Impact of tweets on Indian movies performance: An integrative approach using sentiment analysis and emotion detection

doi:10.21203/rs.3.rs-3432798/v1

Download PDF

Research Article

Impact of tweets on Indian movies performance: An integrative approach using sentiment analysis and emotion detection

https://doi.org/10.21203/rs.3.rs-3432798/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Due to social media's rapid growth, moviegoers primarily rely on internet reviews before buying. Studios must determine how reviews affect sales. This study examines how online reviews affect movie box office earnings using sentiment analysis and emotion recognition. Text reviews affect attendance and revenue as much as rating does. This data can help filmmakers and advertising target demographics. Films met audience expectations, with increased trust and happiness and lower wrath, surprise, disgust, and sadness. However, some films scored better for specific emotions, suggesting they elicited different audience responses. Most film tweets were happier and more trusting after release than before. However, tweets about some films showed more wrath and dread, suggesting marketing or audience reception concerns.

Sentiment Analysis

Emotion Identification

Movie Promotion

Twitter Scraping

Natural Language Processing

Artificial Intelligence

The rapid growth of various social media platforms has enabled consumers to share product experiences and specifics online. According to research, prior to making a purchase decision, consumers peruse online reviews. Due to brief life cycles, the film industry is extremely competitive. The film's genre, director, actors, and plot summary, as well as the marketing techniques used to advertise the film, can all influence box-office earnings. Most prospective moviegoers read movie evaluations, and electronic word of mouth (eWOM) influences box office success (Lee & Choeh, 2018). Nevertheless, movie evaluations contain a wealth of information regarding the perspectives of moviegoers. Numerous studies used sentiment analysis to determine the overall polarity of a text to determine what consumers were thinking (Hu & Chen, 2016; Hu et al., 2018; Liu, B., 2012).

1.1 Motivation and Research Objectives

The business problem is defined as: With immense dependence on social media, especially Twitter, more and more people are sharing their views and opinions, which in turn influence their decision-making. The goal is to find out how people feel about movies on Twitter and how that affects how much money they make at the box office. This will be done using NRCLex's sentiment analysis and emotion detection tools to find patterns and give the movie makers important information.

The motivations for this research are as follows: Text-based sentiment analysis is a hard business problem that has many uses and effects for both academics and business professionals. Because of the internet's quick expansion and the proliferation of social media, moviegoers can easily locate and share information about films, as well as other people's opinions and reviews, which could then affect other readers' decisions to see the film or forgo seeing it in theaters. We decided to investigate social media's ability to predict actual outcomes because it can also be interpreted as collective wisdom. Also, when making marketing and advertising campaigns, it can be helpful to gather information about how people talk about certain products. This paper discusses a similar study. We look at the task of using text from Twitter, which is one of the fastest-growing social networks on the Internet, to predict how much money movies will make at the box office. Twitter, which is a microblogging network, has become more popular in recent months. Because of this, it now has tens of millions of active users who make and share content all the time.

In this study, movies are our primary focus for two reasons:

The social media community is particularly interested in movies, as indicated by the vast number of individuals who discuss them and the diversity of their perspectives.
The real outcomes can be easily noticed through movie box office income, and the trend in attitudes can be traced over time, particularly before and after the release.

The research objectives of this work are stated as:

Analyze the effectiveness of the deep learning-based NLP (Natural Language Processing) framework ‘FLAIR’ for categorizing movie reviews as positive, negative, or neutral in sentiment.
Establish the connection between the movie's quality and the precision of the sentiment classification. When sentiment analysis models are used to analyze reviews of movies with high and low box office performance, compare the models' accuracy.
Examine how well emotion recognition models work when dealing with various emotions (such as joy, anger, sadness, fear) and determine which emotions are more identified in movie reviews than others.

1.2 Research Questions and Contributions

The research questions are stated as:

RQ1: How well can emotion analysis foretell a movie's box office performance?
RQ2: Is there a shift in the sentiment of movie reviews over time, and if yes, what variables contribute to these changes?
RQ3: What are the most prevalent emotions conveyed in Twitter movie reviews, and how do they change depending on the genre and the film? Also, why are some emotions more common in Twitter movie reviews than others?
RQ4: Is it possible to find prospective marketing strategies for the film, such as focusing on emotional reactions in certain audiences, by emotion identification in Twitter movie reviews?

As Twitter data is unstructured, cleaning and preprocessing is necessary. Three months before the release of 10 selected films, 10,000 tweets will be preprocessed, and three months after, another 10,000 tweets will be preprocessed. The sentiment analysis was then performed in Python using the 'nltk' library. Additionally, we used the Flair Class distribution to divide the tweet text into positive and negative categories that correspond to how individuals felt about the film. In addition, we intend to perform emotion detection on these tweets to accurately identify the eight fundamental emotions: fear, anger, sadness, trust, surprise, sadness, disgust, joy, and anticipation. Emotion detection can be used in box office analysis to determine how people feel about films. Analysis of online conversations on social media platforms such as Twitter is one method for measuring box office emotions. Using sentiment analysis algorithms and natural language processing techniques, it is possible to identify and monitor the emotions expressed in online movie-related conversations. Using sentiment analysis, for instance, tweets about a specific film can be analyzed, and the trend in sentiment can be monitored over time. This can help predict a film's performance at the box office and provide valuable insight into the audience's overall emotional response. By identifying the emotions that resonate with their target audience, filmmakers can create more successful marketing campaigns using emotion detection. By analyzing social media conversations about their films, filmmakers can determine which emotions are most strongly associated with their films. They can then use this data to construct marketing campaigns that appeal to their target audience more effectively. Using emotion detection to determine the emotions most strongly associated with a movie and leveraging those emotions in marketing campaigns, marketers can increase chatter surrounding the film.

The paper is organized into seven sections. The first section talks about Introduction. The second section is about Literature Review. The third section talks about the Research Methodology employed. The fourth section is about Experimental Setup and Findings. The fifth and sixth sections are Implications and Conclusion. The last section talks about Limitations of the study.

