Multimedia Lu Xun literature online learning based on deep learning

As a great Chinese thinker and writer in the twentieth century, Lu Xun and his literary works are widely known. However, as a successful cultural communication activist, editor and publisher, we still need to conduct in-depth research on Lu Xun in many aspects. Therefore, based on deep learning, this paper constructs an online multimedia learning system of Lu Xun literature. This system takes the relationship between classical Lu Xun literature and modern multimedia technology as the research object, and compare the calculation effect of other different types of algorithms and this dhraa algorithm. Through the availability of data, the dhraa algorithm is significantly better than other algorithms in the recommendation accuracy, thus proving its effectiveness. This system is managed by two servers and one system. The two servers are database server and web server, respectively. After testing, the system has good bearing capacity, can make up for the limited processing capacity of the server, and ensure the system has high performance. Its performance characteristics also show that the system achieved the expected performance. This paper systematically combines Lu Xun’s literature with modern multimedia, which can provide online learning services for Lu Xun’s literature lovers, thus helping scholars to expand Lu Xun’s research field and academic vision. This paper designs an effective online learning system of Lu Xun’s literature by combining deep learning, multimedia technology and Lu Xun’s literature.


Introduction
Although Lu Xun is a famous writer, he is different from many other writers. His original intention of becoming a writer is not to pursue literary achievements, but to have more specific and lofty ideas: namely, ''promoting social development'', ''describing life'' and ''enlightening thoughts'' (Liu 2009). The most important difference between Lu Xun and other writers is that he not only pays attention to the content of his writing, but more importantly, he will pay attention to the social problems caused by the creation of his works from the perspective of communication, and how to spread the progressive ideas of ''promoting social development'' and ''describing life'', so as to enlighten the people's wisdom (Wang and Zhang 2017). Therefore, he took an active part in the field of editing and publishing to spread sacred culture. The introduction of modern multimedia into the field of Lu Xun's literary research will certainly change the research methods and perspectives, and thus change the overall concept of Lu Xun's research in the history of modern Chinese literature (Tang 1992). This paper proposes to integrate Lu Xun's literature with online multimedia, not to study the official media, that is, the carrier materials and communication methods of modern newspapers and magazines on Lu Xun's literature, but to study the impact of this media on Lu Xun's early works, as well as the impact of Lu Xun's creation and creative thinking and the internal relationship between the study of modern multimedia and Lu Xun's literary creation (Seuk-Pyo 2020; Hashimoto 2019). Since Lu Xun's works were published in modern newspapers and magazines, there have been many articles on Lu Xun's works and thoughts, but few works examining Lu Xun and his creative thoughts from the perspective of the media (Kaldis 2020). Therefore, the new research method of Lu Xun's literature will certainly rewrite the research field of modern Chinese literature. Therefore, based on deep learning, this paper constructs a multimedia online learning system of Lu Xun's literature. By adding the personal recommendation mode to the system, users can have more enthusiasm and initiative in the process of learning Lu Xun's literary works. The system can establish a user model according to the user's learning behavior. Therefore, we can recommend the Lu Xun Literature Resources to users in a targeted way to enhance their learning motivation and improve their learning enthusiasm. The online multimedia Lu Xun Literature learning system not only provides users with a platform to learn Lu Xun's literary works anytime and anywhere, but also recommends Lu Xun's literature resources to Lu Xun's literature lovers according to their interest characteristics, effectively solving the problem of text overload in the online learning system, thus improving the user experience and learning efficiency.

