Intelligent content recommendation method and system

Abstract

The present invention discloses an intelligent content recommendation system, which relates to the technical field of recommendation methods, including an information capture module, a sentiment analysis module, a data processing module, a recommendation module, a login module, a detection module, an exception handling module and a display module. The information capture module crawls user data and data on the type of content to be recommended; the sentiment analysis module is used to obtain the current sentiment type of the user; the data processing module is used for data processing and uploading content keywords to a total list of content keywords. The present invention adopts a sentiment analysis module to classify user sentiment types, increases the diversity of user samples on the basis of traditional user browsing content information and historical behavior data, expands the application scope of traditional collaborative filtering algorithms, recommends content to users according to changes in user sentiment, adopts a combination of a detection module, a recommendation module and a screening module, eliminates content that users are not interested in in detail, satisfies user search needs more meticulously, and improves the accuracy of recommendation preferences for users.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation methods, and in particular to an intelligent content recommendation method and system. Background Art

[0002] With the rapid development of the Internet and digital media, the amount of information users face has greatly increased. Traditional recommendation methods often fail to meet the personalized needs of users. Intelligent recommendation systems aim to improve user satisfaction and stickiness by analyzing user behavior data and interest preferences and providing content that better meets user needs.

[0003] At present, the Chinese invention patent for artificial intelligence-based content recommendation method and artificial intelligence content recommendation system with publication number CN113360762A provides an artificial intelligence-based content recommendation method and artificial intelligence content recommendation system, which extracts rich open topic relationships of the conversation service process through the conversation interest path matching conversation service process, the interest point mining data of the conversation service process and the weighted directed operation business graph of the conversation service process, so as to facilitate efficient key interest object mining based on the relationship of open topics in the key interest object mining process, and facilitate subsequent content recommendation and improve push accuracy by extracting key interest objects from the conversation service process. Memory-based collaborative filtering is easy to implement, but requires a lot of memory and CPU time, information acquisition is incomplete, and the model cannot accurately capture the evolution of user interests. Model-based collaborative filtering does not require so much memory and CPU time, and has better sparsity and scalability, but the related technology has the problems of insufficient application scope and insufficient diversity of user samples, and cannot make real-time recommendations based on changes in user emotional states, and cannot eliminate content that users are not interested in in detail. Summary of the invention

[0004] The technical problem solved by the present invention is that the existing technology cannot classify user emotion types and does not take into account the diversity of user samples. The application scope of traditional collaborative filtering algorithms is very small, and it is impossible to recommend content to users according to changes in user emotions and it is impossible to eliminate content that users are not interested in in detail.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent content recommendation system, including an information capture module, a sentiment analysis module, a data processing module, a recommendation module, a login module, a detection module, an exception handling module and a display module:

[0006] The information crawling module crawls user data through hook functions and collects data on the types of content to be recommended through big data;

[0007] The sentiment analysis module is used to analyze and process the user's multimodal information to obtain the user's current sentiment type;

[0008] The data processing module is used to process the user data, user emotion type and content type data to be recommended respectively, weight all the processed data and merge them to perform matrix decomposition, extract the m content keywords with the highest matrix decomposition scores and forward them to the recommendation module, and randomly select n content keywords from the m content keywords each time and upload them to the total content keyword list of the background of the intelligent content recommendation system;

[0009] The recommendation module is used to receive the m content keywords sent by the data processing module, and publish the m content keywords on the personalized recommendation interface of the display module;

[0010] The login module is used for users to register accounts and log in to the intelligent content recommendation system;

[0011] The detection module is used to detect the time when a user observes a certain content page, and adjust the recommended content to the user according to the time, including excluding all the content included in the content page from the workflow of the data processing module if the time when the user observes a certain content page is lower or higher than a preset browsing time interval;

[0012] The exception handling module is used to perform exception detection and update the content keyword list when the user enters the login module;

[0013] The display module is used to display the detailed interface of user information and content information.

[0014] Preferably, the information capture module includes:

[0015] The user data includes a user ID and user multimodal information, and the user multimodal information includes historical search records in the form of text and pictures;

[0016] The content type data to be recommended includes text content types and image content types collected from the entire network using big data, and a set of content keywords corresponding to each content type.

