Content recommendation method and system based on flow monitoring

Abstract

The invention provides a content recommendation method and system based on flow monitoring, and relates to the technical field of content recommendation, and the method comprises the steps: determining related content information related to a search keyword; determining a first recommendation list according to the correlation between the related content information and the search keyword, the historical page view and the historical interaction quantity; determining a first preset number of pieces of to-be-tested content information, obtaining historical interaction information of the to-be-tested content information, and further determining key negative factors; and determining a second recommendation list according to the historical browsing record of the user, the introduction information of the to-be-tested content information and the key negative factors, and displaying the related content information. According to the method and the device, the key negative factors can be searched, and the conflict degree of the user to the key negative factors can be judged, so that the preference of the user to various factors is reflected, the second recommendation list is obtained, the content information in the second recommendation list is related to the search keyword, and the recommendation sequence considers the quality of the content information and the preference of the user; and user experience is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of content recommendation, and in particular to a content recommendation method and system based on traffic monitoring. Background Art

[0002] In related technologies, when a user enters a keyword to search for content, a recommendation list is usually generated based on the relevance between the content and the keyword, and the content is displayed to the user according to the recommendation list. Although this recommendation method can search for content related to the keyword, it does not take into account the user's preferences for the content, nor the quality of the content itself. Therefore, it may recommend content to the user that is highly relevant but of poor quality or not of interest to the user, resulting in a poor user experience.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0004] The present invention provides a content recommendation method and system based on traffic monitoring, which can solve the technical problem that related technologies are difficult to pay attention to user preferences and content quality when recommending content based on keywords.

[0005] According to a first aspect of the present invention, a content recommendation method based on traffic monitoring is provided, comprising:

[0006] Upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword;

[0007] Determining a first recommendation list of the plurality of related content information based on the relevance of the plurality of related content information to the search keyword, the historical page views and the historical interaction counts of the plurality of related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information;

[0008] In the first recommendation list, a first preset number of content information to be tested is determined, and historical interaction information of the content information to be tested is obtained, wherein the historical interaction information includes content of historical evaluation information of the content information to be tested;

[0009] Determining key negative factors based on the historical interaction information;

[0010] Determining a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors;

[0011] According to the second recommendation list, multiple related content information is displayed.

[0012] According to the present invention, upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword includes:

[0013] Acquire the title, introduction information and attribute tag of each content information, wherein the attribute tag includes tag information added by the author of the content information to the content information for describing the attributes of the content information;

[0014] Using a semantic recognition model, obtain the title semantic information of the title, obtain the introduction semantic information of the introduction information, and obtain the label semantic information of the attribute label;

[0015] Obtain keyword semantic information of search keywords through semantic recognition model;

[0016] For each piece of content information, determining a first similarity between the keyword semantic information and the title semantic information, a second similarity between the keyword semantic information and the introduction semantic information, and a third similarity between the keyword semantic information and the tag semantic information;

[0017] Obtaining a maximum similarity among the first similarity, the second similarity, and the third similarity;

[0018] If the maximum similarity value corresponding to the content information is greater than or equal to a preset first similarity threshold, the content information is determined to be relevant content information.

[0019] According to the present invention, a first recommendation list of multiple related content information is determined based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, including:

[0020] Determining the maximum similarity value corresponding to the relevant content information as the relevance between the relevant content information and the search keyword;

[0021] Determine the interaction attraction coefficient of the relevant content information based on the ratio of the historical interaction number and the historical pageview number of the relevant content information;

[0022] Determining a first recommendation coefficient for relevant content information based on the interaction attraction coefficient, the number of historical interactions, the historical page views, and the relevance;

[0023] A first recommendation list of multiple related content information is determined according to the first recommendation coefficient.

[0024] According to the present invention, determining key negative factors based on the historical interaction information includes:

[0025] Determine the sentiment classification information of each historical evaluation information of the content information to be tested through the sentiment classification model;

[0026] Filtering negative evaluation information from multiple historical evaluation information based on the sentiment classification information;

[0027] Determine the negative semantic information of negative evaluation information through semantic recognition model;

[0028] performing clustering processing on the negative semantic information to obtain a plurality of negative semantic information clusters;

[0029] The cluster centers of multiple negative semantic information clusters are identified as key negative factors.

[0030] According to the present invention, a second recommendation list of multiple related content information is determined based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors, including:

[0031] Determine, based on the user's historical browsing history, the progress of playback of historical content information browsed by the user, the user's interaction history with the historical content information, and the interaction history of the historical content information;

[0032] determining the user's resistance coefficient to the key negative factor based on the playback progress, the user interaction record, the interaction information record, and the key negative factor;

[0033] Determining a publicity negation coefficient of the content information to be tested based on the introduction information of the content information to be tested and the key negative factors;

[0034] Obtain historical interaction information of multiple related content information in the first recommendation list;

[0035] Determining a supplementary recommendation coefficient for each piece of relevant content information based on historical interaction information of the plurality of relevant content information and the key negative factors;

[0036] A second recommendation list of multiple related content information is determined based on the resistance coefficient, the publicity negation coefficient and the supplementary recommendation coefficient.

