism, the relevant words were extracted and presented in Table 1.n to the relation of element in set CD, PD, and PR.

Table 1

A list of sample words extracted from customer complaints and organized into predefined groups.
Communication (CO) 15 elements	Product (PD) 12 elements	Process (PR) 27 elements	Impact (IM) 28 elements	person/org (OB) 16 elements
report	credit	payment	stress	bank
inform	account	consume	dispute	agent
call	loan	request	time	staff
receiv	debt	collect	day	representative
tell	card	charge	month	creditor
state	mortgage	remove	year	court
letter	money	service	law	supervisor
:	:	:	:	:

To determine the feature set and its elements related to reliability from the review synthesis, the authors utilized confirmatory factor analysis and conducted a KMO and Bartlett test to assess all available data. A KMO value greater than 0.5 and a significance level below 0.05 for the Bartlett's test indicate significant correlation in the data (Vogt & Johnson, 2011). The collinearity of variables shows the level of correlation between a single variable and other variables. And the results are listed in Table 2.

Table 2

Evaluation of the reliability feature set and its components using confirmatory factor analysis, KMO, and the Bartlett Test
Factor	KMO	Bartlett's Test of Sphericity			Factor score
		Approx. Chi-Square	df	Sig	Minimum	Maximum
CO (15 elements)	0.593	1016.916	105	0.000	0.453	0.652
PD (12 elements)	0.530	560.615	66	0.000	0.431	0.756
PR (12 elements)	0.624	2707.820	351	0.000	0.458	0.665
IM (28 elements)	0.558	2523.963	378	0.000	0.398	0.785
OB(16 elements)	0.548	820.995	120	0.000	0.402	0.707

Sentence Segmentation

In Natural Language Processing (NLP), determining where sentences start, and end is called "paragraph division." We divide a paragraph into individual sentences, known as sentence segmentation. The authors used the spacy library in Python to carry out this NLP task. In Fig. 5, sample messages from records 1 was segmented into sets of sentences as shown in Fig. 6. The messages in records 1 was divided into six sentences, respectively. The system only selects the first five sentences of a message if it has more than five sentences.

The symbol set representation of record number 1 is represented in Table 2. The symbol "St_i,j" represents the sentence number i of message j.

Table 3

Word selection in each sentence using rule 1 and 2.
St_1,1	contac	supv	Friday	fall	payment
St_2,1	chat	representit	account	credit	account
St_,3,1	advic	extention	payment
St_4,1	review	finance	payment	Monday
St_5,1	represent	advic	payment	account	supv	loan
Note: The messages used to identify were passed through a preprocessing process in which certain words were adjusted, such as "supervisor" being shortened to "supv" and "advice" being shortened to "advic."

Labeling, Transformation And Matrix Manipulation

Table 3 shows how the system automatically selected the relevant words (from Table 2) and converted them into group names (CO, OB, IM, PR, and PD) based on their location in each sentence. In Table 3, the first column lists the sentence number, and the further columns list the number of words in each sentence that were labeled using predefined decision rules.

Table 4

Word labeling and location in each sentence
Sentence							No of words
St_1,1	CO	OB	IM	PR	PD		5
St_2,1	CO	OB	PD	PR	PD	PD	6
St_,3,1	CO	PR	PD				3
St_4,1	CO	PD	PR	IM			4
St_5,1	OB	CO	PR	PR	OB	PD	6

Root Cause Detection

The results of the calculation are presented in Table 4, with the final row displaying the total weight of each factor. The highest value of 14.83 in the "CO" column indicates that the primary cause of the compliance issue is "communication between the bank and customer." This can be further expanded by combining it with the second highest factor, "OB," and identifying it as "communication related to bank service units." However, when comparing this study to the label provided by the customer from the pull-down menu in the CFPB database, "Struggling to pay your loan," it is not clear how this label specifically relates to the identified root cause.

Table 5

The details and weight of five factors of a sample message.
Sentence (St_i)	CO (co_i)	PD (pd_i)	PR (pr_i)	IM (im_i)	OB (ob_i)	sentence weight (lst_i)	no word in a sentence
1	1	0.20	0.40	0.60	0.80	5	5
2	1	1.16	0.50	0.00	0.83	4	6
3	1	0.33	0.66	0.00	0.00	3	3
4	1	0.75	0.50	0.25	0.00	2	4
5	0.83	0.17	1.17	0.00	1.17	1	6
	14.83	8.3	8.15	3.5	8.49

Max_i𝟄GR (weight of CO, weight of PD, weight of PR, weight of IM, weight of OB)

weight of CO = 14.83 = > The root cause is the bank's communication.