[0017] Preferably, the sentiment analysis module comprises:

[0018] Analyzing the current emotion type of the user using the emotion type detection model according to the first text feature vector obtained by the data processing module includes:

[0019] Sorting the first text feature vectors obtained by the data processing module according to time, and extracting time-frequency domain features from the sorted first text feature vectors to obtain time-frequency domain first text feature vectors, inputting the time-frequency domain first text feature vectors into a convolutional neural network to obtain high-level features of all texts, inputting the high-level features into the convolutional neural network again, performing self-learning on the high-level features to achieve fusion of all content types, extracting deep features from the processed valid data again through the convolutional neural network, and using the output deep features to classify the emotion types contained in the content through a support vector machine, wherein the emotion types include a first emotion type, a second emotion type, ... and an nth emotion type, where n is the number of emotion types classified by the support vector machine;

[0020] The emotion type detection model is used to perform real-time detection on the content searched or browsed by the user to obtain the user's current emotion type.

[0021] Preferably, the current emotion type of the user at a certain time point and the user behavior currently performed by the user are recorded in real time, and the user behavior is processed by the data processing module to obtain the user behavior text features, and the processed user behavior text features are added to the background database of the intelligent content recommendation system, and the emotion type detection model is input into the time by queue mode for pre-training, and the emotion type detection model is updated in real time;

[0022] The user behavior includes searched content, browsed content, and the emotion type corresponding to the user browsing the content.

[0023] Preferably, the data processing module includes:

[0024] Extract visual features of all image content types through a convolutional neural network model, reduce the dimension of the visual features, map the visual features to text space using a CLIP model, and generate text describing the image content type using an ImageCaptioning generation model;

[0025] The generated text describing the image content type and the text content type are converted into a first text feature vector using a word vector model.

[0026] Preferably, the data processing module further includes:

[0027] The results of obtaining different recommended keyword lists based on the user's current emotion type and user data include:

[0028] The method of converting the content type data to be recommended into a first text feature vector is used to convert the user multimodal information into a second text feature vector. The first text feature vector, the second text feature vector and the user's current emotion type are respectively reduced in dimension using an embedding method. The first text feature vector and the second text feature vector after the dimension reduction are respectively concatenated with the user's current emotion type after the dimension reduction to obtain a user initial vector and a content initial vector. An item similarity matrix is ​​established, the similarity between the content type data to be recommended is calculated, the similarity between the user initial vector and the content type data to be recommended is matrix decomposed, the scores after the matrix decomposition are ranked, and the m content keywords with the highest scores with the user are extracted and forwarded to a recommendation module. N content keywords are randomly selected from the m content keywords and uploaded to the total content keyword list of the background of the intelligent content recommendation system.

[0029] Preferably, m and n are both positive integers and m>n;

[0030] When the number of content keywords with the highest ranking scores is less than m and greater than n, the number of content keywords forwarded to the recommendation module should be equal to the number of the content keywords with the highest ranking;

[0031] When the number of content keywords with the highest ranking scores is not greater than n, the number of content keywords forwarded to the recommendation module and the number of randomly selected content keywords should be equal to the number of content keywords with the highest ranking.

[0032] Preferably, the exception handling module includes an exception detection unit and a screening unit:

[0033] The abnormality detection unit comprises:

[0034] When a user registers an account, the intelligent content recommendation system also runs the exception handling module, including:

[0035] The abnormality detection unit is used to detect and search the list of registered user name IDs in the background, and compare the user ID filled in by the current user. If there is a same, it will be displayed that the user ID is duplicated and cannot be registered;

[0036] The anomaly detection unit is used to compare the two passwords. If the two passwords are the same, the registration is successful. If they are different, the registration cannot be performed.

[0037] The screening unit comprises:

[0038] The user fills in the content keyword to be shielded, and obtains a total list of content keywords excluding the content keyword to be shielded.

[0039] Preferably, the display module includes a popular content section, a personalized recommendation section and a personal record section;

[0040] The popular content section is used to display the content keyword list published by the recommendation module;

[0041] The personalized recommendation section includes:

[0042] When a user logs into the intelligent content recommendation system, he / she fills in the content type label that he / she is interested in, and obtains the user's recent interest feedback;

[0043] According to the user's recent interest feedback and the latest news of the to-be-recommended content type data captured by the information capture module at intervals, updating several content keywords with the highest current search times in the popular content section;

[0044] The personal record section is used to display personal information, which includes user ID, the number of content keywords searched by the user, and the content keywords recommended to the user by the intelligent content recommendation system.