Related work
A deep recommendation algorithm (dhraa) is proposed in the literature, which has a hierarchical attention mechanism combined with auxiliary information. To solve the problem of data sparsity and score prediction, this paper proposes a dhraa algorithm, which uses convolutional neural network (CNN) to process the user and course comment information, as well as the course name, course description information, teacher organization and other auxiliary information (Yin and Zhang 2020). The NRPA based model uses word level attention mechanism and comment level attention mechanism to extract the information of user comments and vector features, and uses the attention factor decomposition machine (AFM) algorithm to calculate the cross combination features to complete score prediction (BenÍtez et al. 2013).Comparing the baseline model with the dhraa algorithm proposed in this paper in the environment of MOOC dataset and Amazon public dataset, we can see that the algorithm designed in this paper can better recommend the target requirements (Kellogg and Edelmann 2015). The literature combines the proposed dhraa algorithm with the online learning platform, and uses the front end and back end separation design model to complete the system development based on the Django framework (Kastrati et al. 2020). The system includes the online learning module, the course collection module, the course recommendation module, the resource download module, the course module, and the backend management module. Finally, the online learning system is tested by using the hyload automatic test tool. The test results show that the system is practical, stable and achieves the expected goal (Xu and Jaggars 2013). The literature selects the distributed crawler scrapy framework for data collection by comparing different data crawling methods, and selects the MOOC platform of China University for data collection by analyzing the online learning platform. Use the scrapy framework to design hierarchical collection rules, simulate browser request behavior, obtain non interface crawling data, analyze course list data, course comment data, user information, comment score information, and realize corresponding data collection and data preprocessing operations (Garcia-Loro et al. 2020). This paper designs a system structure with low coupling, easy expansion, high availability and high performance, and describes and implements the module design of various servers in the system (Kim 2015). The literature has designed a class of course recommendation engines, which analyze based on the course information data and obtain the final recommendation candidate sets by clustering implicit feature vectors and calculating various recommendation scenarios, including relevant course Recommendation sets, popular course recommendation sets and user-defined Recommendation sets. Such Recommendation sets can provide a data base for the personalized recommendation of the system (Kim et al. 2002;Jain et al. 2022;Sangaiah et al. 2023;Samantra et al. 2022).

Deep learning algorithm
The amount of online data continues to expand with the continuous development of 5G technology, and its structure is gradually complex and dynamic. Therefore, recommender systems that can operate in the context of information overload emerge as the times require. At the same time, deep learning technology has also been widely used in the fields of image analysis and speech recognition. Deep learning technology can promote the transformation of recommendation system and promote the industrial application revolution. This technology can directly extract features based on the target content and process noisy data. It has better anti-noise ability for dynamic or sequential data. This is because the neural network has the above characteristics and can achieve good results when input into the recommendation system. MLP (Multilayer Perceptron) is a feedforward neural network model. Although the model structure is simple, it runs efficiently. It has many hidden layers and is widely used and has good performance in the application of recommendation systems. The three-layer MLP network structure is shown in Fig. 1.
The self-encoder can learn the correlation of the target data based on this system structure and create a data compression type representation in the hidden layer. Wherein, encoding Ø is used to represent the data exchange process between the input layer and the hidden layer, and decoding / is used to represent the data exchange process.

Mathematical model
NRPA model is usually divided into two parts: comment encoder and user item encoder. The comment encoder calculates the feature vector representation of the comment as a function of the word level attention vector of the hierarchical attention mechanism. The comment level attention vector mechanism aggregates all comments to obtain vector representations of users and objects. The hierarchical attention mechanism of NRPA maps the user ID and item ID to a low dimensional space based on the uniqueness of the user and item ID, and then calculates and generates word level attention vectors and comment level attention vectors using a multi-layer perceptron model. The hierarchical attention mechanism extracts word level attention vectors and comment level attention vectors from user comments, which, respectively, represent the importance of each word in the comments and the importance of the comments as feature vectors of users and objects.
The generated word level attention vector of the user is calculated using the following formula: where w represents the mlp1 parameter, b1 represents the offset value, and ReIU is the activation function. The attention weight of each word in the comment text is calculated using the following formula: where W a represents the weight matrix, and gk represents the kth word vector after encoding. Then, according to the attention weight of each word in the user comment text and the vector representation point of each word, the vector representation of the user comment on the object is obtained: The comment level attention vector is calculated using the following formula: The attention weight of the j-th comment of the user u is calculated from the comment text using the following formula: where W b is the user's comment level attention parameter, d u , j is the user's comment vector u on the jth item. Adding the comment text vector and the attention weight to obtain the user feature vector: The input sequence of LSTM network is the implicit eigenvector of user I according to the time sequence skill course, and the output is the implicit eigenvector of user I.
where W 1 , W 2 , W 3 and W 4 are mapping matrices. B 3 , and B 4 are offsets.
Suppose the system has N users and M courses, U [ R D*N is the implicit eigenvector matrix of all users, and V [ R D*M is the implicit eigenvector matrix of all courses. The probability matrix decomposition model assumes that the real rating matrix R satisfies the conditional probability shown in Eq. (10): The probability matrix decomposition model assumes that the implicit eigenvectors of users and courses are independent of each other, and the prior probability distribution follows the Gaussian distribution of zero mean: The prior probability distribution of implicit eigenvectors of users and courses is the key to the combination of deep learning model and probability matrix decomposition model. The traditional probability matrix decomposition model relies on Gaussian distribution to randomly initialize the implicit eigenvectors of users and courses. However, this paper uses LSTM network and attention CNN network to initialize the implicit eigenvectors of users and courses respectively, so the default user eigenvectors and initialization courses become: where W u is a general mapping matrix representation of the LSTM network. W V is a general representation of the attention CNN network mapping matrix. X u is the input set of the LSTM network. X V is the input set of the attention CNN network. After modifying the above formula, we fuse the deep learning model and the probabilistic matrix decomposition model to obtain a probabilistic matrix decomposition recommendation model integrating deep learning. According to the Bayesian formula, the posterior probability of the implicit eigenvector matrix of users and courses can be obtained, which satisfies the following formula: The probability matrix decomposition recommendation model integrating deep learning solves the posterior maximum likelihood estimation of the implicit eigenvector matrices u and V of users and courses related to the known score matrix R, namely: max p U; V; W U ; W 1 jR; X; r 2 ; r 2 V ; r 2 1 ; r 2 H C ; r 2 To facilitate the derivation, we take the natural logarithm of formula (16) to obtain: Since the mapping matrices of LSTM and attention CNN networks are randomly initialized according to the Gaussian distribution, the elements of the mapping matrix in Eq. (17) can be ignored. Then, the maximum value of Eq. (17) can be replaced by the equivalent value by solving the minimum value of the following equation: In this paper, the coordinate descent algorithm is used to find the minimum value of F. first, the partial derivatives of the default user feature vector UI and the default course feature vector VJ of F are calculated, and the expressions of UI and VJ are updated to obtain Eqs. (19) and (20). Then, the default user feature vector and the default course feature vector are iteratively updated according to Eqs. (19) and (20) until the value of F converges to obtain the default end user feature vector and the default course feature vector. Finally, user implicit eigenvector and course implicit eigenvector are used for recommendation.
where IK denotes a k-dimensional identity matrix.

