Content recommendation method based on artificial intelligence

Abstract

The invention discloses a content recommendation method based on artificial intelligence, relates to the technical field of artificial intelligence recommendation, and solves the technical problems that the traditional recommendation method is single, the recommended content is lack of accuracy, the requirements of different users cannot be met, and the user experience is reduced. The interest direction of a user can be preliminarily known by obtaining a user tag, new users can quickly contact current concerned information by combining hot content recommendation based on the recommended related content, the interest and participation degree of the users in a platform are increased, historical data of old users are deeply analyzed, and the user experience is improved. Similar users are found by calculating the same type of pushing proportions, more potential interest contents can be mined, and the pushing contents are screened based on similar user preference value calculation and correlation analysis, so that recommendation is more fit for actual interest preferences of old users, repeated recommendation is avoided, and the quality of recommended contents and user satisfaction are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence recommendation technology, and in particular to an artificial intelligence-based content recommendation method. Background Art

[0002] With the rapid development and popular application of e-commerce, personalized recommendations in the Internet era have penetrated into all aspects of people's lives. For example, in e-commerce and news clients, the Internet can cater to users' preferences, recommend the content they are most interested in, and grasp users' interests in real time and accurately.

[0003] According to the publication number CN111949869A, a content information recommendation method and system based on artificial intelligence are disclosed, S1. Establish a classification channel model; S2. Collect user behavior data to analyze user interests and / or intentions; S3. According to the user interests and / or intentions, screen and match with the classification channels, and push the content information under the classification channel to the user; S4. Obtain user behavior records to divide users, and recommend the preferences of other users to users in the same group.

[0004] As the number of users and content continues to increase, how to accurately recommend content that suits users' interests becomes a key issue. Traditional recommendation methods may be too single or lack precision and cannot meet users' increasingly diverse and personalized needs. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides an artificial intelligence-based content recommendation method, which solves the problem that the traditional recommendation method is single, resulting in a lack of accuracy in the recommended content, inability to meet the needs of different users, and reduced user experience.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a content recommendation method based on artificial intelligence, the method specifically comprises the following steps: Step S001, obtaining a target user, and classifying and identifying the target user according to the user information to obtain a person identification result, wherein the person identification result includes: a recorded user and a new user; Step S002, analyzing the target user classified and identified as a recorded user, calculating similar users associated with the target user based on historical data, and analyzing the preferred contents of the similar users; Step S003, based on the similar users obtained by screening, calculating the preference of the push content corresponding to the similar users, and performing correlation analysis in combination with the historical data of the target user, while screening the push content and generating push information; Step S004, analyzing the target users classified and identified as new users, performing comprehensive push based on the user tags and popular contents of the new users, and generating push information.

[0007] As a further solution of the present invention: the specific method of obtaining the personnel recognition result in step S001 is: Try to obtain the historical data of the target user. If the historical data is successfully obtained, it means that the target user has a usage record and is classified as a recorded user. If the historical data cannot be obtained, it means that the target user has not had any usage behavior and is classified as a new user.

[0008] As a further solution of the present invention: the specific method of analyzing the target user classified and identified as the recorded user in step S002 is: Obtain the target user identified as the recorded user, then extract the historical data of the target user, and then obtain the push content associated with the historical data. For each classification type, count the corresponding push quantity. After completing the quantity statistics, calculate the proportion of each type of push content in the total push through a specific calculation method, and record it as the target push proportion; Get the historical records, label each historical user in the historical records as i, i from 1 to j, j is the total number of historical users, calculate the proportion of the same type of push for each historical user i, and then match the target push proportion of the target user with the corresponding proportion of the historical users to filter out similar users.

[0009] As a further solution of the present invention: the specific method of matching and screening similar users in step S002 is: Obtain the target push ratio of the target user, recorded as Za, where a=1,2,…,b, b is the number of groups of the same type of content, select any historical user for analysis, record the same type of push ratio of the historical user as Zn, where n=1,2,…,m, m is the number of groups of the same type of pushed content of the historical user, then calculate the difference between Za and Zn under the same type, and then compare the difference with the preset value, which is set by the operator; If the difference is greater than the preset value, it indicates that the corresponding push contents of the same type are similar, and a content similarity signal is generated; if the difference is less than the preset value, it indicates that they are not similar, and a content dissimilarity signal is generated; The number of similar signals and dissimilar signals is obtained, and the number of similar signals and dissimilar signals is compared. If the number of similar signals is greater than the number of dissimilar signals, the corresponding historical user is marked as a similar user. If the number of similar signals is less than the number of dissimilar signals, no processing is performed.