[0045] An intelligent content recommendation method, wherein the information capture module collects multimodal information of users and contents, the data processing module pre-processes the multimodal information to obtain a text feature vector, the text feature vector is input into the sentiment analysis module to obtain a classified sentiment type, the sentiment type, user data and data of the type of content to be recommended are input into the data processing module to obtain a content keyword list, the content keyword list is transmitted to the recommendation module, the content keyword list is published on the display module through the recommendation module, and the user logs in to the intelligent content recommendation system through the login module, and the content keyword list is updated and displayed through the detection module and the screening module.

[0046] The beneficial effects of the present invention are as follows: the present invention adopts a sentiment analysis module to classify user sentiment types, increases the diversity of user samples on the basis of traditional user browsing content information and historical behavior data, expands the application scope of traditional collaborative filtering algorithms, recommends content to users according to changes in user emotions, adopts a combination of a detection module, a recommendation module and a screening module, eliminates content that users are not interested in in detail, better understands user interests, meets user search needs more meticulously, and improves the accuracy of recommendation preferences for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A basic flow chart of an intelligent content recommendation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0049] Reference Figure 1 , is an embodiment of the present invention, and provides an intelligent content recommendation system, including an information capture module, a sentiment analysis module, a data processing module, a recommendation module, a login module, a detection module, an exception handling module and a display module:

[0050] The information crawling module crawls user data through hook functions and collects data on the types of content to be recommended through big data;

[0051] The sentiment analysis module is used to analyze and process the user's multimodal information to obtain the user's current sentiment type;

[0052] The data processing module is used to process the user data, user emotion type and content type data to be recommended respectively, weight all the processed data and merge them to perform matrix decomposition, extract the m content keywords with the highest matrix decomposition scores and forward them to the recommendation module, and randomly select n content keywords from the m content keywords and upload them to the content keyword list of the background of the intelligent content recommendation system each time, and the content keyword list is stored in the form of a stack;

[0053] The recommendation module is used to receive the m content keywords sent by the data processing module, and publish the m content keywords on the personalized recommendation interface of the display module; and recommend content keywords to the user according to the user's emotional changes.

[0054] The login module is used for users to register accounts and log in to the intelligent content recommendation system;

[0055] The detection module is used to detect the time when a user observes a certain content page, and adjust the recommended content to the user according to the time, including excluding all the content included in the content page from the workflow of the data processing module if the time when the user observes a certain content page is lower or higher than a preset browsing time interval;

[0056] The exception handling module is used to perform exception detection and update the content keyword list when the user enters the login module;

[0057] The display module is used to display the detailed interface of user information and content information.

[0058] The present invention adopts a sentiment analysis module to classify user sentiment types, increases the diversity of user samples based on traditional user browsing content information and historical behavior data, recommends content to users according to changes in user emotions, and adopts a combination of a detection module, a recommendation module and a screening module to eliminate content that users are not interested in in detail, thereby improving the accuracy of recommendation preferences for users.

[0059] The information capture module includes:

[0060] The user data includes a user ID and user multimodal information, and the user multimodal information includes historical search records in the form of text and pictures;

[0061] The content type data to be recommended includes text content types and image content types collected from the entire network using big data, and a set of content keywords corresponding to each content type.

[0062] The sentiment analysis module includes:

[0063] Analyzing the current emotion type of the user using the emotion type detection model according to the first text feature vector obtained by the data processing module includes:

[0064] Sorting the first text feature vectors obtained by the data processing module according to time, and extracting time-frequency domain features from the sorted first text feature vectors to obtain time-frequency domain first text feature vectors, inputting the time-frequency domain first text feature vectors into a convolutional neural network to obtain high-level features of all texts, inputting the high-level features into the convolutional neural network again, performing self-learning on the high-level features to achieve fusion of all content types, extracting deep features from the processed valid data again through the convolutional neural network, and using the output deep features to classify the emotion types contained in the content through a support vector machine, wherein the emotion types include a first emotion type, a second emotion type, ... and an nth emotion type, where n is the number of emotion types classified by the support vector machine;

[0065] The emotion type detection model is used to perform real-time detection on the content searched or browsed by the user to obtain the user's current emotion type.