Robustness Test

The robustness test was conducted on a sample size of 300 using systematic sampling (by time frame) from data that had passed the preprocessing step. This test aimed to assess the validity and reliability of the methodologies used in the study. The results of this test are presented in Table 5. It shows the experimental messages that have been passed through the sentence segmentation process and then processed by the root cause analyzer. The robustness test is important in ensuring the validity and reliability of the results obtained from the study, and it is a common practice in academic research. This test helps to ensure that the results of the study are not influenced by any external factors and that the results can be replicated in future studies.

Table 6

Shows the experimental messages that have been passed through the sentence segmentation process and then processed by the root cause analyzer.
No	Message	CO	PD	PR	IM	OB	Root Cause
1	I contacted Ally on Friday after falling behind on ……	14.83	8.30	8.15	3.50	8.49	Communication-related to Service unit
2	SOC SEC # DOB ADDRESS ………….	3.90	0.70	2.40	4.50	0.00	Impact related to Communication
3	Hello, This complaint is against …………..	11.00	12.30	8.35	1.60	0.00	Product related to Communication
4	I am a victim of Identity Theft & currently have an Experian account that ……….	17.56	4.44	4.71	13.67	1.67	Communication-related to Impact
5	Although I am checking for and addressing missing and or deficient aspects of ………	0.00	4.90	2.20	0.00	0.00	Product- related to Process
6	Two accounts are still on my credit history after contacting the credit bureau……	1.00	4.67	1.33	0.00	0.00	Product related to Process
7	I opened a City double cash card at the beginning and still have not received it. ………	9.00	8.25	9.92	3.33	0.00	Process related to Communication
8	In an attempt to purchase concert tickets communicated with an internet seller.	20.33	1.92	9.58	2.00	1.67	Communication-related to Process
9	I am attempting to sell my home and pay off my home loan with PHH Mortgage. ………	5.80	10.00	12.20	1.20	4.80	Process- related to Product
10	We are writing this letter to dispute a delinquent payment reported for 2018 under our GM ……….	12.83	10.33	13.58	1.00	0.75	Process- related to Communication.
:	:	:	:	:	:	:

According to the results of an analysis of 300 data samples judged by three experts who read the documentation, the root cause detection accuracy is 82% (determined by maximum winning vote). The system's validity when applied in other industries is approximately 75%.

Root Cause Knowledge Representation

Root Cause Knowledge Representation refers to the technical problem of encoding human knowledge and reasoning into a symbolic language that can be processed by information systems. The representation is specifically designed to represent information about the study domain and is used to solve complex tasks such as service or resolution management. It also incorporates findings from psychology (Ronald, 1983) on how and what humans use to solve problems and represent knowledge to make it easier to design and build. To represent the problem causes in terms of relevant factors in the context of management information, the authors use a fishbone diagram in Fig. 7 that illustrates the factors that contribute to the development of complaint handling. According to the synthesis of both quantitative and qualitative data from the customer complaint document database, five main problem factors were identified in financial business: communication, product, process/operation, defects/impact, and stakeholder involvement. The details of words related to each main problem can be represented by a fishbone diagram, as shown in Fig. 7. The sample analysis led to the conclusion that the most commonly encountered problems were "communication" (50%), and "banking processes" (30%). The four most common causes that contribute to 50% of the "communication" issues are: (1) customer failure to contact; (2) missed call; (3) speaking ignorantly; and (4) miscommunication between staff and customers.

Visualization Of Monitoring Customer Churn

The range for the sentiment value is -1 to 1, and the minimum value for the magnitude is 0. A negative sentiment value indicates a negative emotion. If there is a great deal of negative language, the value approaches − 1. The magnitude scale is determined by the number of negative words and the document's length. as an example of data from the data analysis (Table 7).

Table 7

Statics from 10 sample complaint messages
Complaint message	Len of message (characters)	Sentiment	Magnitude
1	1695	-0.65	8.53
2	1256	-0.64	6.57
3	2255	-0.54	10.31
4	1117	-0.58	4.24
5	145	-0.56	1.19
6	447	-0.57	5.07
7	1812	-0.59	4.00
8	1444	-0.54	7.22
9	1621	-0.63	10.31
10	1692	-0.55	8.15

From the table, it can be seen that the relationship between magnitude and document length is 0.690, which is high significant. It is often natural that if the customer does not receive the service, the value of the service is lower than the customer's expectations. It will inevitably show many reactions, for example, if it is transmitted through a document, it will write a long description (Best & Andreasen, 1977; Day & Landon, 1976).

The magnitude of negative emotions, combined with the magnitude of those emotions, can be used to predict which customers are likely to switch brands. To gain a better understanding of customer complaints, it was important to calculate sentiment and magnitude values, which were then presented graphically in Fig. 8. The results showed that the customers under observation belong to the orange circle group.