[0037] According to the present invention, determining the user's resistance coefficient to the key negative factor based on the playback progress, the user interaction record, the interaction information record, and the key negative factor includes:

[0038] According to the interactive information record of the historical content information, select the target historical content information having the i-th key negative factor from the multiple historical content information;

[0039] According to the formula , Determine the user's resistance coefficient to the i-th key negative factor ,in, is the playback progress of the jth target historical content information with the i-th key negative factor, is the average playback progress of historical content information, is the semantic similarity between the user's historical evaluation information of the jth target historical content information with the i-th key negative factor and the i-th key negative factor. If the user has not evaluated the jth target historical content information with the i-th key negative factor, then , is the number of target historical content information with the i-th key negative factor, if (*) is the conditional function, j≤ , and i, j and All are positive integers.

[0040] According to the present invention, determining the publicity negation coefficient of the content information to be tested based on the introduction information of the content information to be tested and the key negative factors includes:

[0041] Based on the historical interaction information of the content to be tested, the target content to be selected is screened for the presence of the i-th key negative factor;

[0042] Segmenting the introduction information of the target content information and removing stop words to obtain multiple introduction words of the introduction information;

[0043] Determine the introductory sentiment classification information of each introductory word through the sentiment classification model;

[0044] According to the introduction sentiment classification information, determine the positive introduction words of the introduction word type;

[0045] Add negative affixes to positive introductory words to obtain negative introductory words;

[0046] Obtain negative introduction semantic information of negative introduction words through semantic recognition model;

[0047] According to the formula , Determine the negative coefficient of the kth target content information for the ith key negative factor ,in, is the negative introduction semantic information of the sth negative introduction word of the kth target content information to be tested, is the i-th key negative factor, for and where N is the number of key negative factors, max(*) is the maximum value function, i≤N, and s, M, i and N are all positive integers.

[0048] According to the present invention, based on the historical interaction information of a plurality of relevant content information and the key negative factors, a supplementary recommendation coefficient of each relevant content information is determined, including:

[0049] Determine the interactive sentiment classification information of historical interactive information through the sentiment classification model;

[0050] According to the interaction sentiment classification information, positive historical interaction information is filtered out from the historical interaction information;

[0051] Add negative affixes to positive historical interaction information to obtain modified historical interaction information;

[0052] Obtaining modification interaction semantic information of modification history interaction information through a semantic recognition model;

[0053] According to the formula , Determine the supplementary recommendation coefficient of the tth relevant content information for the ith key negative factor ,in, is the modified interaction semantic information of the xth historical interaction information of the tth related content information, is the i-th key negative factor, for and The semantic similarity of is the number of historical interactive information of the t-th related content information, x≤ , and x and All are positive integers.

[0054] According to the present invention, a second recommendation list of a plurality of related content information is determined based on the conflict coefficient, the publicity negation coefficient and the supplementary recommendation coefficient, including:

[0055] If the tth related content information is not the target content information to be tested, then according to the formula , Determine the gain recommendation coefficient of the tth relevant content information ,in, is the number of positive historical interaction information of the t-th related content information, is the number of historical interactive information of the t-th related content information, is the average positive interaction rate of multiple related content information, is the supplementary recommendation coefficient of the t-th relevant content information for the i-th key negative factor, is the user's resistance coefficient to the i-th key negative factor, N is the number of key negative factors, i≤N, and both i and N are positive integers;

[0056] If t relevant content information is the target content information to be tested, then according to the formula , Determine the gain recommendation coefficient of the tth relevant content information ,in, is the publicity negation coefficient of the t-th relevant content information for the i-th key negative factor;

[0057] A second recommendation list is determined according to the gain recommendation coefficients of the respective related content information.

[0058] According to a second aspect of the present invention, there is provided a content recommendation system based on traffic monitoring, comprising:

[0059] A related content information module is configured to, upon receiving a user's search keyword, determine a plurality of related content information related to the search keyword;

[0060] A first recommendation list module is configured to determine a first recommendation list of multiple related content information based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information;

[0061] A historical interaction information module is configured to determine a first preset number of pieces of content information to be tested in the first recommendation list, and obtain historical interaction information of the content information to be tested, wherein the historical interaction information includes historical evaluation information of the content information to be tested;

[0062] a key negative factor module, configured to determine key negative factors based on the historical interaction information;

[0063] A second recommendation list module is used to determine a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors;

[0064] A display module is used to display multiple related content information according to the second recommendation list.

[0065] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0066] According to the present invention, after obtaining relevant content information related to the search keyword, the quality of the relevant content information can be indirectly reflected based on the historical number of views and historical interactions of the relevant content information, thereby combining the quality of the relevant content information to make recommendations to users, and the key negative factors of the content information to be tested can be found, and the degree of user resistance to the key negative factors can be determined, thereby reflecting the user's likes and dislikes of various factors of the content, so as to obtain a second recommendation list, so that the content information in the second recommendation list is not only related to the search keyword, but the recommendation order takes into account the quality of the content information itself and the user's preferences, thereby improving the user experience. When determining the resistance coefficient, the impact of the key negative factors on the playback can be determined through the playback progress of the historical content information, thereby determining whether the user has a resistance to the key negative factors. The intensity of the user's resistance to the key negative factors can also be amplified through the user's interaction records, thereby accurately reflecting the user's resistance to the key negative factors and providing an accurate data basis for recommending content that the user is interested in. When determining the negative advertising coefficient, negative introductory words can be obtained by adding negative affixes to positive introductory words in the introductory information. The semantic similarity between the negative introductory words and key negative factors can be calculated to determine whether the content under test contains false advertising. This can reduce the recommendation probability of content containing false advertising and improve the user experience. When determining the supplementary recommendation coefficient, negative affixes can be added to positive historical interaction information in historical user objective comments. The semantic similarity between the modified historical interaction information and key negative factors can be determined to determine the likelihood that the relevant content does not contain key negative factors. The supplementary recommendation coefficient can be used to increase the recommendation probability of relevant content without key negative factors, thereby improving the user experience. When determining the gain recommendation coefficient, the target content under test and other related content can be classified and calculated to further reduce the recommendation coefficient of the target content under test that contains false advertising. The recommendation coefficient of each relevant content can also be adjusted based on multiple factors, such as the quality of the relevant content, the presence of false advertising, the presence of key negative factors, and the user's resistance to key negative factors. This can prioritize the recommendation of relevant content with high quality, no false advertising, and no negative key factors, thereby improving the user experience.