Emotion involves feelings, experience, physiology, behavior, cognitions, and conceptualization (Ortony, 2009) (M.Cabanac, 2002). According to WordNet Search 3.0, all intense feelings are considered emotions. An emotion is a mental condition that is "related to a variety of feelings, ideas, behavioral responses, and a degree of pleasure or dissatisfaction". (Pempek, 2009). Emotions play a significant role in human nature and are a form of inheritance (Sharma, 2017). Different users' expressions of a particular emotion are identified. This study's author proposed a model for identifying emotions, which has numerous applications in the digital age. It can be used by businesses to gauge consumer sentiment regarding their products or services. Therefore, they can endeavor to improve them according to the needs of their customers (Alswaidan, 2020). Today, the design of a textual emotion detection model is a very fascinating problem to solve because the majority of human-computer interfaces, such as emails, forums, and social media platforms like Twitter, YouTube, and Facebook, rely primarily on text for communication (Kušen, 2019) (Angel Deborah, 2021). The requirement for natural language understanding (NLU) (Serrano-Guerrero J., 2022) (Kusal, Patil, Kotecha, Aluvalu, & Varadarajan, 2021) is one of the most challenging aspects of the framework for emotion detection in text processing. The process of emotion detection analyzes the text, which contains several well-arranged words. The text classification procedure can determine a person's emotions simply by examining the text. This is because facial expressions are typically divided into distinct categories, such as happiness, anger, disgust, sadness, fear, astonishment, and shock.

The complexity of the network environment is increasing as the volume of data grows. The study of emotions and how people feel about things has been a top priority in the field of natural language processing because it aids in determining what individuals believe. Since artificial intelligence is still improving, precise emotion detection and sentiment analysis currently have a great deal of scientific value. (Zahid Anwar, 2022). The most difficult aspect of working with lexicons is locating high-quality resources. Regarding emotion analysis, well-known tools include SenticNet (Cambria et al. 2020), WordNet-Affect (Strapparava and Valitutti, 2004), SentiStrength (Baccianella et al. 2010), ANEW (Bradley and Lang 1999), AFFIN (Nielsen 2011), and the NRC Word-Emotion Association Lexicon (Mohammad and Turney 2013), among others. Nonetheless, these instruments are predominantly English based. In other languages, this variety of tools is of lower quality. Due to this, researchers frequently translate these instruments, despite the expected quality loss. The Arabic language is one of these (Guo, 2022).

In any text-based emotion detection system, datasets are compiled by utilizing APIs to obtain data from social media websites. Tweepy on Python, for instance, can be used to obtain Twitter data. After data generation, the text is prepared for any machine learning or deep learning algorithm in the text preprocessing phases. Text pre-processing consists of tokenization, text cleansing, normalization, and the generation of feature vectors or embeddings. Text from social media, product or customer reviews, etc., contains slang terms, emoticons, hashtags, HTML tags, short text, incomplete words, etc., which require preprocessing. The generated feature vectors are then subjected to either machine learning or deep learning (Kusal S. P., 2021). Sentiment analysis (SA) is a subfield of data mining and natural language processing (NLP) problems that examines how people feel about something using computers. In sentiment analysis, machine learning (ML) is the foundation for artificial intelligence (AI) and natural language processing (NLP), and its function is to predict and direct the sentiment. ML systems perform numerous tasks, including part-of-speech (POS) labeling, lemmatization, stemming, co-reference, spam detection, word sense disambiguation (WSD) parsing, and detection of dataset popularity. SA's primary objective is to identify public sentiment toward targeted entities (Anwar, 2022).

Humans, upon reading Why don't you text me more often? ('tum ne mujhe text kyun nahi kya!') We can interpret it as either a furious or sad emotion in the absence of context, and so can machines. The absence of facial expressions and vocal modulations makes identifying emotions in text difficult. (Majeed, 2022). The computer should be able to obtain face information, such as emotions and gender, immediately. Due to the complexity of facial information, however, too many parameters arise during the machine learning process. This eliminates the possibility of instant recognition. (Wei, 2019)

The multimedia content is then appropriately processed to recognize emotions/sentiments, such as by analyzing expressions and postures in images/videos using machine learning techniques or by converting speech to text to perform emotion detection using natural language processing (NLP) techniques. (Graterol, 2021). Natural Language Processing (NLP) techniques are more effective now that text data is available in the psychological domain. New technological developments are occurring in pre-trained language models based on deep learning that require fine-tuning on limited datasets; BERT is one such model. (Kumar, 2022)

The majority of people use social media (such as Facebook and Twitter) in a variety of ways, including posting videos, images, and texts, sharing the posts of others to express their emotions and discover information of interest, and obtaining the most recent international news. In a variety of contexts, including treatment, pre-election outcomes, development processes, and medical status, the sentiments of social media users are crucial. (Swe, 2021)

Real-time sentiment analysis enables us to anticipate potentially problematic situations, such as civil disturbance, so we can take measures to prevent or control them. The integration of large deep learning models, which considerably slows them down, prevents the use of current effective models for real-time emotion detection. (Vijayvergia, 2021). There is a correlation, according to the U.S., between the sentiments and emotions identified in tweets and ILI visits to health facilities in the same regions. Across all U.S. locations, the strongest correlations are negative for anger and surprise, positive for sadness, and mixed for dread, disgust, and happiness. (Volkova, 2017)

Before purchasing specific products or services, consumers rely on the reviews/comments of other users and businesses consider the opinions of their customers. Typically, these reviews/comments are provided in non-normative colloquial language in various contexts and domains (social media, forums, news sites, blogs, etc.). By designating each of these texts positive, negative, and sometimes neutral sentiment values, sentiment classification plays a crucial role in the analysis of user-generated texts. In addition, these texts typically contain many expressed or concealed emotions (such as happiness, sadness, etc.) that could considerably contribute to identifying sentiments. (Tesfagergish, 2022). In a related study, we found that when extended texts are provided to an algorithm like Word Sense Disambiguation (WSD), we can improve the accuracy of assigning correct word-senses, but we also likely obtain less reliable category ratings. Similar to this, it may be more difficult to discern the correct word senses in shorter input texts, but we would likely achieve greater precision in scoring. (Dessì, 2021)