The research on aggressive marketing in customer retention suggests that when faced with limited business resources and a large customer base, adopting an automated, user-friendly, and cost-effective operation is a wise choice. One such operation involves (1) regularly reviewing customer-posted documents to identify any potential defection, and (2) if a customer is determined to be at risk of leaving, utilizing a root cause investigation process to address the issue and regain the customer's trust.

The research investigates previous academic literature on customer compliance and its impact on the marketing paradigm. Research has found that incorrect detection of root causes can lead to inefficient management and the loss of intangible assets such as corporate reputation and customer retention. Identifying the root cause of a customer's complaint is crucial for improving customer retention as it allows for targeted and effective solutions to be implemented. Traditional methods of collecting dissatisfaction data, such as quantitative or qualitative interviewing or questionnaires, are passive in nature. The study of communication theory and self-expectancy theory as they relate to driving customer-compliant behavior is crucial. Complaint behaviors are an active way that customers acquire to resolve the negative discrepancy between expectations and perceived satisfaction. The higher the perceived importance, the more is at stake, and thus, the more serious the perceived negative gap between expected and realized utility (Best & Andreasen, 1977; Day & Landon, 1976).

The study collected data on 65,536 customers who had message complaints, randomly selected from the CFPB between 2021 and 2022. A total of 28,820 words were extracted after the preprocessing step. The descriptive statistics (min, max, and mean) of the sample are 5 words, 5358 words, and 185 words, respectively. The relevant words were selected for further semi-automated clustering based on a frequency count greater than 500 (an approximately 0.05 threshold that was computed from the ratio between the documents that contain that word and the total number of documents). The 54 relevant words were clustered into strong word sets (CD, PD, and PR). While 34 words were identified using rough set theory and classified as IM and OB by experts, The reasoning clustering algorithms were developed based on financial event-oriented and word-meaning dictionaries. From the purpose system of the financial event, the authors were able to cluster these words into 5 groups, which are "communication," "product," "process," "impact," and "object." There are many differences between the factors in SERVQUAL and this study. For example, tangibles are related to the accuracy, security, consistency, and faster operation of financial transactions, and assurance quantifies a sense of security in the transaction, competitive loan rates, interest rates on savings and time deposits, and friendly employees. It can be seen that "security" and "sense of security" are the same concept, while "competitive loan rates" and "interest rates on savings" are financial products in the study. In addition, the use of SERVQUAL does not apply to persons or entities involved in financial transactions. The word grouping according to this study method was consistent with the evidence of the information in the complaints (Corey & Friedman, 2022). The feature clustering results, as shown in Table 1, indicate that each factor has a semantic concept in the financial-event dimension. The clarity and unambiguity of the category led to an effective complaint management solution (Lovelock et al., 2009; Salim et al., 2018).

The transcription of complaint messages into sentences in this research results in a customer journey. The customer’s journey aligns with the compliant message because it can visualize the events, touchpoints, and pain points. Likewise, in other businesses, an event on a journey map is related to the stakeholders and the activities. Additionally, the "stakeholders" and the "pain points" are the elements of the factors "OB" and "IM." This is supported by Følstad et al. (2013) and Dove et al. (2016), which extract a set of events that define key experiences in the life cycles of customers. In addition, Baker et al. (2015) found that customers who experience negative emotions with their services usually do not respond verbally and tend to lodge complaints with third parties. Since there are many dissatisfaction problems on a customer's journey map, a serial complaint message consists of compound causes. Dissatisfaction is one of the reasons that customers switch to other banks (Manrai & Manrai, 2007). When customers feel that the bank is not handling their complaints properly, they file a complaint with the bank's supervisory agency. The results (Table 1) are consistent with previous academic work (Bougie et al., 2003).

The study utilizes a mixed method of identifying the root cause of customer complaints through analyzing words and sentences at both local (word) and global (sentence) levels, which is supported by the theories of cause proximity and argumentation theory, a crucial theory in communication theory. According to argumentation theory, the arguments, sub-arguments, and potential arguments are applied at the word and sentence levels in this study (Corbett & Connors, 1999; Miller & Levine, 1996). The accuracy of root cause detection is 82%, using the maximum win vote of three experts. The benefit of this study is demonstrated by comparing an example of the root cause from the system, "Communication-Related to Product" (record no. 1 in Table 3), with the label from the submenu (selected by the customer from the pull-down menu): "Struggling to pay your loan" (the CFPB database). This comparison highlights that the handling of complaints in this example focuses on the communication system, while "Struggling to pay your loan" cannot be identified. This finding also supports customer complaints against airlines that misunderstood and did not handle complaints properly (Gronhaug, 1977; Heung & Lam, 2003; Kotler et al., 2004). The increase in consumer complaints makes research on the CFPB database crucial for protecting consumers and enhancing the financial industry's sustainability. The study of root cause detection uses simple supervised learning in part for weak words, while strong words are judged by experts (Aggarwal & Zhai, 2012; Kaveh et al., 2016). Furthermore, the proposed system is a straightforward churn analysis, which does not require supervised data (Danet, 1978; Umayaparvathi & Iyakutti, 2016; Xie et al., 2009). The prototype system is designed to be simple, cost-effective, and adaptable.