[0067] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts.

[0069] Figure 1 A flow chart showing an exemplary method for content recommendation based on traffic monitoring according to an embodiment of the present invention is provided;

[0070] Figure 2 A flowchart of searching for relevant content information according to an embodiment of the present invention is exemplarily shown;

[0071] Figure 3 The following is an exemplary flowchart of determining a first recommendation list according to an embodiment of the present invention;

[0072] Figure 4 A flowchart of determining key negative factors according to an embodiment of the present invention is exemplarily shown;

[0073] Figure 5 The following is an exemplary flowchart of determining a second recommendation list according to an embodiment of the present invention;

[0074] Figure 6 A block diagram of a content recommendation system based on traffic monitoring according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0075] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings 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 making creative efforts shall fall within the scope of protection of the present invention.

[0076] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0077] Figure 1 A flow chart of a content recommendation method based on traffic monitoring according to an embodiment of the present invention is exemplarily shown. The method includes:

[0078] Step S1, upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword;

[0079] Step S2, determining a first recommendation list of multiple related content information based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information;

[0080] Step S3, determining a first preset number of content information to be tested in the first recommendation list, and obtaining historical interaction information of the content information to be tested, wherein the historical interaction information includes historical evaluation information of the content information to be tested;

[0081] Step S4, determining key negative factors based on the historical interaction information;

[0082] Step S5, determining a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors;

[0083] Step S6: Display multiple related content information according to the second recommendation list.

[0084] According to the content recommendation method based on traffic monitoring of an embodiment of the present invention, after obtaining relevant content information related to the search keyword, it can indirectly reflect the quality of the relevant content information based on the historical browsing volume and historical interaction volume of the relevant content information, thereby recommending to users in combination with the quality of the relevant content information, and can find key negative factors of the content information to be tested, and can judge the degree of user resistance to the key negative factors, thereby reflecting the user's likes and dislikes for various factors of the content, so as to obtain a second recommendation list, so that the content information in the second recommendation list is not only related to the search keyword, but the recommendation order takes into account the quality of the content information itself and the user's preferences, thereby improving the user experience.

[0085] According to one embodiment of the present invention, in step S1, after the user enters the search keyword, multiple related content information related to the search keyword can be searched in the server. For example, the user opens a video website and enters the search keyword in the search engine of the website, then the server can push multiple videos related to the search keyword to the user.

[0086] Figure 2 The flowchart of searching for relevant content information according to an embodiment of the present invention is exemplarily shown.

[0087] According to one embodiment of the present invention, in step S1, upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword includes:

[0088] Step S11, obtaining the title, introduction information and attribute tags of each content information, wherein the attribute tags include tag information added by the author of the content information to the content information for describing the attributes of the content information;

[0089] Step S12, obtaining the title semantic information of the title, obtaining the introduction semantic information of the introduction information, and obtaining the label semantic information of the attribute label through the semantic recognition model;

[0090] Step S13, obtaining keyword semantic information of the search keyword through a semantic recognition model;

[0091] Step S14, for each content information, determining a first similarity between the keyword semantic information and the title semantic information, a second similarity between the keyword semantic information and the introduction semantic information, and a third similarity between the keyword semantic information and the tag semantic information;

[0092] Step S15, obtaining the maximum similarity among the first similarity, the second similarity and the third similarity;

[0093] Step S16: If the maximum similarity value corresponding to the content information is greater than or equal to a preset first similarity threshold, the content information is determined as relevant content information.

[0094] According to one embodiment of the present invention, in step S11, each piece of content information may include a title, introductory information, and attribute tags. The introductory information is a brief introduction to the content information, and the attribute tags are tags added by the author to describe the attributes of the content information. In this example, the content information of the video website is videos. Each video may have a title and introductory information. The author of the video may also add attribute tags to the video, such as "game video" or "animation video."

[0095] According to one embodiment of the present invention, in step S12, the semantic recognition model may be a recursive neural network model, etc. The present invention does not limit the specific type of the semantic recognition model. The semantic recognition model processes the title of the content information to obtain title semantic information. The semantic recognition model processes the introduction information of the content information to obtain introduction semantic information. The semantic recognition model processes the attribute tags of the content information to obtain tag semantic information. The title semantic information, introduction semantic information, and tag semantic information are all information in vector form, used to express the meaning of the title, introduction information, and attribute tag, respectively.

[0096] According to an embodiment of the present invention, in step S13, the search keyword may be processed by a semantic recognition model to obtain keyword semantic information. The keyword semantic information is information in a vector form and is used to express the meaning of the keyword.