[0010] As a further solution of the present invention: the specific method of generating the push information in step S003 is: Get all groups marked as similar users, label similar users as o, where o = 1, 2, ..., p. Then, get the push content corresponding to similar user o, and calculate the preference value of these push contents. The specific calculation method is: select any similar user as the analysis object, get the score value of the analysis object for different push contents and the number of views of the same type of push contents, substitute the score value and the number of views into the formula Calculate the preference value Q of the same type of pushed content, where R is the score of the pushed content, c is the number of views of the same type of pushed content, and α is the preset proportional coefficient; Similarly, the preference values ​​of all types of pushed content are calculated and sorted from large to small according to the preference values. At the same time, the preference values ​​of all similar users corresponding to different pushed content are calculated. Then, the preference values ​​of the target user corresponding to different types of pushed content are calculated. At the same time, the pushed content of similar users is screened based on the pushed content of the target user, and the content to be analyzed is obtained. Then, the correlation value between the content to be analyzed and the same type of pushed content of the target user is calculated.

[0011] As a further solution of the present invention: the specific method of calculating the correlation value between the content to be analyzed and the same type of pushed content of the target user in step S003 is: Obtain the content to be analyzed, obtain the same type of pushed content of the target user, count the number of similar content to the content to be analyzed in the same type of pushed content, calculate the preference values ​​of similar content and find the average as the association value, calculate the average preference value of all content to be analyzed as the threshold, if the association value is greater than the threshold, mark the content to be analyzed as related content; if it is less than the threshold, do not process it; All the related contents obtained through analysis are obtained, and the preference values ​​corresponding to the related contents are obtained at the same time, and they are sorted from large to small according to the preference values, and push information is further generated according to the sorting order.

[0012] As a further solution of the present invention: the specific method of analyzing the target user classified and identified as a new user and generating push information in step S004 is: Obtain a target user identified as a new user, then obtain the user tag selected by the target user when registering, and at the same time, obtain popular content with high attention and appeal, and classify the popular content into the same type in a manner that matches the tag content to obtain popular content of the same type; The target users are then secondary classified to obtain secondary classified users. Based on the secondary classification results, the corresponding users of the same type are obtained. Next, the push content corresponding to the users of the same type is obtained, and the push content with the largest preference value is screened out and marked as pre-selected content. At the same time, the pre-selected content is matched with the same type of tagged content and the same type of popular content.

[0013] As a further solution of the present invention: the specific method of matching the pre-selected content in step S004 is as follows: If there is corresponding pre-selected content in the content with the same tag, that is, the two match, the push information is generated based on the pre-selected content. If the pre-selected content does not match the content with the same tag, the pre-selected content is matched with the popular content of the same type. If the pre-selected content matches the popular content of the same type, push information is generated based on the pre-selected content; if the pre-selected content does not match the popular content of the same type, push information is generated based on the popular content of the same type.

[0014] The present invention provides a content recommendation method based on artificial intelligence. Compared with the prior art, it has the following beneficial effects: The present invention classifies users. For new users, by obtaining user tags during registration, the invention can preliminarily understand the user's interest direction and recommend relevant content based on this. At the same time, combined with popular content recommendations, new users can quickly access currently popular information, increase user interest and participation in the platform, conduct in-depth analysis of historical data of old users, find similar users by calculating the proportion of the same type of push notifications, and discover more potential content of interest. Based on similar user preference value calculation and association analysis, push content can be filtered, so that recommendations are more in line with the actual interest preferences of old users, duplicate recommendations can be avoided, and the quality of recommended content and user satisfaction can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the steps of the present invention; Figure 2 This is a block diagram of the system principle of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] For example, see Figure 1 and Figure 2 , the present application provides a content recommendation method based on artificial intelligence, the method specifically comprises the following steps: In step S001, a target user is acquired, and the target user is classified and identified according to user information to obtain a person identification result, wherein the person identification result includes: recorded users and new users.

[0018] Specifically, the method for classifying and identifying the target user is: first obtain the corresponding target user, and the target user is customized by the system, and the selection method is random selection, and then obtain the historical data of the target user. If the historical data of the target user can be obtained, it means that the target user has used it, and it is classified as a recorded user. If the historical data of the target user cannot be obtained, it means that the target user has not used it, and it is classified as a new user.

[0019] For example, in an online music platform, the system randomly selects a user A. The platform tries to find historical data such as user A's play history and favorites list. If it is found that user A has past play records and favorites, then user A is classified as a record user. For another randomly selected user B, if the platform cannot find any historical operation data about user B, user B is identified as a new user.