The framework validity test (reviewing the sub-process) focuses on two dimensions, in addition to testing the accuracy of the root cause detection process. The review is rated by three experts (0–100 scores), and the average score is 75%. Finally, from the perspective of managerial implication, the study shows that the bank must try to find the root of the problem of any failure or error in service so that the company can formulate the best solution to any problems or complaints that arise.

The study utilizes compliant messages from the CFPB database, which mostly use spoken language, to identify the root cause of customer complaints. The text processing level results are consistent with the study by Kaveh et al. (2016), which similarly found that "communication" was a common label. However, many previous studies have used probability models that only consider word frequency and neglect the importance of language and psychological communication, resulting in mislabeling. As the goal of labeling is to efficiently address customer pain points, previous studies have not been suitable to solve this problem (Aggarwal & Zhai, 2012; Ayres et al., 2013; Littwin, 2015; Kaveh et al., 2016). Furthermore, there are some topics that do not correspond to any labels, such as "family" and "divorce and ex-spouse." This suggests that there may be new issues that have been neglected by the CFPB but have been identified using the proposed approach, which uses Latent Dirichlet Allocation (LDA). This ability to identify new and evolving issues not possible under the CFPB labeling conventions is a key benefit of this proposed approach.

Since no consumer complaint results in a favorable outcome for the buyer (Gronhaug, 1977; Heung & Lam, 2003), not even those handled by the consumer protection agency (Danet, 1978), it may be interpreted that consumers have partly unrealistic expectations. In most previous research, it is assumed that consumer satisfaction or dissatisfaction and complaining behavior can provide useful feedback for businesses, government, and consumer agencies (Best & Andreasen, 1977). But studies (Day & Landon, 1976; Best & Andreasen, 1977) show that only a small number of complaints ever get to government or consumer agencies.

Managerial implication

Bank managers aiming to improve customer service can benefit significantly from the findings of this study. The mixed method of analyzing words and sentences at both local and global levels achieved an accuracy rate of 82%, demonstrating high precision in detecting the root causes of complaints. This highlights the importance of conducting a comprehensive analysis of customer complaints to identify underlying issues accurately. The proposed system offers a flexible and cost-effective solution for banks to proactively monitor customer complaints and address issues using a simple churn analysis approach that does not require supervised data. Additionally, the system can be adapted to any business domain by modifying the feature extraction and categories and can handle real-time processing in dynamic environments.

Theoretical implication: The research contributes to the theoretical implications by integrating three theories: the SERVQUAL model, total quality management (TQM), and communication theory. The SERVQUAL model, which is based on the customer's assessment of service quality, highlights the key factors that contribute to customer satisfaction and identifies areas for improvement. TQM focuses on improving all aspects of a business through data-driven decision making and root cause analysis to prevent similar problems from occurring in the future. Communication theory states that communication is not just the flow of information but also the shared thoughts and emotions of the sender and the receiver's reactions and emotions. Argumentation theory, relevant to communication theories, suggests that effective complaints should be based on thorough investigation, supported by well-designed sub-arguments, and presented with only relevant and persuasive information. This research contributes to these theories by showing how the principles of argumentation theory can be applied to enhance the development of effective consumer complaints and improve overall business performance.

Limitation

The limitations of using English and other languages in the marketing paradigm, especially in comparison with the native language, can have a significant impact on the interpretation and effectiveness of marketing messages. Cultural dimensions can cause miscommunications and misunderstandings due to differences in idioms, expressions, and cultural references. In addition, different languages may have different grammatical structures, which can make it difficult to properly convey the intended meaning of a marketing message. This can lead to ineffective or misinterpreted marketing strategies that do not resonate with the target audience. To overcome these limitations, it is crucial to understand the cultural and linguistic differences and to use language that is appropriate and effective in communicating with the target audience.

Further study

One area of future work in the field of marketing is the exploration of the impact of language on consumer behavior and the effectiveness of marketing messages in different cultural contexts. This research would aim to better understand the role of language in shaping consumer attitudes and behaviors and how marketing messages can be crafted to effectively reach and engage with different cultural groups. The study could include an examination of the use of language in advertising, product labeling, and other marketing materials, and how these elements can impact consumer perception and decision-making. Additionally, the research could look at how cultural dimensions, such as values and beliefs, influence the way language is used in marketing and the impact of these cultural dimensions on consumer behavior. The ultimate goal of this research would be to provide insights and guidance for marketers on how to effectively craft and deliver culturally sensitive and impactful marketing messages.