Social media posts are prominent and frequently loaded with emotion. By analyzing and examining these social media posts, it may be possible to identify emotional states and their underlying causes. Nonetheless, the overwhelming volume of data makes this analysis extremely challenging. Artificial intelligence can aid in the automated discovery of emotions, feelings, personality traits, beliefs, and their influences on societal trends. (Mubeen, 2022). Long short-term memory (LSTM) deep learning was used to classify all tweets into six emotion classes and two categories (symptom and non-symptom tweets) in covid-19 study. The monitoring system indicates that the majority of tweets were published in March of 2020. After joy, the anger and fear emotions have the greatest number of tweets and user interactions. The results of monitoring user interaction indicate that people use likes and replies to interact with tweets that do not contain symptoms, while they use retweets to spread tweets that contain any COVID-19 symptoms. (Al-Laith, 2021)

The classification of emotions by NLP systems remains challenging, but its significance has grown in recent years. As anticipated, deep learning techniques are now prevalent in this text classification task. The combination of linguistic data supports the benefits of hybrid solutions even further. Unquestionably, class imbalance and how to implement additional variables with deep neural network encodings were among the most intriguing issues for the participants. (Plaza-del-Arco, 2021). The sentiment analysis of a given text determines whether it is positive, negative, or neutral. However, emotion analysis goes further, as evidenced by the distribution of categories under sentiment analysis. Keyword-based and lexical affinity approaches have been used to some extent, despite their drawbacks and inferior accuracy compared to the learning-based approach. In that they classify emotions differently, machine learning and deep learning methodologies are distinct. (Bharti, 2022)

Cognitive science is an interdisciplinary field of study that examines various cognitive processes and sophisticated human behaviors, including perception, thinking, remembering, learning, reasoning, and emotions. Emotions are among the most essential faculties for identifying human social behaviors. (Asghar, 2017). Textual information is sequential; therefore, the order of words and their relationship in a sentence, i.e., context, play an important role in deriving the full meaning of a sentence. Not only does the traditional unsupervised machine learning models disregard the occurring order or relation- ship of words in written texts but are also limited by the relatively small, fixed input size. Hence, the rationale for applying computationally sophisticated methods to textual data. (Acheampong, 2021)

When authors document their own emotions, they may neglect socially-created emotions that annotators may recognize. Such ML-based systems have their own limitations. In addition, it is currently uncertain how much text and label data is necessary for training models based on self-reported mood data sets.. (Lee 2023) . Social networking sites and microblogging services, such as Facebook and Twitter, provide an unprecedented quantity of (politically pertinent) user-generated content. In recent years, the number of users on social networking sites and microblogging services such as Facebook and Twitter has increased exponentially. As a form of social media with a lengthier history, weblogs are also not to be overlooked. Twitter is a significant communication platform for evaluations and discussions, according to (Stieglitz, 2012).

A lexicon-based technique, a machine learning-based method, or a hybrid method will be utilized to conduct sentiment analysis. In this connected world, studies have been conducted to determine the exact accuracy, but the results appear ineffective. The effectiveness of sentiment classification is dependent on the size of the lexicon (dictionary), because as the lexicon expands, this method becomes less accurate and more time-consuming. (Mitra, 2020) People and companies are depending more and more on the easily accessible content in social media channels for decision-making as a result of the social media industry's explosive growth in terms of blogs, microblogs, comments, and postings on social networks, among other things. These days, if someone wants to buy a product, they don't need to ask friends and family what they think because there are a ton of user reviews and forum conversations available. A firm can now obtain information from potential clients without having to run opinion polls because a wealth of this information is readily available on the Internet (Nedjah , 2019).

While firms are aware of the influence of e-WoM, it is difficult for them to interpret and evaluate the significance of e-WoM in enhancing their sales performance. Therefore, recognizing reputation and status as two separate moderators may be important from the perspective of firms that sell products or services. Here, e-WoM refers to electronic word of mouth, which can be used in the context of movie evaluations, and movie producers are aware of how word of mouth can influence box office sales (Mun, 2022). Internet's influence has become extremely dominant. 33 percent of the younger generation relies on social media, according to a recent survey. Nineteen percent of the millennial generation obtained their news from social media platforms such as Twitter, Snapchat, and LinkedIn (Sivasankari, 2021).

It has long been acknowledged that trend analysis is crucial for formulating future projections. Recent technological advancements have facilitated this (Fehrenbacher, 2022). As a consequence, managers are more skeptical of social media data. A retweet is the forwarding of an original tweet by one user to another. These retweets are a common way for the Twitter community to spread fascinating posts and links. Twitter has garnered a great deal of attention from corporations due to its immense viral marketing potential. Twitter is increasingly used by news organizations to disseminate news updates through the community due to its vast reach. A number of businesses and organizations use Twitter and other micro-blogging services to advertise their products and distribute information to their stakeholders (Asur, 2010).

The correlation between review rating and review frequency and continuity is positive, as the motivation to express positive emotions positively influences the number of reviews (Lee S. &., 2018). In the equation for box office revenue, the following characteristics of movies are used as control variables: sales of same actors (star power), sales of same directors (director power), award, rating of movie, genre, nationality, sequel, scheduling of release, and sales of competing movies. We estimate the effects of movie characteristics in order to account for variations in average box office receipts. In general, the film industry has a life cycle of a few months, during which it is extremely challenging to be self-sufficient. The box office is the most significant source of revenue for a movie or the production team as a whole, particularly the first-week box office. If the estimated total box office cannot guarantee that the investment in the film will yield positive returns, the issuer can conduct official marketing and celebrity marketing on various social media platforms to mitigate the risks posed by the distribution; alternatively, they can reduce the investment in publicity and lower the cost (Zhang, 2022).

Firms may decide whether to adopt predictive analytics with emotion analysis based on the level of information complexity in reviews that their targeted customers encounter. Another implication is that online review platforms may design features to make information processing more fluent, which contributes to review helpfulness. Finally, the effects of emotions may be different across cultures. (Yu, 2023)

This paper talks about how Twitter text about movies was analyzed for mood and emotion. The model detects the user sentiments, whether ‘positive’ or ‘negative’ from Twitter users' posts using the Flair Class distribution and then using the NRC Lexicon, which classifies the emotions into eight basic emotions. Twitter data, being highly unstructured, needs pre-processing. Data gathering, data pre-processing, sentiment analysis, and emotion identification utilizing the NRC Lexicon are the four processing processes included in the system development to achieve this. The overall system design is shown in Figure 1.