[0020] Next, in step S002, the target user classified and identified as a recorded user is analyzed, and similar users associated with the target user are calculated based on historical data, and the preferred contents of the similar users are analyzed.

[0021] Obtain a target user who is identified and classified as a record user, then obtain historical data of the target user, and obtain push content corresponding to the historical data, and classify the push content into the same type. For example, the push content is classified into the following types: entertainment type, news type, and knowledge type, and the number of pushes corresponding to the same type of content is obtained. Specifically, the number of pushes corresponding to the entertainment type, the number of pushes corresponding to the news type, and the number of pushes corresponding to the knowledge type are obtained respectively, and then the proportion of pushes of the same type is calculated and recorded as the target push proportion; Obtain historical records, and the historical records here have different meanings from historical data. Historical records include users and the push content corresponding to the users. At the same time, different historical users in the historical records are labeled as i, and i=1, 2, ..., j, where j represents the number of historical users, and the proportion of the same type of push corresponding to historical user i is calculated. Then, the target push proportion of the target user is matched with the same type of push proportion of the historical user to analyze and filter to obtain similar users, and the specific matching and filtering method is as follows: Get the target push ratio of the target user, and record the target push ratio as Za, and a=1, 2, ..., b, where b represents the same type of content. For example, if there are three groups of the same type of content, then the value of b is 3. It is found that there are 100 push contents, of which 30 are entertainment type pushes, 40 are news type pushes, and 30 are knowledge type pushes. Then the entertainment type push ratio is 30%, the news type push ratio is 40%, and the knowledge type push ratio is 30%. Get any historical user for analysis, and then record the same type of push ratio of the historical user obtained as Zn, and n=1, 2, ..., m, where m represents the same type of push content corresponding to the historical user obtained. For example, if there are 7 groups of the same type of content, then the value of m is 7. Assume that among the push contents of historical user B, there are 25 entertainment-type, 45 news-type, and 30 knowledge-type. The calculation shows that the proportion of entertainment-type push is 25%, the proportion of news-type push is 45%, and the proportion of knowledge-type push is 30%. At the same time, the difference between the target push proportion Za and the same-type push proportion Zn is calculated. For example, the push proportion of the target user belonging to the entertainment type is subtracted from the push proportion of the historical user, and the difference is compared with the preset value. The specific value of the preset value is set by the operator. If the difference is greater than the preset value, it means that the corresponding push contents of the same type are similar, and a content similarity signal is generated. If the difference is less than the preset value, it means that the corresponding push contents of the same type are not similar, and a content dissimilar signal is generated. The preset value is set to 5%. When calculating the difference in the proportion of the same-type pushes of target user A and historical user B, the difference in entertainment type is 5%, the difference in news type is 5%, and the difference in knowledge type is 0%. Since the difference between the entertainment type and the news type is equal to the preset value, one content similarity signal and one content dissimilarity signal are generated according to the rules; Similarly, the target push ratio of the target user is calculated separately, and the number of similar signals and dissimilar signals is obtained. At the same time, the number of similar signals and dissimilar signals is compared. If the number of similar signals is greater than the number of dissimilar signals, the corresponding historical user is marked as a similar user. If the number of similar signals is less than the number of dissimilar signals, no processing is done.

[0022] Next, in step S003, based on the similar users obtained by screening, the push content preference levels of the similar users are calculated, and a correlation analysis is performed in combination with the historical data of the target user, and the push content is screened at the same time.

[0023] Get all similar users and label them as o, where o=1, 2, ..., p. Then get the push content corresponding to similar user o and calculate the preference value of the push content. The specific calculation method is as follows: Take any similar user as the analysis object, and obtain the ratings of the analysis object for different pushed contents and the number of views of the same type of pushed contents. The ratings obtained here are the ratings of the same type of pushed contents, and the ratings are specifically expressed as the user's ratings for the pushed contents, with a full score of 5. Substitute the ratings and views into the formula Calculate the preference value Q of the same type of pushed content, where R is the score of the pushed content, c is the number of views of the same type of pushed content, and α is a preset proportional coefficient. The specific value is set by the operator. Similarly, the preference values ​​of all types of pushed content are calculated and sorted from large to small according to the preference values. At the same time, the preference values ​​of all similar users corresponding to different pushed content are calculated; Then, the preference values ​​of the target user corresponding to different types of pushed content are calculated. At the same time, the pushed content of similar users is screened based on the pushed content of the target user, and the content to be analyzed is obtained. Here, the content to be analyzed is the pushed content that does not exist in the target user. Then, the correlation value between the content to be analyzed and the same type of pushed content of the target user is calculated. The specific calculation method is as follows: Obtain the content to be analyzed, then obtain the same type of pushed content corresponding to the target user, and obtain the number of similar contents between the content to be analyzed and the same type of pushed content, and the similar content here is represented as the content to be analyzed existing in the same type of pushed content, and then calculate the number of similar contents and the preference values ​​corresponding to the similar contents, and the calculation method of the preference values ​​of the similar contents here is the same as the calculation method of the preference values ​​of the same type of pushed content, and calculate the average of the preference values ​​of all similar contents, and use the calculated average value as the associated value, and compare the associated value of the content to be analyzed with the threshold value, and the specific value of the preset value is set by the operator, and the specific threshold value calculation method is to calculate the average of the preference values ​​of all the content to be analyzed, and use the calculated average value as the threshold value; When the mean value is greater than the threshold value, the corresponding content to be analyzed is marked as related content. Conversely, when the mean value is less than the threshold value, it is not processed.