Experimental simulation
The root mean square error is calculated as the arithmetic square root of the mean square error: The average absolute error is the average value of the absolute error, and the calculation formula is as follows: where observed is the actual value of user I's score on course J, predicted is the predicted value of user I's score on course J, and N is the total number of scores in the test set. Table 1 shows the size of the data sets used in the experiment and the parameter settings of each model.
According to the parameter values set above, deepconn algorithm, NRPA algorithm, NRPA algorithm only mixing auxiliary information and NRPA algorithm are, respectively, used in the crawled MOOC data set of Chinese University and Amazon goods data set_ The AFM algorithm is compared with the dhraa algorithm proposed in this paper. In order to effectively evaluate the experimental effect, the parameters with the best effect in the process of model formation were selected. The comparative experimental results of the two datasets are shown in Table 2.
Through the analysis of the deepconn algorithm, NRPA algorithm, NRPA algorithm only fusing auxiliary information and NRPA algorithm proposed in this paper_ The comparison experiment of five models of AFM algorithm and dhraa algorithm shows that the dhraa model proposed in this paper is superior to other models in terms of recommendation accuracy, thus proving the effectiveness of the model proposed in this paper.
The smaller the RMSE result, the higher the prediction accuracy. The process of calculating RMSE is as follows: where R ij is the actual score of user I on course J, and t is the score in the test set.
Test on the test set, and the final RMSE results are shown in Table 3.
From the experimental results in Table 3, it can be seen that the recommendation model proposed in this paper is superior to the traditional probability matrix decomposition model, the collaborative topic model, the collaborative deep learning model, the matrix convolution decomposition model, the model only combining attention and the CNN model. At the same time, it can be seen that when only attention CNN or LSTM network is combined, although the recommendation accuracy is improved, some defects still appear. The RMSE of the attention CNN model is superior to the traditional probability matrix decomposition model, collaborative topic regression model, collaborative deep learning model and convolution matrix decomposition model, which shows that the attention CNN network can obtain more accurate implicit feature vectors of course description information and improve the accuracy of recommendation.
In order to verify the influence of implicit feature size on the model proposed in this paper, we set different implicit feature sizes to conduct experiments on the data set. The experimental results are shown in Fig. 2. to the present, Lu Xun's works have been included in Chinese middle school textbooks. It is not difficult to see that his works are all good works included in Chinese middle school textbooks and are advanced cultural ideas that nourish the growth of several generations. However, in the new century, the voice of denying Lu Xun gradually appeared. Some people say that Lu Xun's works should be removed from the times. Some people say that Lu Xun's articles are obscure and difficult to understand, and demand that Lu Xun's works be reduced or even completely eliminated in the teaching materials. Therefore, the author cannot help asking: does Lu Xun's literary works and his ideological spirit really suit our modern society? The reform of the newspaper body is a process of constantly adapting to the characteristics of the newspaper itself. With the development of modern newspaper industry, the cycle of news transmission has been shortened, and news comments have also changed. Short, fast and new have become the requirements of news comment writing, and Lu Xun's short essays and comments just conform to the development trend of news style, so that comments can be related to the latest news, and the meaning and information behind the news can be seen in time. Among the newspaper supplements and essays, the most popular ones are Shen Bao and free talk. Lu Xun published many essays for this purpose, and finally compiled three collections of pseudo free book, quasi fengyuetan and qijieting essays. In his essays published in Shen Bao free talk, he often wrote the time of the current events he commented on, or the date of the newspaper that published the news. Some essays were published one day after the news was released, or some essays were published within half a month after the news was released, and all the information and opinions of the whole event were collected. This kind of timely and quick comment fully reflects the artistic characteristics of ''current review'' based on news. Lu Xun's essays are not argumentative in an absolute sense. Its qualitative provisions indicate that they belong to Literature and ''poetry''. Therefore, the artistic nature of essays is irreplaceable, and the ''news and media'' of essays cannot erase its artistic nature. Therefore, the art of essays includes the expression of ''style of essays'' and ''interest of essays'' in the newspapers and periodicals, that is, the remarks with the characteristics of the style of essays described by Lu Xun. It can be seen that the influence of Lu Xun's works has continued to this day.
Shakespeare's works have infected and educated countless people. The British and even many Westerners think that Shakespeare is a cultural symbol. Many people made fun of him and said, ''I would rather lose the British Isles than Shakespeare.'' Shakespeare's Classics have become the eternal source of the English people's life spirit. In China, I believe Lu Xun is the same to us, so we should continue to study and study Lu Xun's literature. The path he took clearly shows that in China at the beginning of the twentieth century, outstanding intellectuals continued to advance on the road of pursuing truth and on the road of China's reform and national rejuvenation. If we want to understand that period, Lu Xun and his works are the best witness. At the same time, Lu Xun's spirit, as his basic proposition of culture, is still an important content of cultural innovation in our times. In today's vulgar fast food culture, Lu Xun's thoughts are like a cup of meaningful liquor, which is still worth stopping to savor. Lu Xun and his works always remind people of the humiliating history of China, which cannot be forgotten. We can become better in the future by summarizing and learning from the previous experience through education. We have the responsibility to educate this generation without worrying about food and clothing, so that they can always keep sober and remember the national shame. Only in this way can we make continuous progress in the process of national development. Therefore, it is of great significance to study and teach Lu Xun's works.