[0066] Record the current emotion type of the user and the user behavior currently performed by the user at a certain time point in real time, and process the user behavior through the data processing module to obtain the user behavior text features, and add the processed user behavior text features to the background database of the intelligent content recommendation system, and input the emotion type detection model into the time by queue mode for pre-training, and update the emotion type detection model in real time;

[0067] The user behavior includes searched content, browsed content, and the emotion type corresponding to the user browsing the content.

[0068] The data processing module comprises:

[0069] The image content type is extracted through a convolutional neural network model to obtain visual features of all image content types. The visual features are reduced in dimension and then mapped to a text space using a CLIP model. The CLIP model can compare and match image and text features in the same semantic space. The Image Captioning model is used to generate text describing the image content type. The model generates a natural language description based on the image features.

[0070] The word vector model is used to convert the generated text describing the image content type and the text content type into a first text feature vector, thereby expanding the application scope of the collaborative filtering algorithm and increasing the diversity of user samples.

[0071] The data processing module also includes:

[0072] The results of obtaining different recommended keyword lists based on the user's current emotion type and user data include:

[0073] The method of converting the content type data to be recommended into a first text feature vector is used to convert the user multimodal information into a second text feature vector. The first text feature vector, the second text feature vector and the user's current emotion type are respectively reduced in dimension using an embedding method. The first text feature vector and the second text feature vector after the dimension reduction are respectively concatenated with the user's current emotion type after the dimension reduction to obtain a user initial vector and a content initial vector. An item similarity matrix is ​​established, the similarity between the content type data to be recommended is calculated, the similarity between the user initial vector and the content type data to be recommended is matrix decomposed, the scores after the matrix decomposition are ranked, and the m content keywords with the highest scores with the user are extracted and forwarded to a recommendation module. N content keywords are randomly selected from the m content keywords and uploaded to the total content keyword list of the background of the intelligent content recommendation system.

[0074] The process of matrix decomposition is:

[0075] The relationship between users and the data of the content type to be recommended can be transformed into a matrix decomposition process. First, the relationship between users and the data of the content type to be recommended is imagined as a whole to form a user-item matrix. In this matrix, the columns represent the data of the content type to be recommended and the rows represent the user ID. If the user has ever searched or browsed for the recommended content type data, then imagine a rating matrix. The intersection of the columns representing the data of the content type to be recommended and the extension lines representing the user ID in the matrix can be identified as the score of the user's evaluation of the recommended content type data. After the matrix decomposition operation, the two low-rank user content matrices will be decomposed from an overall user-item rating matrix. At the same time, the computational complexity will be smoothly reduced in the process. Starting from the idea of ​​linear regression, in order to predict the missing ratings, it is necessary to find the optimal implicit vector representation of the user and item, which leads to the previous step of observing the square of the data in a minimized manner.

[0076] The item similarity matrix is ​​established according to the item similarity formula, including:

[0077] The mathematical expression of the item similarity formula is:

[0078]

[0079] in, represents the item similarity matrix, N(μ) represents the number of users who like item μ, N(ν) represents the number of users who like item ν, and N(μ)∩N(ν) represents the number of users who like both items μ and ν;

[0080] Calculate the user co-occurrence matrix, the mathematical expression is:

[0081] C[ i ][ j ]=N(μ)∩N(ν)

[0082] Wherein, C[i][j] represents the number of users who like both item i and item j;

[0083] C[a][b]=1 means that there is one user who likes both item a and item b, that is user A. Since the number of people who like both item i and item j is the same as the number of people who like both item j and item i, this is a symmetric matrix. Similarly, all users have the same form of table. Because different users have different lists of favorite items, the size of the co-occurrence matrix formed by each user is different. In order to unify, the matrix size is expanded to h. h, where h is the size of the union of all users’ favorite item lists. By accumulating the co-occurrence matrices of different users’ favorite item sets, we get the matrix C, which is the numerator in the similarity formula.

[0084] Counting users The mathematical expression of the interest in item j is:

[0085]

[0086] in, Indicates user The degree of interest in item j, N(μ) represents the set of items that the user likes, S(j,K) represents the set of K items that are most similar to item j, represents the similarity between items j and i, represents the interest of user μ in item i;

[0087] For each item i in the list of items liked by user μ, find the K items most similar to it according to the item similarity matrix, and let the item similarity be , j=1,2,...,K, using item similarity To indicate the user's interest in item j, finally sort the selected items in reverse order according to the user's interest in them, and recommend the first K items in the list to the user, where K is the m.