Source of data: The data used in this study was obtained from the CFPB database, which is publicly available on their website at https://www.consumerfinance.gov/.

Acknowledgments: Authors would like to acknowledge the substantial contribution provided by the Consumer Financial Protection Bureau.

Aggarwal, C. C., & Zhai, C. (2012). A Survey of Text Clustering Algorithms. Mining Text Data. Springer, 77-128. http://dx.doi.org/10.1007/978-1-4614-3223-4_4.
Akash, P. N., Manoj, H., Suhas, S. A., & Abhay, D. L. (2021). Online Product Reviews and Their Impact on Third Party Sellers Using Natural Language Processing. International Journal of Business Intelligence Research, 12(1), 26–47. https://EconPapers.repec.org/RePEc:igg:jbir00:v:12:y:2021:i:1:p:26-47
Akbar, M.M., & Parvez. N. (2009). Impact of Service Quality, Trust, and Customer Satisfaction Loyalty. ABAC Journal, 29(1), 24-38. http://www.assumptionjournal.au.edu/index.php/abacjournal/article/view/526/472
Akhgari, M., Bruning, E. R., Finlay, J., & Bruning, N. S. (2018). Image, Performance, Attitudes, Trust, and Loyalty in Financial Services. International Journal of Bank Marketing, 36(4), 744-763. https://www.proquest.com/scholarly-journals/image-performance-attitudes-trust-loyalty/docview/2057445562/se-2
Andrews, J., & Eade, E. (2013). Listening to students: Customer journey mapping at birmingham city university library and learning resources. New Review of Academic Librarianship, 19(2), 161–177. https://doi.org/10.1080/13614533.2013.800761
Ang, L. & Francis, B. (2006). E Customer retention management processes. A quantitative study. European Journal of Marketing, 40 (1/2), 83-99. https://www.emerald.com/insight/content/doi/10.1108/03090560610637329/full/html
Aspinall, E., Nancarrow, C. & Stone, M. (2001). The meaning and measurement of customer retention. Journal of Targeting, Measurement and Analysis for Marketing, 10 (1), 79-87. DOI: 10.1057/palgrave.jt.5740035
Ayres, I., Lingwall, J., & Steinway, S. (2013). Skeletons in the Database: An Early Analysis of the CFPB's Consumer Complaints. Fordham Journal of Corporate & Finance Law, 19, 343 -391. https://ianayres.yale.edu/sites/default/files/files/Skeletons%20in%20the%20Database(2).pdf
Baker, R.S., Bloom, N., & Davis, J.S. (2015). Measuring Economic Policy Uncertainty. National Bureau of Economic Research, Working Paper 21633. http://www.nber.org/papers/w21633
Best, A., & Andreasen, A. R. (1977). Consumer Response to Unsatisfactory Purchases: A Survey of Perceiving Defects, Voicing Complaints, and Obtaining Redress. Law & Society Review, 11, 701-742. http://dx.doi.org/10.2307/3053179
Blei, D.M., Ng, A.Y., & Jordan, M.I. (2003). Latent Dirichlet allocation. Journal of Machine Learning Research, 3, 993–1022. https://www.jmlr.org/papers/volume3/blei03a/blei03a.pdf
Bougie, R., Pieters, R., & Zeelenberg, M. (2003). Angry customers don’t come back, they get back: The experience and behavioral implications of anger and dissatisfaction in services. Journal of the Academy of Marketing Science, 31(4), 377-393. https://doi.org/10.1177/0092070303254412
CFPB Annual Complaint Report (2021). Retrieved March 19, 2022, from, https://www.consumerfinance.gov/about-us/newsroom/cfpb-annual-complaint-report-highlights-more-than-a-half-million-complaints-received-in-2020/
CFPB, Website, Consumer Complaint Database. Retrieved December 25, 2022, from, https://www.consumerfinance.gov/data-research/consumer-complaints/
Chan, V. H. Y., Chiu, D. K., & Ho, K. K. (2022). Mediating effects on the relationship between perceived service quality and public library app loyalty during the COVID-19 era. Journal of Retailing and Consumer Services, 67, 102960. https://www.sciencedirect.com/science/article/pii/S0969698922000534
Consumer Finance Protection Bureau (2021). Retrieved February 5,2022 from https://www.consumerfinance.gov/
Corbett, E. P. J., & Connors, R.J. (1999). Classical rhetoric for the modern student. Oxford: Oxford University Press.
Corey, B. F. (2022). Root Cause Analysis and Proximate Cause Retrieved March 10, 2022, from, https://romanolawgroup.com/legal-insights/root-cause-analysis-and-proximate-cause/