3.1 Data Collection

People often post short, frequent updates on Twitter, which is a popular microblogging platform. Online social media (OSM) platforms let users express, communicate, and share their opinions, thoughts, views, and points of view about local and international issues, matters, and topics through text, image, audio, and video contributions. The data was collected from Twitter using snscrape, which is a library for social networking services (SNS) in Python. We have taken ten movies as (Figure 2) from the IMDB database wherein we have mentioned the IMDB rating along with the Indian collection. Ten movie reviews and opinions have been considered in this paper. 10000 prerelease tweets and 10000 post-release tweets have been collected over a 3-month period for both pre and post. The data includes the ‘date’, ‘username’ and ‘tweet’ everything in only ‘English’ language for the ease of understanding. Data has been downloaded in ‘.csv’ format. Although images and videos may be posted, This system only permits text posts that are either "emotional" or not. In order to use age, gender, education, and employment data to determine the group's emotional result, the system also collects other fundamental user data.

3.2 Data Preprocessing

To analyze the feelings and find out more about the emotions behind the tweets, the data needs to be pre-processed. Coming from a highly unstructured format, the punctuation was removed using the string. Punctuation which is a pre-initialized string in Python. Next the stop words were removed for which the ‘nltk’ package was imported in python specifically for the English language. Lemmatization is a method for preprocessing text that turns words into their basic or root form, which is called the lemma. The goal of lemmatization is to turn all a word's different forms into one basic form. This can help reduce the data's dimensionality, making it easier to examine and handle. Natural language processing activities such as text classification, sentiment analysis, and information retrieval benefit greatly from lemmatization. It may also be used to improve machine learning model performance by decreasing noise in the data. The WordNet lemmatizer, which is included in the Natural Language Toolkit (NLTK) module for Python, is used in this paper. Corrupted text can cause problems with text analysis and result in erroneous or misleading outcomes; thus, it was eliminated. Regular expressions are a technique for removing damaged text. Using a regular expression, this function eliminates all non-alphanumeric characters from the given text. Following that, frequent words were removed, which can also enhance the effectiveness of text analysis algorithms such as topic modeling or sentiment analysis. Because these algorithms frequently rely on detecting the most significant words in a document, deleting stop words can enhance their accuracy and efficiency. Rare words might be typos, misspellings, or other textual mistakes. Eliminating these terms can assist in minimizing text noise and make it simpler to locate key words. Text analysis methods like topic modeling and sentiment analysis frequently rely on detecting the most relevant elements in the text; deleting unusual words can assist to minimize noise and enhance our results. Finally, we remove the URLs because they have a lot of frequency in the text and URLs provide no meaningful information.¹

3.3 Sentiment Analysis

Sentiment analysis of movie reviews on Twitter can reveal useful information about how people feel about a certain film. It is used for reasons stated as:

Box Office Predictions: Twitter sentiment analysis can be used to forecast a movie's box office performance. Before a movie comes out, studios may look at how people feel about it and use that information to change their marketing plan.
Recognizing Trends: Looking at the tone of movie reviews on Twitter can help you figure out what's going on with popular movies. Studios may detect what sorts of movies are connecting with people and change their production schedule appropriately by monitoring the emotion surrounding different types of movies.
Feedback for improvement: Movie companies can also use Twitter sentiment analysis to find out what people like and dislike about their movies. These comments may be utilized to improve future films and the entire movie-going experience.
Influencer Marketing: Sentiment analysis of movie reviews on Twitter may also assist in discovering important people who are talking about a specific movie. Then, movie producers can work with these "influencers" to build buzz and get the word out about the movie.
Lastly, sentiment analysis of movie reviews on Twitter can assist film companies in listening to what people are saying about their films in real time. This can help them recognize possible problems and respond to them immediately.

In Python, "punkt" refers to the NLTK library's pre-trained tokenizer model. The punkt tokenizer is used to tokenize text data, which is the act of breaking down a string or document into individual words or phrases known as tokens. In this paper, we have used Flair Class Distribution for sentiment analysis.² It is a cutting-edge NLP framework created using PyTorch. It brings together the most recent research and makes it easy to use different embeddings for different NLP tasks. The pretrained sentiment model gives us a way to analyze sentiment without having to train it ourselves. Flair emotion is what creates the projected label with a confidence score. The confidence level runs from 0 to 1, with 1 being extremely confident and 0 being extremely unsure. Sentence tokenization is required before estimating the emotion of a text. Following that, we can observe that the sample text was divided into nine tokens. We get the count of positive and negative words both from the pre-release and post-release of the movies.

3.4 Emotion Detection

NRCLex is a lexicon-based method for detecting emotions in text. It is based on the Affective Lexicon of Canada's National Research Council (NRC), which is a list of words and the emotions they are associated with. The NRC Lex method uses a lexicon to evaluate each word in each text for its relationship with emotions. The emotion scores for each word are added together to form the text’s overall emotion score. This can be accomplished by simply adding the scores together or by employing more advanced aggregation techniques such as weighted averaging. Based on the total emotion score, the text is put into one or more groups based on how it makes you feel. The categories are fear, anger, anticipation, trust, surprise, sadness, disgust, and joy.³ Emotion detection has become a popular method for understanding the audience's reaction to movies. For this paper, we analyzed the pre- and post-release tweets of 10 different movies: "Brahmastra", "Gangubai", "Jug Jug Jiyo", "Kashmir Files", "KGF", "Liger", "Lal Singh Chadha", "RRR", "Runway 34", and "Vikram Veda". The emotion categories we used were fear, trust, anger, joy, surprise, disgust, and sadness.