[0097] According to one embodiment of the present invention, in step S14, for each content information, a first similarity between the keyword semantic information and the title semantic information, a second similarity between the keyword semantic information and the introduction semantic information, and a third similarity between the keyword semantic information and the tag semantic information of each tag information can be determined. For example, the cosine similarity between the keyword semantic information and the title semantic information can be solved as the first similarity, the cosine similarity between the keyword semantic information and the introduction semantic information can be solved as the second similarity, and the cosine similarity between the keyword semantic information and the tag semantic information of each tag information can be solved as the third similarity.

[0098] According to one embodiment of the present invention, in step S15, the maximum similarity value among the first similarity, the second similarity and the third similarity can be obtained as the overall similarity between the content involved in the content information and the search keyword. For example, the third similarity between the tag information of a video and the search keyword is low, and the first similarity between the title and the search keyword is low, but the second similarity between the introduction information of the video and the search keyword is high. It can be considered that the video involves the content that the user wants to search for in the introduction content, that is, the video involves the content that the user wants to search for. Therefore, the maximum value among the first similarity, the second similarity and the third similarity can be used as the overall similarity between the content involved in the video and the search keyword.

[0099] According to one embodiment of the present invention, in step S16, if the maximum similarity value corresponding to the content information (i.e., the overall similarity between the content involved in the content information and the search keyword) is greater than or equal to a preset first similarity threshold, the content information is determined to be relevant content information, that is, it is judged that the content information is relevant to the search keyword.

[0100] According to one embodiment of the present invention, in step S2, multiple related content information may be comprehensively sorted based on their relevance to the search keyword, their historical page views and historical interaction counts to obtain a first recommendation list.

[0101] Figure 3 The following is an exemplary flowchart of determining a first recommendation list according to an embodiment of the present invention.

[0102] According to one embodiment of the present invention, in step S2, a first recommendation list of multiple related content information is determined based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, including:

[0103] In step S21, the maximum similarity value corresponding to the relevant content information is determined as the correlation between the relevant content information and the search keyword;

[0104] In step S22, the interaction attraction coefficient of the relevant content information is determined based on the ratio of the historical interaction number and the historical pageview number of the relevant content information;

[0105] In step S23, a first recommendation coefficient of the relevant content information is determined based on the interaction attraction coefficient, the number of historical interactions, the historical page views, and the relevance;

[0106] In step S24, a first recommendation list of a plurality of related content information is determined according to the first recommendation coefficient.

[0107] According to one embodiment of the present invention, in step S21, as described above, the maximum similarity among the first similarity, the second similarity and the third similarity can be used as the overall similarity between the content involved in the content information and the search keyword, and therefore, can be used as the correlation between the relevant content information and the search keyword.

[0108] According to one embodiment of the present invention, in step S22, the ratio of historical interactions to historical views can be used to describe a user's interest in the relevant content information after viewing it. For example, after viewing the relevant content information, a user may be interested in a detail of the relevant content information and leave a message to discuss it. This process increases the number of historical evaluations of the relevant content information. A higher ratio of historical interactions to historical views indicates a higher likelihood that the relevant content information contains content that attracts the user. Therefore, this ratio can be determined as the interaction attraction coefficient of the relevant content information.

[0109] According to one embodiment of the present invention, in step S23, a first recommendation coefficient for each relevant content information can be determined based on the interaction attraction coefficient, the number of historical interactions, the number of historical views, and the relevance. That is, in addition to considering the relevance of the relevant content information to the search keyword, parameters such as the interaction attraction coefficient, the number of historical interactions, and the number of historical views that can reflect the quality of the relevant content information itself are also considered to obtain the first recommendation coefficient. The first recommendation coefficient is the basis for sorting the relevant content information. Relevant content information with a higher first recommendation coefficient is arranged at the front of the first recommendation list. Therefore, the recommendation order in the first recommendation list is not only related to the relevance of the relevant content information and the search keyword, but also to the quality of the relevant content information itself. In this example, the average of the historical views of multiple relevant content information can be solved, and a first ratio of the historical views of a relevant content information to the average of the historical views can be solved to serve as the ratio of the attractiveness of the relevant content information for browsing to the average attractiveness of the relevant content information for browsing. That is, the relative attractiveness of the relevant content information in attracting users to watch can indirectly reflect the quality of the relevant content information. In this example, the average number of historical interactions across multiple pieces of relevant content can be calculated, and a second ratio of the historical number of interactions for a piece of relevant content to the average number of historical interactions can be calculated to represent the ratio of the relevant content's appeal to comments relative to the average appeal of the multiple pieces of relevant content. This represents the relative appeal of the relevant content in attracting user comments, indirectly reflecting the quality of the relevant content. The weighted sum of the interaction attraction coefficient, the first ratio, the second ratio, and the relevance can be used to obtain the first recommendation coefficient for the relevant content.

[0110] According to one embodiment of the present invention, in step S24, after obtaining the first recommendation coefficient of each relevant content information, it can be sorted according to the first recommendation coefficient to obtain a first recommendation list, so that the sorting of the relevant content in the first recommendation list is not only related to the relevance between the relevant content information and the search keyword, but also related to the quality of the relevant content information itself, thereby improving the quality of the recommended content.