[0024] Finally, in step S004, push information is generated based on the acquired related content and the preference value corresponding to the related content.

[0025] All the related contents obtained through analysis are obtained, and the preference values ​​corresponding to the related contents are obtained at the same time, and they are sorted from large to small according to the preference values, and push information is further generated according to the sorting order.

[0026] Embodiment 2, as Embodiment 2 of the present invention, is implemented on the basis of step S001 of Embodiment 1, and differs from Embodiment 1 in that: In step S002, the target user classified and identified as a new user is analyzed, and a comprehensive push is performed based on the user tag and popular content of the new user, and push information is generated.

[0027] Get the target user who is identified and classified as a new user, and then get the user tag of the target user. The user tag here means that when a new user registers, some common interest tag options are provided for the user to choose from. For example, movies can be divided into science fiction, action, romance and other tags; music can be divided into pop, rock, classical and other tags. Recommend related content based on the tag selected by the user, and obtain the corresponding tag content based on the user tag, and classify the tag content into the same type to obtain the same type of tag content. The same type classification here is based on the user tag as the classification basis for classification processing, and then obtain the corresponding popular content. Popular content means that the content usually has a high degree of attention and appeal. For example, for movies, popular movies can be determined based on indicators such as box office, ratings, and playback volume; for news, popular news can be determined based on indicators such as click-through rate, number of comments, and number of shares. At the same time, popular content is classified into the same type to obtain the same type of popular content; Then, the target user is secondary classified to obtain secondary classified users, and the secondary classified users include young users and professional users. Specifically, professional users are those who have professional knowledge in a certain field, and corresponding users of the same type are obtained based on the secondary classified users. For example, if the secondary classified user is a professional user, the obtained users of the same type are professional users. Then, the push content corresponding to the users of the same type is obtained, and the push content corresponding to the maximum preference value is screened and marked as pre-selected content. At the same time, the pre-selected content is matched with the target user's same type of tagged content. If the pre-selected content matches the same type of tagged content, and the match here indicates that the corresponding pre-selected content exists in the same type of tagged content, the push information is generated based on the pre-selected content. If the pre-selected content does not match the same type of tagged content, the pre-selected content is matched with the same type of popular content. If the pre-selected content matches the popular content of the same type, push information is generated based on the pre-selected content. If the pre-selected content does not match the popular content of the same type, push information is generated based on the popular content of the same type.

[0028] Embodiment 3, as the embodiment 3 of the present invention, focuses on combining the implementation processes of embodiment 1 and embodiment 2 for implementation.

[0029] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0030] The above embodiments are only used to illustrate the technical method 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 method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A content recommendation method based on artificial intelligence, characterized in that: The method specifically comprises the following steps: Step S001, obtaining a target user, and classifying and identifying the target user according to the user information to obtain a person identification result, wherein the person identification result includes: a recorded user and a new user; Step S002, analyzing the target user classified and identified as a recorded user, calculating similar users associated with the target user based on historical data, and analyzing the preferred contents of the similar users; Step S003, based on the similar users obtained by screening, calculating the preference of the push content corresponding to the similar users, and performing correlation analysis in combination with the historical data of the target user, while screening the push content and generating push information; Step S004, analyzing the target users classified and identified as new users, performing comprehensive push based on the user tags and popular contents of the new users, and generating push information.

2. The content recommendation method based on artificial intelligence according to claim 1, characterized in that: The specific method of obtaining the personnel identification result in step S001 is: Try to obtain the historical data of the target user. If the historical data is successfully obtained, it means that the target user has a usage record and is classified as a recorded user. If the historical data cannot be obtained, it means that the target user has not had any usage behavior and is classified as a new user.