Based on the emotion detection results, the movie "Brahmastra" had a higher count of fear-related tweets both before and after release. This suggests that fear was a prominent emotion associated with the movie among audiences on social media. However, there was a slight increase in the count of joy-related tweets after the movie's release, which could indicate that the movie was well-received and enjoyed by audiences. The decrease in the count of sadness-related tweets after the movie's release could also suggest that the movie had a positive impact on audiences and was not overly emotional or depressing.

"Gangubai" generated a lot of buzz on social media, with high counts of tweets across multiple emotions. The pre-release tweets indicate that audiences were anticipating the movie with a mix of fear, anger, and trust, perhaps due to the intense and emotional subject matter. After the movie's release, however, there was a significant shift in the emotions associated with it, with a substantial increase in joy-related tweets. This suggests that audiences found the movie to be entertaining, perhaps due to the strong performances or compelling storyline. The increase in trust-related tweets could also suggest that the movie resonated with audiences on a deeper level, inspiring a sense of confidence or admiration.

For Jug Jug Jiyo pre-release tweets suggest that audiences were experiencing a mix of emotions in anticipation of the movie, with fear and trust being the most prominent. This could indicate that audiences were either unsure or excited about the movie and had high expectations for it. After the movie's release, there was a notable decrease in the count of fear-related tweets, which could suggest that the movie was less suspenseful or intimidating than audiences had anticipated. Instead, there was a significant increase in joy-related tweets, indicating that audiences found the movie to be enjoyable and entertaining. The increase in trust-related tweets could also suggest that the movie resonated with audiences on a deeper level, perhaps inspiring a sense of confidence or admiration.

For Kashmir Files, pre-release tweets indicate that audiences had mixed emotions about the movie, with fear and trust being the most prominent. The high count of trust-related tweets could indicate that audiences had faith in the movie's ability to deliver a powerful and impactful story. However, there was also a significant count of anger-related tweets, which suggests that audiences were potentially concerned about the movie's subject matter. After the movie's release, there was a decrease in the count of fear-related tweets, which could suggest that audiences found the movie to be less intimidating or suspenseful than they had anticipated. Additionally, there was a notable increase in the count of anger-related tweets, which could suggest that the movie was successful in creating a strong emotional response among audiences. However, there was also an increase in the count of disgust-related tweets, which could indicate that some audiences found the movie to be distasteful or offensive.

Pre-release tweets of KGF show that the movie generated a high level of fear among audiences, which could be interpreted as a positive sign for a movie in this genre. The relatively low count of trust-related tweets could suggest that audiences were uncertain about what to expect from the movie. However, the relatively high count of joy-related tweets indicates that audiences were looking forward to seeing the movie and were excited about it. After the movie's release, there was a decrease in the count of fear-related tweets, which could be interpreted as a sign that the movie met audience expectations in terms of delivering an intense and thrilling experience. The increase in the count of joy-related tweets also suggests that audiences found the movie to be entertaining and enjoyable. The increase in the count of anger-related tweets could indicate that some audiences were disappointed with certain aspects of the movie or that they felt the movie didn't fully meet their expectations.

The high count of fear-related tweets before the release of "Liger" could indicate that audiences were excited about the movie's potential to deliver a thrilling and suspenseful experience. However, the low count of trust-related tweets could suggest that audiences were not entirely sure about what to expect from the movie. After the movie's release, the decrease in fear-related tweets could be interpreted as a sign that the movie delivered on its promise to provide audiences with an exciting experience. The increase in joy-related tweets could also suggest that audiences found the movie to be entertaining and enjoyable. However, the increase in anger-related tweets could indicate that some audiences were disappointed with certain aspects of the movie, which may require further investigation to understand what aspects were responsible for this reaction.

The pre-release tweets of Lal Singh Chadha show a mix of emotions, with fear and trust being the most prominent ones, followed by joy and anger. However, after the release, we see a shift in the emotions, with joy being the most dominant one, followed by fear, trust, and anger. This indicates that the movie has performed well and has been well-received by the audience. The increase in joy and decrease in fear and trust could suggest that the audience found the movie to be enjoyable and entertaining, rather than intense and suspenseful. The increase in anger and disgust in the post-release tweets could suggest that the movie may have been controversial or polarizing among some viewers.

For RRR movie fear was the most prominent emotion among Twitter users before the release of the movie, with over 6,600 tweets expressing fear. However, after the release, the fear tweets reduced by 985, which can be considered a positive outcome. Trust was the second most prominent emotion before the release with 1872 tweets, and it increased after the release by 495 tweets, which is a good sign for the movie's reputation and credibility. Anger and joy were less prominent among Twitter users, with only 100 and 273 tweets respectively before the release. However, after the release, both emotions increased considerably, with anger increasing to 720 tweets, and joy increasing to 544 tweets. Surprise, disgust, and sadness were also present among Twitter users, but with lower counts than other emotions. Surprisingly, the count of tweets expressing surprise increased after the release, whereas the counts for disgust and sadness decreased after the release. Overall, these numbers suggest that the movie "RRR" generated a positive response among Twitter users, with increased trust and positive emotions after the release, and reduced fear and negative emotions.

Fear and anger were the two most prominent emotions of Runway 34 before and after the movie release. However, there was a significant decrease in fear and anger post-release, indicating that the movie might have addressed the issues causing these emotions in the pre-release period. Trust was also a significant emotion among the tweets, but there was a slight decrease in trust post-release. This might suggest that some viewers had different expectations from the movie that were not met, resulting in a decrease in trust. Joy and surprise were relatively low emotions in both pre-release and post-release periods, but there was a slight increase in joy post-release. This indicates that the movie might have been successful in delivering an enjoyable experience to the viewers. Disgust and sadness were relatively low emotions in both pre-release and post-release periods, but there was a slight increase in disgust post-release. This might suggest that some viewers were disappointed with the movie, resulting in an increase in disgust.

For Vikram Vedha fear and anger decreased while trust increased: The pre-release count of tweets for fear and anger were higher than the post-release count, while trust showed an increase in post-release. This suggests that the movie may have met the expectations of the audience in terms of delivering a satisfactory experience, thereby reducing their fear and anger. Joy and sadness increased: The post-release count of tweets for joy and sadness were higher than the pre-release count, which suggests that the movie may have been able to evoke strong emotions from the audience. Disgust showed a significant increase: The post-release count of tweets for disgust was considerably higher than the pre-release count. This may indicate that some aspects of the movie were not well received by the audience, causing them to express their displeasure on social media.