Coussement, K., & Van den Poel, D. (2008). Improving customer complaint management by automatic email classification using linguistic style features as predictors. Decision support systems, 44(4), 870-882. https://www.sciencedirect.com/science/article/pii/S0167923607001820
Crie, D. (2003). Consumers’ complaint behaviour. Taxonomy, typology and determinants: Towards a unified ontology. Database Marketing & Customer Strategy Management, 11(1), 60–79. DOI: 10.1057/palgrave.dbm.3240206
Crolic, C., Thomaz, F., Hadi, R., & Stephen, A. T. (2022). Blame the bot: anthropomorphism and anger in customer–chatbot interactions. Journal of Marketing, 86(1), 132-148. https://journals.sagepub.com/doi/pdf/10.1177/00222429211045687
Danet, B. (1978). Toward a Method to Evaluate the Ombudsman Role. Administration and Society, 10, 335-70. https://doi.org/10.1177/00953997780100030
Day, R. L., Landon, E. & Laird L. (1976). Collecting Comprehensive Consumer Complaint Data by Survey Research. Advances in Consumer Research, 3, eds. Beverlee B. Anderson, Cincinnati, OH: Association for Consumer Research, 263-268. https://www.acrwebsite.org/volumes/9274
Dhruv, G. & Anne L. R. (2020). Understanding Retail Experiences and Customer. Journey Management. Journal of Retailing, 96(1), 3-8. DOI: 10.1016/j.jretai.2020.02.002
Dove, L., Reinach, S., & Kwan, I. (2016). Lightweight journey mapping: The integration of marketing and user experience through customer driven narratives. Proceedings of the 2016 Conference Extended on Human Factors in Computing Systems ACM, 880–888. https://doi.org/10.1145/2851581.2851608
Feldman, R., & Sanger, J. (2007). The text mining handbook: advanced approaches in analyzing unstructured data. Cambridge University Press.
Financial Consumer Protection Center Bank of Thailand (2021). Retrieved January 12,2026 from, https://www.1213.or.th/th/Market%20conduct/stats/ComplaintReport_64.pdf
Følstad, A., Kvale, K., & Halvorsrud, R. (2013). Customer journey measures-state of the art research and best practices. Oslo, Norway: Report A 24488. http://hdl.handle.net/11250/2390670
Ghamri, N. S. (2017). Positive and Negative Effects of Using Electronic Banking on Customers and Small Entrepreneurs: An Exploratory Study in the Western Region of Saudi Arabia. Business and Economic Research, 7(2), 311-331. https://ideas.repec.org/a/mth/ber888/v7y2017i2p311-331.html
Gronhaug, K. (1977). Exploring Consumer Complaining Behavior: A Model and Some Empirical Results. Advances in Consumer Research, 4th ed, William D. Perreault, Jr. Atlanta: Association for Consumer Research, 159-65. https://www.acrwebsite.org/volumes/9348/volumes/v04/NA-04
Griffiths, T. L., Steyvers, M., & Tenenbaum, J. B. (2007). Topics in semantic representation. Psychological Review, 114(2), 211–244. https://doi.org/10.1037/0033-295X.114.2.211
Grönroos, C. (1990). Relationship approach to marketing in service contexts: the marketing and organizational behavior interface. Journal of Business Research, 20(4), 3-11. https://www.academia.edu/18427432/Relationship_approach_to_marketing_in_service_contexts_The_marketing_and_organizational_behavior_interface
Heskett, J. L., Sasser, W. E., Jr, & Schlesinger, L. A. (1997). The service profit chain. New York: Free Press.
Heung, C. S. & Lam, T. (2003). Customer Complaint Behavior towards Hotel Restaurant Services. International Journal of Contemporary Hospitality Management, 15(5): 283-289. https://www.researchgate.net/publication/235293362_Customer_complaint_behaviour_towards_hotel_restaurant_services
Jurafsky, D., & Martin, J. H. (2000). Speech and language processing: An introduction to natural language processing, computational linguistics, and speech recognition. Upper Saddle River, N.J: Prentice Hall.
Kaveh, B., Hamed N., & Jeffry S. L. (2016). Dirichlet Allocation (LDA) for Topic Modeling of the CFPB Consumer Complaints. Retrieved March 19, 2022, from https://doi.org/10.48550/arXiv.1807.07468
Kotler, P. (2003). Marketing Management, 11th ed., Prentice-Hall, Englewood Cliffs, NJ.
Kotler, P., Brown, L., Adam, S., & Armstrong, G. (2004). Marketing, 6th ed. French Forest, NSW, Pearson Education Australia.