3. The content recommendation method based on artificial intelligence according to claim 1, characterized in that: The specific method of analyzing the target user classified and identified as a recorded user in step S002 is: Obtain the target user identified as the recorded user, then extract the historical data of the target user, and then obtain the push content associated with the historical data. For each classification type, count the corresponding push quantity. After completing the quantity statistics, calculate the proportion of each type of push content in the total push, and record it as the target push proportion; Get the historical records, label each historical user in the historical records as i, i from 1 to j, j is the total number of historical users, calculate the proportion of the same type of push for each historical user i, and then match the target push proportion of the target user with the corresponding proportion of the historical users to filter out similar users.

4. The content recommendation method based on artificial intelligence according to claim 3, characterized in that: The specific method of matching and screening similar users in step S002 is: Obtain the target push ratio of the target user, recorded as Za, where a=1,2,…,b, b is the number of groups of the same type of content, select any historical user for analysis, record the same type of push ratio of the historical user as Zn, where n=1,2,…,m, m is the number of groups of the same type of pushed content of the historical user, then calculate the difference between Za and Zn under the same type, and then compare the difference with the preset value, which is set by the operator; If the difference is greater than the preset value, it indicates that the corresponding push contents of the same type are similar, and a content similarity signal is generated; If the difference is less than a preset value, it indicates dissimilarity and generates a content dissimilarity signal; The number of similar signals and dissimilar signals is obtained, and the number of similar signals and dissimilar signals is compared. If the number of similar signals is greater than the number of dissimilar signals, the corresponding historical user is marked as a similar user. If the number of similar signals is less than the number of dissimilar signals, no processing is performed.

5. The content recommendation method based on artificial intelligence according to claim 1, characterized in that: The specific method of generating the push information in step S003 is: Get all groups marked as similar users, label similar users as o, where o = 1, 2, ..., p. Then, get the push content corresponding to similar user o, and calculate the preference value of these push contents. The specific calculation method is: select any similar user as the analysis object, get the score value of the analysis object for different push contents and the number of views of the same type of push contents, substitute the score value and the number of views into the formula Calculate the preference value Q of the same type of pushed content, where R is the score of the pushed content, c is the number of views of the same type of pushed content, and α is the preset proportional coefficient; Similarly, the preference values ​​of all types of pushed content are calculated and sorted from large to small according to the preference values. At the same time, the preference values ​​of all similar users corresponding to different pushed content are calculated. Then, the preference values ​​of the target user corresponding to different types of pushed content are calculated. At the same time, the pushed content of similar users is screened based on the pushed content of the target user, and the content to be analyzed is obtained. Then, the correlation value between the content to be analyzed and the same type of pushed content of the target user is calculated.

6. The content recommendation method based on artificial intelligence according to claim 5, characterized in that: The specific method of calculating the correlation value between the content to be analyzed and the same type of pushed content of the target user in step S003 is: Obtain the content to be analyzed, obtain the same type of pushed content of the target user, count the number of similar content to the content to be analyzed in the same type of pushed content, calculate the preference values ​​of similar content and find the average as the association value, calculate the average preference value of all content to be analyzed as the threshold, if the association value is greater than the threshold, mark the content to be analyzed as related content; if it is less than the threshold, do not process it; All the related contents obtained through analysis are obtained, and the preference values ​​corresponding to the related contents are obtained at the same time, and they are sorted from large to small according to the preference values, and push information is further generated according to the sorting order.

7. The content recommendation method based on artificial intelligence according to claim 1, characterized in that: The specific method of analyzing the target user classified and identified as a new user and generating push information in step S004 is: Obtain a target user identified as a new user, then obtain the user tag selected by the target user when registering, and at the same time, obtain popular content, and classify the popular content into the same type in a manner that matches the tag content to obtain popular content of the same type; The target users are then secondary classified to obtain secondary classified users. Based on the secondary classification results, the corresponding users of the same type are obtained. Next, the push content corresponding to the users of the same type is obtained, and the push content with the largest preference value is screened out and marked as pre-selected content. At the same time, the pre-selected content is matched with the same type of tagged content and the same type of popular content.

8. The content recommendation method based on artificial intelligence according to claim 7, characterized in that: The specific method of matching the pre-selected content in step S004 is as follows: If there is corresponding pre-selected content in the content with the same tag, that is, the two match, the push information is generated based on the pre-selected content. If the pre-selected content does not match the content with the same tag, the pre-selected content is matched with the popular content of the same type. If the pre-selected content matches the popular content of the same type, push information is generated based on the pre-selected content; if the pre-selected content does not match the popular content of the same type, push information is generated based on the popular content of the same type.

Images