[0088] Both m and n are positive integers and m>n;

[0089] When the number of content keywords with the highest ranking scores is less than m and greater than n, the number of content keywords forwarded to the recommendation module should be equal to the number of the content keywords with the highest ranking;

[0090] When the number of content keywords with the highest ranking scores is not greater than n, the number of content keywords forwarded to the recommendation module and the number of randomly selected content keywords should be equal to the number of content keywords with the highest ranking.

[0091] The exception handling module includes an exception detection unit and a screening unit:

[0092] The abnormality detection unit comprises:

[0093] When a user registers an account, the intelligent content recommendation system also runs the exception handling module, including:

[0094] The abnormality detection unit is used to detect and search the list of registered user name IDs in the background, and compare the user ID filled in by the current user. If there is a same, it will be displayed that the user ID is duplicated and cannot be registered;

[0095] The anomaly detection unit is used to compare the two passwords. If the two passwords are the same, the registration is successful. If they are different, the registration cannot be performed.

[0096] The screening unit comprises:

[0097] The user fills in the content keyword to be shielded, and obtains a total list of content keywords excluding the content keyword to be shielded.

[0098] The display module includes a popular content section, a personalized recommendation section and a personal record section;

[0099] The popular content section is used to display the content keyword list published by the recommendation module;

[0100] The personalized recommendation section includes:

[0101] When a user logs into the intelligent content recommendation system, he / she fills in the content type label that he / she is interested in, and obtains the user's recent interest feedback;

[0102] According to the user's recent interest feedback and the latest news of the to-be-recommended content type data captured by the information capture module at intervals, updating several content keywords with the highest current search times in the popular content section;

[0103] The personal record section is used to display personal information, which includes user ID, the number of content keywords searched by the user, and the content keywords recommended to the user by the intelligent content recommendation system.

[0104] An intelligent content recommendation method, wherein the information capture module collects multimodal information of users and contents, the data processing module pre-processes the multimodal information to obtain a text feature vector, the text feature vector is input into the sentiment analysis module to obtain a classified sentiment type, the sentiment type, user data and data of the type of content to be recommended are input into the data processing module to obtain a content keyword list, the content keyword list is transmitted to the recommendation module, the content keyword list is published on the display module through the recommendation module, and the user logs in to the intelligent content recommendation system through the login module, and the content keyword list is updated and displayed through the detection module and the screening module.

[0105] The present invention adopts a sentiment analysis module to classify user sentiment types, increases the diversity of user samples on the basis of traditional user browsing content information and historical behavior data, expands the application scope of traditional collaborative filtering algorithms, recommends content to users according to changes in user emotions, and adopts a combination of a detection module, a recommendation module and a screening module to eliminate content that users are not interested in in detail, better understand user interests, more meticulously meet user search needs, and improve the accuracy of recommendation preferences for users.