Lekham, L.A., Wang,Y., Hey, E., Lam, S.S. , & Khasawneh, M.T. (2020). A multi-stage predictive model for missed appointments at outpatient primary care settings serving rural areas IISE. Transactions on Healthcare Systems Engineering, 11, 79-94. https://doi.org/10.1080/24725579.2020.1858210
Littwin, A. K. (2015). Examination as a Method of Consumer Protection. Temple Law Review, 87(807). https://www.templelawreview.org/lawreview/assets/uploads/2016/01/Braucher-Littwin-87-Temp.-L.-Rev.-807.pdf
Lovelock, C., Wirtz, J., & Patricia, C. (2009). Essential of Services Marketing Management. Prentice Hall, Peasron Education South Asia Pte Ltd.
Luo, L., Ratchford, B.T., & Yang, B. (2013). Why we do what we do: A model of activity consumption. Journal of Marketing Research, 50(1), 24-43. https://doi.org/10.1509/jmr.10.0416
Manrai, L.A., Manrai, A.K. (2007). A field study of customers' switching behaviour for bank services. Journal of Retailing and Consumer Services, 14(3), 208-215. DOI: 10.1016/j.jretconser.2006.09.005
Matusitz, J. & Breen Gerald-Mark (2011). Applying Communication Theories toward Designing Compliance-Gaining Techniques in Customer Dissatisfaction. www.pasosonline.org, 9(1), 37-46. https://doi.org/10.25145/j.pasos.2011.09.004
Miller, M. D., & Levine, T. R. (1996). Persuasion. In M. B. Salwen, & D.W. Stack (Eds.), An integrated approach to communication theory and research. Oxford: Oxford University Press, 261-276.
Monge, P. R., & Kirste, K. K. (1980). Measuring proximity in human organizations. Social Psychology Quarterly, 43, 110-115. https://doi.org/10.2307/3033753
Monge, P., Rothman, L., Eisenberg, E., Miller, K., & Kirste, K. (1985). The dynamics of organizational proximity. Management Science, 31, 1129-1141. https://www.jstor.org/stable/2631736
Nadda, W., Boonchieng, W. & Boonchieng, E. (2022). Influenza, dengue and common cold detection using LSTM with fully connected neural network and keywords selection. BioData Mining, 15, 5. https://doi.org/10.1186/s13040-022-00288-9
Omoregie, O.K., Addae, J.A., Coffie, S., Ampong, G.O.A., & Ofori, K.S. (2019). Factors influencing consumer loyalty: Evidence from the Ghanaian retail banking industry. International Journal of Bank Marketing, 37(3), 798-820. https://fardapaper.ir/mohavaha/uploads/2019/09/Fardapaper-Factors-influencing-consumer-loyalty-evidence-from-the-Ghanaian-retail-banking-industry.pdf
Parasuraman, A., Zeithaml, V.A. & Berry, L.L. (1988) SERVQUAL: A Multiple-Item Scale for Measuring Consumer Perceptions of Service Quality. Journal of Retailing, 64, 12-40. https://www.researchgate.net/journal/Journal-of-Retailing-0022-4359
Payne, A., Holt, S. & Frow, P. (2001). Relationship value management: exploring the integration of employee, customer and shareholder value and enterprise performance models. Journal of Marketing Management, 17 (7/8), 785-817. https://doi.org/10.1362/026725701323366827
Pawlak, Z. (2002). Rough Set Theory and Its Applications. Journal of Telecommunications and Information Technology, 3, 7-10. https://bibliotekanauki.pl/articles/309289.pdf
Pessoa, A. & Stephany, S. (2022). An innovative approach for attribute reduction in rough set theory. Intelligent Information Management, 6, 223-239. DOI: 10.4236/iim.2014.65022
Prajogo, D. I., & Sohal, A. S. (2006). The relationship between organization strategy, total quality management (TQM), and organization performance––the mediating role of TQM. European journal of operational research, 168(1), 35-50. https://www.sciencedirect.com/science/article/pii/S0377221704003121
Radice, R. (2018, January 9). How to transform your customer experience with storytelling. Retrieved June 20, 2022, from, https://medium.com/@RebekahRadice/how-to-transform-your-customer-experience-with-storytelling-5d05d9c0d537
Ronald, J. B. (1983). What IS-A is and isn't. An Analysis of Taxonomic Links in Semantic Network. IEEE Computer, 16 (10), October 1983. DOI:10.1109/MC.1983.1654194
Rumler, F., & Waschiczek, W. (2010). The impact of economic factors on bank profits. Monetary Policy & the Economy, 4, 49-67. https://www.oenb.at/dam/jcr:1372d37d-a783-4612-9369-aba50006da5d/mop_2010_q4_analyses03_tcm16-221663.pdf
Salim, A., Setiawan, M., Rofiaty, R., & Rohman, F. (2018). Focusing on Complaints Handling for Customer Satisfaction and Loyalty: The Case of Indonesian Public Banking. European Research Studies Journal, 21(3), 404- 416. https://ersj.eu/journal/1071