The results show that fear and trust were the most prevalent emotions across all movies. This indicates that audiences were concerned about what they would see and whether it would meet their expectations. Additionally, the post-release tweets showed higher levels of trust, suggesting that the movies generally met their audience's expectations.

Anger was a relatively low emotion for all movies, except for "Kashmir Files" and "Lal Singh Chadha". For "Kashmir Files", the anger increased post-release, indicating that the movie may have stirred controversy. For "Lal Singh Chadha", the anger increased significantly post-release, suggesting that some audience members did not enjoy the movie. Joy was another high-scoring emotion across all movies. This indicates that audiences found the movies to be enjoyable and engaging. The movies that received the highest joy scores were "Lal Singh Chadha", "Vikram Veda", and "Jug Jug Jiyo". Surprise, disgust, and sadness were relatively low-scoring emotions across all movies. This suggests that audiences were not particularly surprised, disgusted, or saddened by any of the movies.

Overall, the results indicate that the movies generally met audience expectations, with higher levels of trust and joy and lower levels of anger, surprise, disgust, and sadness. However, some movies received higher scores in certain emotions than others, suggesting that they may have elicited different reactions from the audience.

For a manager or producer, these results can be used to understand how their movie performed with the audience and to identify areas for improvement in future projects. For example, if a movie received high scores in anger, it may be necessary to address the controversy or issues raised in the movie. Similarly, if a movie received low scores in joy, it may be necessary to reconsider the plot or characters to make it more engaging for the audience.

The results of sentiment analysis and emotion detection of the pre-release and post-release tweets of the ten movies provide valuable insights for the film industry, especially in terms of marketing and audience engagement.

Firstly, the pre-release sentiment analysis suggests that movie promotions should focus more on building positive sentiments than addressing negative ones. This is because nine out of ten movies had more pre-release positive tweets than negative ones, indicating that the audience was generally excited about the upcoming releases. Therefore, film marketers should leverage this enthusiasm to create a buzz around the movies, rather than trying to counteract negative perceptions.

Secondly, the post-release sentiment analysis reveals that the actual movie experience can influence audience sentiments, as shown by the changes in tweet counts. For instance, in the case of the movies Brahmastra and Liger, the post-release negative tweets exceeded the pre-release negative tweets, indicating that the audience's expectations were not met. On the other hand, for movies like Gangubai and KGF, the post-release positive tweets outnumbered the pre-release ones, indicating that the movies exceeded the audience's expectations. Therefore, film producers and directors should pay attention to the audience's preferences and feedback to improve their future productions.

Thirdly, the emotion detection results shed light on the audience's emotional responses to the movies. For instance, the post-release emotion detection of the movie Runway34 revealed that despite having a similar number of positive and negative tweets, the majority of the negative tweets expressed sadness and fear. This suggests that the movie might have had a poignant or scary theme that resonated with some audience members but not others. Therefore, film producers and marketers should consider the emotional appeal of their movies and target the appropriate emotions in their promotions.

Overall, the sentiment analysis and emotion detection results provide a nuanced understanding of the audience's perceptions and emotional responses to the movies, which can guide film industry professionals in developing effective marketing strategies, improving movie quality, and engaging with the audience more effectively.

In this study, we conducted sentiment analysis and emotion detection on the pre- and post-release tweets of 10 popular Bollywood movies, with the aim of understanding the audience perception and emotional response towards these movies.

Overall, our results showed that most of the movies had a more positive sentiment and higher levels of joy and trust in the post-release tweets compared to the pre-release tweets. However, some movies had more negative sentiment and higher levels of anger and fear in the post-release tweets, indicating potential issues with the movie's marketing or audience reception.

Specifically, the movies "Brahmastra", "Kashmir Files", "Lal Singh Chadha", and "RRR" showed a decline in positive sentiment and an increase in negative sentiment in the post-release tweets. On the other hand, the movies "Gangubai", "Jug Jug Jio", "KGF", "Runway34", and "Vikram Veda" had a more positive sentiment and higher levels of joy and trust in the post-release tweets.

Emotion detection has become a popular method for understanding the audience's reaction to movies. For this paper, we analyzed the pre- and post-release tweets of 10 different movies: "Radhe", "Gangubai", "Jug Jug Jiyo", "Kashmir Files", "KGF", "Liger", "Lal Singh Chadha", "RRR", "Runway 34", and "Vikram Veda". The emotion categories we used were fear, trust, anger, joy, surprise, disgust, and sadness.

Interestingly, our emotion detection results showed that the most prevalent emotions in both the pre- and post-release tweets were joy and trust, followed by surprise and anger. Fear and disgust were the least prevalent emotions. This suggests that audiences generally respond positively to movies that evoke feelings of joy and trust, and that surprise and anger can also contribute to a movie's emotional impact.

Based on these findings, we recommend that movie marketers and producers pay close attention to the audience perception and emotional response towards their movies and take necessary steps to improve the marketing and audience engagement strategies. This may include early and effective promotion, engaging with the audience on social media platforms, and taking feedback from the audience to make necessary changes in the movie production process.

In conclusion, our study provides valuable insights into the audience perception and emotional response towards popular Bollywood movies. Our results suggest that both sentiment analysis and emotion detection can be useful tools for movie marketers and producers to understand the audience's feedback and improve the movie's performance.

The study only analyzes Twitter data, which may not represent the entire population's sentiments towards the movies. Therefore, the findings of this study cannot be generalized to the entire population. The pre-processing of the Twitter data might not have been entirely accurate, which can impact the accuracy of the sentiment and emotion analysis. The pre-processing method used in this study might have missed out on some crucial information or erroneously removed relevant information. The NRC Lexicon used in this study only classifies emotions into eight basic emotions, which might not be sufficient to capture the complex emotions associated with movies. It is possible that some emotions might have been overlooked or classified inaccurately. This study only focuses on sentiment analysis and emotion detection related to movies. It does not consider other aspects of movies such as storyline, acting, cinematography, etc. The Twitter users who were included in this study might not be representative of the general population, which could lead to biased results. The study only analyzes English language tweets, which might not reflect the sentiments of non-English speakers. Additionally, the emotions and sentiments expressed on Twitter might be influenced by cultural and regional factors, which could impact the accuracy of the findings.