[0111] According to one embodiment of the present invention, in step S3, a first preset number of relevant content information with the highest first recommendation score can be obtained in the first recommendation list as the content information to be tested. The first preset number can be 10% of the total number of relevant content information in the first recommendation list, and the present invention does not impose any restrictions on this. The first recommendation score of the content information to be tested is relatively high, that is, the content information to be tested has a high correlation with the search keyword and has good quality itself, but the content information to be tested does not take into account factors of user likes and dislikes. When the content information to be tested is given priority recommendation, if the content information to be tested contains factors that the user dislikes or resists, the user may be given priority to watch the disliked content, thereby causing a decline in the user experience. Therefore, the factors of user likes and dislikes can be analyzed, and the order of the first recommendation list can be adjusted based on this, so that content that is relevant to the search keyword, has good quality, and is liked by the user is given priority.

[0112] According to one embodiment of the present invention, in step S4, it is possible to check whether the content information to be tested that is ranked high in the first recommendation list has key negative factors, that is, to find out the factors in the content information to be tested that may be disliked by users. If there are factors that users dislike, the order of the first recommendation list is changed so that the relevant content information that the user likes is recommended first.

[0113] Figure 4 The flowchart for determining key negative factors according to an embodiment of the present invention is exemplarily shown.

[0114] According to one embodiment of the present invention, in step S4, determining key negative factors based on the historical interaction information includes:

[0115] Step S41, determining the sentiment classification information of each historical evaluation information of the content information to be tested through the sentiment classification model;

[0116] Step S42, filtering negative evaluation information from multiple historical evaluation information based on the sentiment classification information;

[0117] Step S43, determining negative semantic information of the negative evaluation information through a semantic recognition model;

[0118] Step S44, clustering the negative semantic information to obtain a plurality of negative semantic information clusters;

[0119] Step S45 : determining the cluster centers of the multiple negative semantic information clusters as key negative factors.

[0120] According to one embodiment of the present invention, in step S41, the sentiment analysis model may be a recursive neural network model, which may be used to analyze the sentiment information expressed in the text to determine whether the text is positive, negative, or neutral. The sentiment analysis model may be used to classify the sentiment of each historical evaluation information of the content to be tested, for example, to classify each historical evaluation information as positive, negative, or neutral.

[0121] According to an embodiment of the present invention, in step S42, negative evaluation information with derogatory meanings may be screened out from a plurality of historical evaluation information of a plurality of content information to be tested.

[0122] According to one embodiment of the present invention, in step S43, the negative evaluation information with derogatory meaning is processed by the above-mentioned semantic recognition model to obtain negative semantic information. The semantic information can be used to determine which negative factors the negative evaluation information has. In other words, it can be used to determine which factors have caused historical users to make negative evaluations on the content information to be tested.

[0123] According to one embodiment of the present invention, in step S44, the negative semantic information may be clustered to obtain multiple negative semantic information clusters. The clustering process may include K-means clustering. The present invention does not limit the specific method of clustering, and the number of cluster centers (i.e., the number of negative semantic information clusters) can be manually set. For example, it can be set to 3, thereby obtaining three negative semantic information clusters. In subsequent processing, three key negative factors can also be obtained, i.e., the three key reasons why the content information to be tested receives negative evaluations. Of course, the number of cluster centers can also be determined using methods such as the silhouette coefficient, and other parameters of the clustering process can be set to default values.

[0124] According to one embodiment of the present invention, in step S45, the cluster centers of the multiple negative semantic information clusters may be determined as key negative factors. The key negative factors are information in vector form, and the dimension of the vector is the same as the dimension of the negative semantic information vector.

[0125] According to one embodiment of the present invention, in step S5, based on the user's historical browsing history, the introduction information of the content information to be tested and the key negative factors, it can be comprehensively determined which factors the user is more disgusted with, so as to adjust the order of the first recommendation list to give priority to recommending relevant content information that the user is interested in.

[0126] Figure 5 The following is an exemplary flowchart of determining the second recommendation list according to an embodiment of the present invention.

[0127] According to one embodiment of the present invention, in step S5, a second recommendation list of multiple related content information is determined based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors, including:

[0128] Step S51, determining the playback progress of the historical content information browsed by the user, the user interaction record of the historical content information, and the interaction information record of the historical content information based on the user's historical browsing record;

[0129] Step S52, determining the user's resistance coefficient to the key negative factor based on the playback progress, the user interaction record, the interaction information record, and the key negative factor;

[0130] Step S53, determining the publicity negation coefficient of the content information to be tested based on the introduction information of the content information to be tested and the key negative factors;

[0131] Step S54, obtaining historical interaction information of multiple related content information in the first recommendation list;

[0132] Step S55, determining a supplementary recommendation coefficient for each relevant content information based on the historical interaction information of the plurality of relevant content information and the key negative factors;

[0133] Step S56: determining a second recommendation list of multiple related content information based on the conflict coefficient, the publicity negation coefficient, and the supplementary recommendation coefficient.

[0134] According to an embodiment of the present invention, in step S51, the playback progress and user interaction records of the historical content information browsed by the user in the past can be obtained from the user's historical browsing history to help determine whether the user is resistant to key negative factors.

[0135] According to one embodiment of the present invention, in step S52, the resistance coefficient can be used to describe the user's resistance level to key negative factors. For example, by analyzing the user's historical browsing history, it can be determined that the user has made derogatory comments on historical content information with the same or similar negative factors, or that the historical content information with the same or similar negative factors has been played less, indicating that the user is not interested in the negative factor or even resists it.