System development requirements
The requirements analysis puts forward precise and specific requirements for the target system, and shows the functions of the software system, that is, to determine the needs of users, and finally meet the following basic requirements: Reality The existing software and hardware facilities can meet the development needs of the project.
Consistency All system requirements should be identical and there should be no difference between them.
Integrity The requirements of each user must be collected completely.
Follow up Each function to be implemented can be customized according to the collected initial customer requirements.
Unambiguous Developers and users should have no conflict in understanding different system requirements.
Non verifiability A defined single functional requirement can be clearly verified.
Different system users have different permissions. Users of such multimedia learning websites can be divided into three categories, namely, system administrators, members and non-members.
(1) System administrator After successfully logging in the system, the system administrator can retrieve, query and manage all user information and resource information within the system, such as video information, and can retrieve the attention of major video resources. The system administrator can update the favorite resources of users according to the attention information to improve their learning efficiency and enthusiasm; (2) Member After successfully registering as a member, the user can log in to the system and enjoy all the services of the member. You can browse, check and manage all the courses and video resources in the system, you can freely watch the video resources you are interested in and leave messages according to your personal preferences, and you can also communicate with other users in this way to enjoy the interactive learning process. In addition, member users can also upload or download various learning resources according to their own needs, so that the system can realize resource sharing and interactive communication, and can effectively cultivate their self-learning ability; (3) Non-member If the user fails to register successfully, it is a non-member. After logging in the system as a member, you can only view all video learning resources or messages, but cannot upload or download.

System architecture design
In this Lu Xun literature online multimedia learning system, two servers are set up for system management. The two servers are database server and web server, respectively. The resource information cost can be reduced based on the load flow in the actual system application process. The general structure diagram of the overall network design is shown in Fig. 3 .