Acheampong, F. A., Nunoo-Mensah, H., & Chen, W. (2021, February 8). Transformer models for text-based emotion detection: a review of BERT-based approaches. Artificial Intelligence Review, 54(8), 5789–5829. https://doi.org/10.1007/s10462-021-09958-2
Al-Laith, A., & Alenezi, M. (2021, February 19). Monitoring People’s Emotions and Symptoms from Arabic Tweets during the COVID-19 Pandemic. Information, 12(2), 86. https://doi.org/10.3390/info12020086
Alswaidan, N., & Menai, M. E. B. (2020). A survey of state-of-the-art approaches for emotion recognition in text. Knowledge and Information Systems, 62, 2937-2987.
Angel Deborah, S., Mirnalinee, T. T., & Rajendram, S. M. (2021). Emotion analysis on text using multiple kernel gaussian. Neural Processing Letters, 53, 1187-1203.
Anwar, Z. et al. (2022) “Context-based emotion predictor: A decision- making framework for Mobile Data,” Mobile Information Systems, 2022, pp. 1–12. Available at: https://doi.org/10.1155/2022/6488848
Asghar, M. Z., Khan, A., Bibi, A., Kundi, F. M., & Ahmad, H. (2017, September 5). Sentence-Level Emotion Detection Framework Using Rule-Based Classification. Cognitive Computation, 9(6), 868–894. https://doi.org/10.1007/s12559-017-9503-3
Asur, S., & Huberman, B. A. (2010, August). Predicting the Future with Social Media. 2010 IEEE/WIC/ACM International Conference on Web Intelligence andIntelligent Agent Technology. https://doi.org/10.1109/wi-iat.2010.63
Bharti, S. K., Varadhaganapathy, S., Gupta, R. K., Shukla, P. K., Bouye, M., Hingaa, S. K., & Mahmoud, A. (2022, August 23). Text-Based Emotion Recognition Using Deep Learning Approach. Computational Intelligence and Neuroscience, 2022, 1–8. https://doi.org/10.1155/2022/2645381
Cabanac M. (2002). What is emotion?. Behavioural processes, 60(2), 69–83. https://doi.org/10.1016/s0376-6357(02)00078-5
Dessì, D., & Diego, R. R. (2021, November 1). LexTex: a framework to generate lexicons using WordNet word senses in domain specific categories. Journal of Intelligent Information Systems, 59(1), 21–44. https://doi.org/10.1007/s10844-02100679-0
Farhat Ullah , Xin Chen , Syed Bilal Hussain Shah ,*, Saoucene Mahfoudh , Muhammad Abul Hassan , and Nagham Saeed, A Novel Approach for Emotion Detection and Sentiment Analysis for Low Resource Urdu Language Based on CNN-LSTM, Electronics 2022, 11, 4096. https://doi.org/10.3390/electronics11244096
Fehrenbacher, D. D., Ghio, A., & Weisner, M. (2022, November 23). Advice Utilization from Predictive Analytics Tools: The Trend is Your Friend. European Accounting Review, 1–26. https://doi.org/10.1080/09638180.2022.2138934
Graterol, W., Diaz-Amado, J., Cardinale, Y., Dongo, I., Lopes-Silva, E., & Santos-Libarino, C. (2021, February 13). Emotion Detection for Social Robots Based on NLP Transformers and an Emotion Ontology. Sensors, 21(4), 1322. https://doi.org/10.3390/s21041322
Guo, J. (2022) Deep learning approach to text analysis for human emotion detection from big data. Journal of Intelligent Systems, Vol. 31 (Issue 1), pp. 113-126. https://doi.org/10.1515/jisys-2022-0001
Hu, Y. H., & Chen, K. (2016). Predicting hotel review helpfulness: The impact of review visibility, and interaction between hotel stars and review ratings. International Journal of Information Management, 36(6), 929–944. https://doi:10.1016/j.ijinfomgt.2016.06.003
Kumar, A., & Jain, A. K. (2022, July 11). Emotion detection in psychological texts by fine-tuning BERT using emotion–cause pair extraction. International Journal of Speech Technology, 25(3), 727–743. https://doi.org/10.1007/s10772-022-09982-9
Kusal, S., Patil, S., Kotecha, K., Aluvalu, R., & Varadarajan, V. (2021). AI Based Emotion Detection for Textual Big Data: Techniques and Contribution. Big Data and Cognitive Computing, 5(3), 43. https://doi.org/10.3390/bdcc5030043
Kušen, E., Strembeck, M., & Conti, M. (2019). Emotional valence shifts and user behavior on Twitter, Facebook, and YouTube. Influence and Behavior Analysis in Social Networks and Social Media, 63-83.Mao X, Chang S, Shi J, Li F, Shi R (2019) Sentiment-aware word embedding for emotion classification. Appl Sci 9(7):1334. https://doi.org/10.3390/app9071334
Lee, S. J., Lim, J., Paas, L., & Ahn, H. S. (2023, January 26). Transformer transfer learning emotion detection model: synchronizing socially agreed and self-reported emotions in big data. Neural Computing and Applications. https://doi.org/10.1007/s00521-023-08276-8
Lee, S., & Choeh, J. Y. (2018, January 31). The interactive impact of online word-of-mouth and review helpfulness on box office revenue. Management Decision, 56(4), 849–866. https://doi.org/10.1108/md-06-2017-0561
Majeed, A., Beg, M. O., Arshad, U., & Mujtaba, H. (2022, May 21). Deep-EmoRU: mining emotions from roman urdu text using deep learning ensemble. Multimedia Tools and Applications, 81(30), 43163–43188. https://doi.org/10.1007/s11042-02213147-w
Mitra, A. (2020, September 18). Sentiment Analysis Using Machine Learning Approaches (Lexicon based on movie review dataset). Journal of Ubiquitous Computing and Communication Technologies, 2(3), 145–152. https://doi.org/10.36548/jucct.2020.3.004