[0136] According to one embodiment of the present invention, in step S52, based on the played progress, the user interaction record, the interaction information record and the key negative factor, the user's resistance coefficient to the key negative factor is determined, including: based on the interaction information record of the historical content information, selecting the target historical content information with the i-th key negative factor from multiple historical content information; determining the user's resistance coefficient to the i-th key negative factor according to formula (1) , (1), in, is the playback progress of the jth target historical content information with the i-th key negative factor, is the average playback progress of historical content information, is the semantic similarity between the user's historical evaluation information of the jth target historical content information with the i-th key negative factor and the i-th key negative factor. If the user has not evaluated the jth target historical content information with the i-th key negative factor, then , is the number of target historical content information with the i-th key negative factor, if (*) is the conditional function, j≤ , and i, j and All are positive integers.

[0137] According to one embodiment of the present invention, a semantic recognition model can be used to process interaction information records (including not only the user's historical evaluation information but also evaluation information from other users) of historical content information in a user's historical browsing history to obtain a semantic vector for each interaction information record. The semantic similarity (e.g., cosine similarity) between the semantic vector and a key negative factor is then determined. If, among multiple interaction information records for a piece of historical content information, there exists an interaction information record whose semantic similarity between the semantic vector and the key negative factor is greater than or equal to a threshold value (e.g., 0.8), the historical content information is determined to contain a key negative factor. The presence of the i-th key negative factor can be determined for each piece of historical content information using the aforementioned method.

[0138] According to one embodiment of the present invention, in formula (1), is the conditional function, In the case of Otherwise, the conditional function value is 0. If the playback progress of the jth target historical content information with the i-th key negative factor exceeds the average playback progress, it indicates that the user's playback of historical content information is not affected by the key negative factor, that is, the impact of the i-th key negative factor on the playback of the j-th target historical content information is 0, and the intensity of the user's resistance to the i-th negative factor is also 0, and the conditional function value can be determined to be 0. If the playback progress of the j-th target historical content information with the i-th key negative factor is lower than the average playback progress, it indicates that the i-th key negative factor has an impact on the playback of the j-th target historical content information, that is, the user has aversion or resistance to the i-th key negative factor, resulting in a low playback progress of the target historical content information. is the relative difference between the average playback progress and the playback progress of the target historical content information. The lower the playback progress of the target historical content information, the greater the relative difference, indicating that the i-th key negative factor has a greater impact on the playback of the j-th target historical content information. The calculation method of is similar to the semantic similarity between the semantic vector of the interactive information record and the key negative factors, which will not be repeated here. It can be used as the amplification coefficient of the impact of the i-th key negative factor on the playback of the j-th target historical content information, that is, The higher the value, the more likely the user is to make an evaluation of the target historical content information similar to the i-th key negative factor. In other words, the more likely the user is to have a low playback progress due to the i-th key negative factor, and the stronger the user's resistance to the i-th key negative factor. It can be used as the intensity of the user's resistance to the i-th negative factor. The average value of the intensity of the user's resistance to the i-th negative factor, determined based on multiple target historical content information, can be used as the user's resistance coefficient to the i-th key negative factor. The higher the resistance coefficient, the stronger the user's resistance to the i-th key negative factor. When recommending content to the user, the number of recommendations for content containing the i-th key negative factor should also be reduced.

[0139] In this way, the impact of key negative factors on playback can be determined through the playback progress of historical content information, thereby judging whether the user has resistance to the key negative factors. The intensity of the user's resistance to the key negative factors can also be amplified through the user's interaction records, thereby accurately reflecting the user's resistance to the key negative factors and providing an accurate data basis for recommending content that users are interested in.

[0140] According to one embodiment of the present invention, in step S53, the content to be tested may include introductory information, typically a positive introduction to the content to be tested. It is determined whether this introductory information conflicts with key negative factors. If so, the introductory information of the content to be tested is false, and the probability of recommending the content to be tested should be reduced. The publicity negation coefficient can be used to describe the likelihood that the introductory information is false.

[0141] According to one embodiment of the present invention, in step S53, based on the introduction information of the content information to be tested and the key negative factors, the publicity negation coefficient of the content information to be tested is determined, including: based on the historical interaction information of the content information to be tested, screening the target content information to be selected that has the i-th key negative factor; performing word segmentation and stop word removal on the introduction information of the target content information to be tested to obtain multiple introduction words of the introduction information; determining the introduction sentiment classification information of each introduction word through the sentiment classification model; determining the positive introduction words in the introduction words based on the introduction sentiment classification information; adding negative affixes to the positive introduction words to obtain negative introduction words; obtaining the negative introduction semantic information of the negative introduction words through the semantic recognition model; and determining the publicity negation coefficient of the k-th target content information to be tested for the i-th key negative factor according to formula (2) , (2), in, is the negative introduction semantic information of the sth negative introduction word of the kth target content information to be tested, is the i-th key negative factor, for and where N is the number of key negative factors, M is the number of target content information to be tested, max(*) is the maximum value function, s≤M, i≤N, and s, M, i and N are all positive integers.