System function module design
The foreground function structure of Lu Xun literature multimedia online learning system is shown in Fig. 4: When a user logs into the website for the first time and tries to register as a member, he needs to click the login module of the webpage, that is, new user registration. The system will automatically jump to the user registration interface after the user clicks. In this interface, the user needs to fill in all the information and meet the requirements before he can become a full member. When registering, the user needs to fill in the user name first. This user name cannot be repeated with the user name of others; Secondly, a password needs to be set. The password needs to have certain security, that is, the combination of numbers, letters and symbols, and the two password filling settings must be consistent; Then fill in the commonly used e-mail box to receive mail; Finally, enter the ID card number correctly. Only if the above information is filled in correctly can the registration be successful.
After the registration is completed, the member can enter the system normally and enjoy relevant services and permissions by entering the user name and password again. The parameters filled in by the member will be sent to the background entry Aspx, and then in TB_ Look up the received user name in the login table. If the detection result is zero, it means that the user does not exist, and a corresponding prompt is given. If the user name appears in the table, it will then determine whether the password is correct. If the same method detects a password mismatch, a password error will be displayed. When the user name and password match, the user can enter the website to enjoy the membership rights. The system uses the session object to store the login information of the user.
Users can view the click through rate, release date, publisher and video content attribution of the video on the viewing video tutorial information page. At the same time, users can express their feelings after watching the video in the message box and communicate with other users. (Since the information of courses, audio tutorials and video tutorials are the same, the introduction of courses and audio tutorials is omitted here.) After watching the video tutorial, users can leave a message on the tutorial. Messages can be executed through Fig. 3 Network topology the click event of the ''talk'' button. Users can communicate and chat with each other through messages to improve the quality of tutoring and learning. When visitors browse and leave messages on the website, the speaker we see becomes a visitor. When a member user leaves a message, the speaker is the registered name of the member. In order to prevent illegal users from leaving too many messages on this site, occupying the database space and affecting the normal operation speed of the server, users must input the verification code every time they leave a message.

System test
The Lu Xun literature multimedia online learning system designed in this paper has a wide audience. Therefore, to fully meet its needs, the system itself needs to have excellent performance. Based on this point, this paper conducts a pressure test on the system's compressive performance, and the results are shown in Table 4. Since the background management system is only used by the administrator, it is not tested. Table 4 shows the average response time test results of the system pressure test: Because the high concurrent high-voltage measurement method can evaluate the average response time of the system under high load pressure, this paper uses this method to test the system performance. Test the response time of the system under high concurrency conditions by building different numbers of threads. The test data is shown in Fig. 5, and the average response time is in ms.
Two different types of stress tests were designed for experiments, which were clustered and non-clustered for all modules. Based on the experimental results, it can be seen that if all modules are non-clustered, the higher the number of concurrency, the faster the response time of the system increases, thus greatly reducing the system performance. However, if all modules are clustered, the higher the number of concurrency, the system response time will only gradually increase rather than surge. This shows that the cluster deployment scheme has a good tolerance to high concurrent access, can make up for the limited processing capacity of the server, and ensure that the system has high performance characteristics. It also shows that the system implemented in this paper can withstand high concurrent access pressure and achieve the expected goal in performance.

Conclusion
Lu Xun is different from other writers. The starting point of his literary career is to consider the Enlightenment of ideas, the change of national character and the development of society from the perspective of communication. He was not only born as a writer. Therefore, he not only paid attention to literary and artistic creation, but also paid more attention to the dissemination of culture, so as to guide his thoughts in a wide range and reach the extreme. He believes that no matter what the form and content are, they should be spread through the media. Only through the media can we spread new literature and new ideas, and play the role of awakening the national will. Lu Xun saw the importance of the media. He devoted himself to establishing various newspapers and magazines, and called on many knowledgeable Patriots to open up new cultural battlefields and actively spread new ideas and culture. Therefore, based on the deep learning technology, this paper establishes an online multimedia learning system of Lu Xun's literature, so that users can view and discuss Lu Xun's literary works without being limited by time and space, so that every user who loves Lu Xun's literature can exchange what they have learned, and let Lu Xun's literary works spread further and faster.
Funding The authors have not disclosed any funding.
Data availability Data will be made available on request.

Declarations
Conflict of interest The authors declare that they have no conflict of interests.
Ethical approval This article does not contain any studies with human participants performed by any of the authors.