Mubeen, S., Kulkarni, D. N., Tanpoco, M. R., Kumar, D. R., M, L. N., & Dhope, T. (2022, December 31). Linguistic Based Emotion Detection from Live Social Media Data Classification Using Metaheuristic Deep Learning Techniques. International Journal of Communication Networks and Information Security (IJCNIS), 14(3), 176–186. https://doi.org/10.17762/ijcnis.v14i3.5604
Mun, H., Lee, C. H., Jung, H., & Yasin, C. (2022, September 1). Clash of reputation and status in online reviews. Information Technology and Management, 24(1), 55–77. https://doi.org/10.1007/s10799-022-00374-8
Nedjah, N., Santos, I., & de Macedo Mourelle, L. (2019, April 3). Sentiment analysis using convolutional neural network via word embeddings. Evolutionary Intelligence,15(4), 2295–2319. https://doi.org/10.1007/s12065-019-00227-4
Ortony, A., Clore, G., & Collins, A. (1988). The Cognitive Structure of Emotions. Cambridge: Cambridge University Press. doi:10.1017/CBO9780511571299B. Garett, “Brain and behaviour: an introduction to biopsychology,” 2nd ed., SAGE Publications, 2009.
Pempek, T. A., Yermolayeva, Y. A., & Calvert, S. L. (2009, May). College students’ social networking experiences on Facebook. Journal of Applied Developmental Psychology, 30(3), 227–238. https://doi.org/10.1016/j.appdev.2008.12.010
Plaza-del-Arco, F. M., Jiménez Zafra, S. M., Montejo Ráez, A., Molina González, M. D., Ureña López, L. A., & Martín Valdivia, M. T. (2021). Overview of the EmoEvalEs task on emotion detection for Spanish at IberLEF 2021. https://doi.org/10.26342/2021-67-13
Serrano-Guerrero J., Alshouha B., Romero F.P., Olivas J.A. Affective Knowledge-enhanced Emotion Detection in Arabic Language: A Comparative Study. Journal of Universal Computer Science, vol. 28, no. 7 (2022), 733-757. DOI: 10.3897/jucs.72590.
Sharma, S., Kumar, P., & Kumar, K. (2017, November). LEXER: Lexicon based emotion analyzer. In Pattern Recognition and Machine Intelligence: 7th International Conference, PReMI 2017, Kolkata, India, December 5-8, 2017, Proceedings (pp. 373-379). Cham: Springer International Publishing.
Sivasankari, S., & Vadivu, G. (2021, July 24). Tracing the fake news propagation path using social network analysis. Soft Computing, 26(23), 12883–12891. https://doi.org/10.1007/s00500-021-06043-2
Stieglitz, S., & Dang-Xuan, L. (2012, August 25). Social media and political communication: a social media analytics framework. Social Network Analysis and Mining, 3(4), 1277–1291. https://doi.org/10.1007/s13278-012-0079-3
Swe, T. M., & Wah, N. L. (2021, April 1). Emotion detection on Myanmar texts. International Journal of Electrical and Computer Engineering (IJECE), 11(2), 1570. https://doi.org/10.11591/ijece.v11i2.pp1570-1577
Tesfagergish, S. G., Kapočiūtė-Dzikienė, J., & Damaševičius, R. (2022, August 29). Zero-Shot Emotion Detection for Semi-Supervised Sentiment Analysis Using Sentence Transformers and Ensemble Learning. Applied Sciences, 12(17), 8662. https://doi.org/10.3390/app12178662
Vijayvergia, A., & Kumar, K. (2021, June 3). Selective shallow models strength integration for emotion detection using GloVe and LSTM. Multimedia Tools and Applications, 80(18), 28349–28363. https://doi.org/10.1007/s11042-021-10997-8
Volkova, S., Charles, L. E., Harrison, J., & Corley, C. D. (2017, June 8). Uncovering the relationships between military community health and affects expressed in social media. EPJ Data Science, 6(1). https://doi.org/10.1140/epjds/s13688-017-0102-z
Wei, W. J. (2019, June 22). Development and evaluation of an emotional lexicon system for young children. Microsystem Technologies, 27(4), 1535–1544. https://doi.org/10.1007/s00542-019-04425-z
Zahid Anwar, Rashid Jahangir, Muhammad Asif Nauman, Roobaea Alroobaea, Sabah M. Alzahrani, Ihsan Ali, "Context-Based Emotion Predictor: A Decision- Making Framework for Mobile Data", Mobile Information Systems, vol. 2022, Article ID 6488848, 12 pages, 2022. https://doi.org/10.1155/2022/6488848
Zhang, C., Tian, Y. X., & Fan, Z. P. (2022, April). Forecasting the box offices of movies coming soon using social media analysis: A method based on improved Bass models. Expert Systems with Applications, 191, 116241. https://doi.org/10.1016/j.eswa.2021.11624
Yu, Y., Yang, Y., Huang, J., & Tan, Y. (2023, March 1). Unifying Algorithmic and Theoretical Perspectives: Emotions in Online Reviews and Sales. MIS Quarterly, 47(1), 127–160. https://doi.org/10.25300/misq/2022/16600

[1] https://colab.research.google.com/drive/15FUbFBkTTlNEIYMj-4GJM26EvdhpbMKW#scrollTo=Xh4Z20JpyeTu

[2] https://colab.research.google.com/drive/147V-TTJITsNjGqo0RSciEG4YT_CQNjAg#scrollTo=10zz4F3xgodk

[3] https://colab.research.google.com/drive/1CfCP9l140YU3buO4XZQ3R6JFDzq8EBgT#scrollTo=2v4UTpA9yAWy

No competing interests reported.

Table123.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Impact of tweets on Indian movies performance: An integrative approach using sentiment analysis and emotion detection

Status:

Version 1

Abstract

Figures

1. Introduction

2. Literature Review

3. Research Methodology

4. Implications

5. Conclusion

6. Limitations

References

Footnote

Additional Declarations

Supplementary Files

Status:

Version 1