[0142] According to one embodiment of the present invention, the method for screening target content information to be selected is similar to the above method for screening target historical content information, and will not be described in detail here. The introduction information of the content information to be tested is usually a paragraph, including a large number of words. The introduction information can be pre-processed by word segmentation and removal of stop words to obtain multiple introduction words. The introduction sentiment classification information of each introduction word can be determined by the sentiment classification model, and words with commendatory meanings, that is, positive introduction words, can be screened out. Negative affixes are added to the positive introduction words to obtain negative introduction words, that is, the commendatory words are modified to derogatory words, for example, "clear" is modified to "unclear", and then the negative introduction words can be processed by the semantic recognition model to obtain negative introduction semantic information.

[0143] According to one embodiment of the present invention, in formula (2), Can be and The higher the cosine similarity between the two, the higher the semantic similarity between the negative introduction word and the i-th key negative factor, indicating that the closer the semantics of the negative introduction word is to the i-th key negative factor, the greater the possibility that the positive introduction word corresponding to the negative introduction word is false advertising. This is the maximum semantic similarity between multiple negative description words and the i-th key negative factor. The larger this maximum value, the more likely the description contains false advertising. For example, if the description contains the word "clear picture quality" and the i-th key negative factor indicates unclear picture quality, then the description contains false advertising.

[0144] In this way, negative introduction words can be obtained by adding negative suffixes to the positive introduction words in the introduction information, and the semantic similarity between the negative introduction words and the key negative factors can be solved to determine whether the content information to be tested contains false propaganda, thereby reducing the recommendation probability of content with false propaganda and improving user experience.

[0145] According to one embodiment of the present invention, in step S54, historical interaction information of multiple related content information can be obtained, and in step S55, a supplementary recommendation coefficient for each related content information is determined. The supplementary recommendation coefficient can increase the probability of related content information without key negative factors being recommended, thereby reducing the probability of related content information with key negative information being recommended first, thereby improving user experience.

[0146] According to one embodiment of the present invention, in step S55, based on the historical interaction information of the plurality of relevant content information and the key negative factors, a supplementary recommendation coefficient of each relevant content information is determined, including: determining the interactive emotion classification information of the historical interaction information through the emotion classification model; screening out positive historical interaction information from the historical interaction information based on the interactive emotion classification information; adding negative affixes to the positive historical interaction information to obtain modified historical interaction information; obtaining modified interaction semantic information of the modified historical interaction information through the semantic recognition model; and determining the supplementary recommendation coefficient of the t-th relevant content information for the i-th key negative factor according to formula (3) , (3), in, is the modified interaction semantic information of the xth historical interaction information of the tth related content information, is the i-th key negative factor, for and The semantic similarity of is the number of historical interactive information of the t-th related content information, x≤ , and x and All are positive integers.

[0147] According to one embodiment of the present invention, the interactive sentiment classification information of historical interactive information can be determined through a sentiment classification model, that is, whether each piece of historical interactive information is commendatory or derogatory. Furthermore, based on the interactive sentiment classification information, positive historical interactive information with commendatory meaning can be screened out. A negative affix can be added to the positive historical interactive information (if the positive historical interactive information is a commendatory word, a negative affix can be added to the commendatory word; if the positive historical interactive information is a sentence including multiple commendatory words, a negative affix can also be added to each commendatory word in the positive historical interactive information) to obtain derogatory modified historical interactive information, and the modified interaction semantic information of the modified historical interactive information can be obtained through a semantic recognition model.

[0148] According to one embodiment of the present invention, in formula (3), Can be and The higher the cosine similarity between the two, the higher the semantic similarity between the modified historical interaction information and the i-th key negative factor, indicating that the semantics of the modified historical interaction information is closer to the i-th key negative factor, and the positive historical interaction information corresponding to the modified historical interaction information is more able to indicate that the relevant content information does not have the problem corresponding to the i-th key negative factor. To modify the maximum semantic similarity between historical interaction information and the i-th key negative factor, the larger the maximum value, the more likely the relevant content information does not have the issue corresponding to the i-th key negative factor. For example, if there is a comment "clear image quality" in historical interaction information, and the i-th key negative factor is used to indicate unclear image quality, since historical interaction information is objective comments from multiple historical users on the relevant content information, it is highly likely that the relevant content information does not have unclear image quality issues. The probability of relevant content information without key negative factors being recommended can be increased by supplementing the recommendation coefficient.

[0149] In this way, negative suffixes can be added to the positive historical interaction information in the objective comments of historical users, and the semantic similarity between the modified historical interaction information and the key negative factors can be determined, thereby determining the possibility that the key negative factors do not exist in the relevant content information, and obtaining a supplementary recommendation coefficient, which can be used to increase the probability of relevant content information without key negative factors being recommended, thereby improving user experience.

[0150] According to one embodiment of the present invention, in step S56, the first recommendation coefficient of each relevant content information can be adjusted based on the above-determined resistance coefficient, publicity denial coefficient and supplementary recommendation coefficient, so as to give priority to recommending content with good quality and user interest.

[0151] According to one embodiment of the present invention, in step S56, a second recommendation list of multiple related content information is determined based on the conflict coefficient, the publicity negation coefficient, and the supplementary recommendation coefficient, including: if the t-th related content information is not the target content information to be tested, then the gain recommendation coefficient of the t-th related content information is determined according to formula (4). , (4), in, is the number of positive historical interaction information of the t-th related content information, is the number of historical interactive information of the t-th related content information, is the average positive interaction rate of multiple related content information, is the supplementary recommendation coefficient of the t-th relevant content information for the i-th key negative factor, is the user's resistance coefficient to the i-th key negative factor, N is the number of key negative factors, i≤N, and both i and N are positive integers; if t relevant content information is the target content information to be tested, then the gain recommendation coefficient of the t-th relevant content information is determined according to formula (5) , (5), in, is the publicity negation coefficient of the t-th relevant content information for the i-th key negative factor; and a second recommendation list is determined according to the gain recommendation coefficient of each relevant content information.