SEA, Supporting English Acquisition. RIT and NTID are registered trademarks of Rochester Institute of Technology. Retrieved March 10, 2022, from https://www.rit.edu/ntid/sea/processes/paragraph/process/placement
Sila, I., & Ebrahimpour, M. (2003). Examination and comparison of the critical factors of total quality management (TQM) across countries. International journal of production research, 41(2), 235-268. https://www.tandfonline.com/doi/abs/10.1080/0020754021000022212
Singh, J. (1988). Consumer complaint intentions and behaviour: Definitional and taxonomical issues. Journal of Marketing, 52(1), 93-107. https://doi.org/10.2307/1251688
Singh, J., & Wilkes, R. E. (1996) When consumers complain: A path analysis of the key antecedents of consumer complaint response estimates. Journal of the Academy of Marketing Science, 24, 350-365. https://link.springer.com/article/10.1177/0092070396244006
Soto-Sanfiel, M. T. & Igartua, J. (2014). Cultural proximity and interactivity in the processes of narrative reception. International Journal Arts and Technology, 9(2):87-107. https://www.researchgate.net/publication/268811187
Soto-Sanfiel, M. T. & Igartua, J. (2014). Cultural proximity and interactivity in the processes of narrative reception. International Journal Arts and Technology, 9(2):87-107. https://www.researchgate.net/publication/268811187
St. Louis Communication College, Topic Sentence and Paragraph. Retrieved March 10, 2022, from https://stlcc.edu/student-support/academic-success-and-tutoring/writing-center/writing-resources/topic-sentence-paragraph.aspx
Umayaparvathi, V. & Iyakutti, K. (2016). A survey on customer churn prediction in telecom industry: Datasets, methods and metrics. International Research Journal of Engineering and Technology (IRJET), 4(4), 1065–1070. https://www.irjet.net/archives/V3/i4/IRJET-V3I4213.pdf
Vogt, W. P., & Johnson, B. (2011). Dictionary of statistics & methodology: A nontechnical guide for the social sciences. Sage. https://books.google.com/books?hl=en&lr=&id=rDib3X4YsSQC&oi=fnd&pg=PR1&dq=Vogt,+W.P.+(2005).+Dictionary+of+Statistics+%26+Methodology:+A+Nontechnical+Guide+for+the+Social+Sciences.+SAGE.&ots=9d1DgQeV1W&sig=ESYUjyyspNZ0edx7fQibf7ZlWUQ
Vroom, V.H. (1964), Work and Motivation, John Wiley and Sons, New York, NY.
Walker, R.C., Cardon W.P., & Aritz, J. (2018). Enhancing Global Virtual Small Group Communication Skills. Journal of Intercultural Communication Research, 47(5), 421-433. https://doi.org/10.1080/17475759.2018.1475292
Widyastuti, R. S., & Armanto, B. (2013). Kompetisi Industri Perbankan Indonesia (Indonesian Banking Industry Competition). Bulletin of Monetary Economics and Banking, 15(4), 417- 439. https://www.bmeb-bi.org/index.php/BEMP/article/view/74
Xie, Y., Li, X., Ngai, E. & Ying, W. (2009). Customer churn prediction using improved balanced randomforests. Expert Systems with Applications, 36(3):5445–5449. DOI: 10.1016/j.eswa.2008.06.121

No competing interests reported.

Download PDF

Journal Publication

published 05 May, 2023

Read the published version in Social Network Analysis and Mining →

Editorial decision: Major revision
22 Mar, 2023
Editor assigned by journal
22 Mar, 2023
Submission checks completed at journal
22 Mar, 2023
First submitted to journal
21 Mar, 2023

You are reading this latest preprint version

One to one marketing management via customer complaint

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Literature review

Methodology

The Study Framework

\({lw}_{m}\) \(=(\frac{l}{l}, \frac{l-1}{l}, \frac{l-2}{l},\dots , \frac{1}{l})\) … (2)

\({lst}_{j}\) \(=(\frac{k}{k}, \frac{k-1}{k}, \frac{k-2}{k},\dots , \frac{1}{k})\) … (3)

Message Preprocessing, Word Selection And Computation

Word Extraction

Monitoring Churn To Improve Customer Retention

Results

Words selection and construction knowledge set

Sentence Segmentation

Labeling, Transformation And Matrix Manipulation

Root Cause Detection

Robustness Test

Root Cause Knowledge Representation

Visualization Of Monitoring Customer Churn

Discussion

Conclusions, Contribution, Limitation, And Further Study

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1