[0106] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent content recommendation system, characterized in that: It includes information capture module, sentiment analysis module, data processing module, recommendation module, login module, detection module, exception handling module and display module: The information crawling module crawls user data through hook functions and collects data on the types of content to be recommended through big data; The sentiment analysis module is used to analyze and process the user's multimodal information to obtain the user's current sentiment type; The data processing module is used to process the user data, user emotion type and content type data to be recommended respectively, weight all the processed data and merge them to perform matrix decomposition, extract the m content keywords with the highest matrix decomposition scores and forward them to the recommendation module, and randomly select n content keywords from the m content keywords each time and upload them to the total content keyword list of the background of the intelligent content recommendation system; The recommendation module is used to receive the m content keywords sent by the data processing module, and publish the m content keywords on the personalized recommendation interface of the display module; The login module is used for users to register accounts and log in to the intelligent content recommendation system; The detection module is used to detect the time when a user observes a certain content page, and adjust the recommended content to the user according to the time, including excluding all the content included in the content page from the workflow of the data processing module if the time when the user observes a certain content page is lower or higher than a preset browsing time interval; The exception handling module is used to perform exception detection and update the content keyword list when the user enters the login module; The display module is used to display the detailed interface of user information and content information; The information capture module includes: The user data includes a user ID and user multimodal information, and the user multimodal information includes historical search records in the form of text and pictures; The content type data to be recommended includes the text content types and image content types collected from the entire network using big data, and the content keyword set corresponding to each content type; The sentiment analysis module includes: Analyzing the current emotion type of the user using the emotion type detection model according to the first text feature vector obtained by the data processing module includes: Sorting the first text feature vectors obtained by the data processing module according to time, and extracting time-frequency domain features from the sorted first text feature vectors to obtain time-frequency domain first text feature vectors, inputting the time-frequency domain first text feature vectors into a convolutional neural network to obtain high-level features of all texts, inputting the high-level features into the convolutional neural network again, performing self-learning on the high-level features to achieve fusion of all content types, extracting deep features from the processed valid data again through the convolutional neural network, and using the output deep features to classify the emotion types contained in the content through a support vector machine, wherein the emotion types include a first emotion type, a second emotion type, ... and an nth emotion type, where n is the number of emotion types classified by the support vector machine; The emotion type detection model is used to perform real-time detection on the content searched or browsed by the user to obtain the user's current emotion type; Record the current emotion type of the user and the user behavior currently performed by the user at a certain time point in real time, and process the user behavior through the data processing module to obtain the user behavior text features, and add the processed user behavior text features to the background database of the intelligent content recommendation system, and input the emotion type detection model into the time by queue mode for pre-training, and update the emotion type detection model in real time; The user behavior includes the searched content, browsed content and the emotion type corresponding to the browsed content by the user; The data processing module comprises: Extract visual features of all image content types through a convolutional neural network model, reduce the dimension of the visual features, map the visual features to text space using a CLIP model, and generate text describing the image content type using an ImageCaptioning generation model; Using a word vector model, the generated text describing the image content type and the text content type are converted into a first text feature vector; The data processing module also includes: The results of obtaining different recommended keyword lists based on the user's current emotion type and user data include: The method of converting the content type data to be recommended into a first text feature vector is used to convert the user multimodal information into a second text feature vector, and the first text feature vector, the second text feature vector and the user's current emotion type are respectively reduced in dimension using an embedding method, and the first text feature vector and the second text feature vector after the dimension reduction are respectively spliced ​​with the user's current emotion type after the dimension reduction to obtain a user initial vector and a content initial vector, and an item similarity matrix is ​​established, and the similarity between the content type data to be recommended is calculated, and the similarity between the user initial vector and the content type data to be recommended is matrix decomposed, and the scores after the matrix decomposition are ranked, and the m content keywords with the highest scores with the user are extracted and forwarded to the recommendation module, and n content keywords are randomly selected from the m content keywords and uploaded to the total content keyword list of the background of the intelligent content recommendation system; m and n are both positive integers and m>n; When the number of content keywords with the highest score ranking is less than m and greater than n, the number of content keywords forwarded to the recommendation module should be equal to the number of the content keywords with the highest ranking; When the number of content keywords with the highest ranking scores is not greater than n, the number of content keywords forwarded to the recommendation module and the number of randomly selected content keywords should be equal to the number of content keywords with the highest ranking.

2. The intelligent content recommendation system according to claim 1, characterized in that: The exception handling module includes an exception detection unit and a screening unit: The abnormality detection unit comprises: When a user registers an account, the intelligent content recommendation system also runs the exception handling module, including: The abnormality detection unit is used to detect and search the list of registered user name IDs in the background, and compare the user ID filled in by the current user. If there is a same, it will be displayed that the user ID is duplicated and cannot be registered; The anomaly detection unit is used to compare the two passwords. If the two passwords are the same, the registration is successful. If they are different, the registration cannot be performed. The screening unit comprises: The user fills in the content keyword to be shielded, and obtains a total list of content keywords excluding the content keyword to be shielded.

3. The intelligent content recommendation system according to claim 1, characterized in that: The display module includes a popular content section, a personalized recommendation section and a personal record section; The popular content section is used to display the content keyword list published by the recommendation module; The personalized recommendation section includes: When a user logs into the intelligent content recommendation system, he / she fills in the content type label that he / she is interested in, and obtains the user's recent interest feedback; According to the user's recent interest feedback and the latest news of the to-be-recommended content type data captured by the information capture module at intervals, updating several content keywords with the highest current search times in the popular content section; The personal record section is used to display personal information, which includes user ID, the number of content keywords searched by the user, and the content keywords recommended to the user by the intelligent content recommendation system.

Images