[0152] According to one embodiment of the present invention, in formula (4), is the gain factor based on the supplementary recommended factor, as described above, The higher the value, the greater the possibility that there are no key negative factors in the relevant content information, which can increase the probability of being recommended. Therefore, a coefficient greater than 1 can be used. Increase the probability of relevant content information without key negative factors being recommended. is the attenuation coefficient based on the user’s resistance coefficient to the i-th key negative factor, The higher the coefficient, the stronger the user's resistance to the i-th key negative factor, and the lower the probability of being recommended first. Therefore, a coefficient less than 1 is used. Reduce the probability of being recommended first. is the adjustment coefficient of the t-th relevant content information for the i-th key negative factor, It is the average value of the adjustment coefficients of the t-th relevant content information for multiple key negative factors, which can be used to adjust the first recommendation score of the t-th relevant content information based on whether the relevant content information has various key negative factors and the user's resistance to various key negative factors. is the positive interaction rate of the t-th related content information, that is, the proportion of historical users who have made positive comments on the t-th related content information, is the average positive interaction rate of each relevant content information, The positive interaction rate of the t-th related content information relative to the whole. If the value is high (for example, greater than 1), it means that the probability of the related content information being positively evaluated by users is high, and the quality of the related content information is relatively good. and By multiplying them together, we can get the gain recommendation coefficient of the t-th relevant content information, so that we can get a comprehensive adjustment coefficient (i.e., gain recommendation coefficient) based on the quality of the relevant content, whether it contains key negative factors, and the user's resistance to negative key factors, thereby increasing the probability that relevant content information with good quality and no key negative factors will be recommended first, thereby improving the user experience.

[0153] According to one embodiment of the present invention, in formula (5), is the attenuation coefficient based on the publicity negation coefficient. When the t-th related content information is the target content information to be tested, The larger the value is, the higher the possibility that the t-th related content information contains false propaganda, and the probability of being recommended first should be reduced. Therefore, by using a coefficient less than 1 Reduce the probability that the t-th related content information is recommended first. In the case that the t-th related content information is the target content information to be tested The adjustment coefficient of the t-th relevant content information for the i-th key negative factor. The gain recommendation coefficient can be obtained by comprehensively considering the quality of the relevant content information, whether there is false propaganda, whether it contains key negative factors, and the user's resistance to negative key factors. .

[0154] According to one embodiment of the present invention, the first recommendation score of each relevant content information is multiplied by the respective gain recommendation coefficient to obtain the second recommendation coefficient of each relevant content information. Sorting is performed by the second recommendation coefficient to obtain a second recommendation list, which can give priority to recommending relevant content information with good quality, no false propaganda and no negative key factors to improve user experience.

[0155] This method allows for the classification and calculation of target content and other related content, further reducing the recommendation coefficient for content that exhibits false advertising. The recommendation coefficient for each piece of relevant content can be adjusted based on multiple factors, including the quality of the relevant content, the presence of false advertising, the inclusion of key negative factors, and user resistance to these key negative factors. This prioritizes recommendations for high-quality content that lacks false advertising and negative key factors, thereby enhancing the user experience.

[0156] According to one embodiment of the present invention, in step S6, multiple relevant content information can be displayed to the user in sequence according to the order in the second recommendation list, and relevant content with good quality, no false propaganda and no negative key factors can be recommended first, which can improve the user experience.

[0157] According to the content recommendation method based on traffic monitoring of an embodiment of the present invention, after obtaining relevant content information related to the search keyword, it can indirectly reflect the quality of the relevant content information based on the historical number of views and historical interactions of the relevant content information, thereby recommending to the user in combination with the quality of the relevant content information, and can also find the key negative factors of the content information to be tested, and can determine the user's resistance to the key negative factors, thereby reflecting the user's likes and dislikes of various factors of the content, so as to obtain a second recommendation list, so that the content information in the second recommendation list is not only related to the search keyword, but the recommendation order takes into account the quality of the content information itself and the user's preferences, thereby improving the user experience. When determining the resistance coefficient, the impact of the key negative factors on the playback can be determined by the playback progress of the historical content information, thereby determining whether the user has a resistance to the key negative factors. The intensity of the user's resistance to the key negative factors can also be amplified by the user's interaction records, thereby accurately reflecting the user's resistance to the key negative factors and providing an accurate data basis for recommending content that the user is interested in. When determining the negative advertising coefficient, negative introductory words can be obtained by adding negative affixes to positive introductory words in the introductory information. The semantic similarity between the negative introductory words and key negative factors can be calculated to determine whether the content under test contains false advertising. This can reduce the recommendation probability of content containing false advertising and improve the user experience. When determining the supplementary recommendation coefficient, negative affixes can be added to positive historical interaction information in historical user objective comments. The semantic similarity between the modified historical interaction information and key negative factors can be determined to determine the likelihood that the relevant content does not contain key negative factors. The supplementary recommendation coefficient can be used to increase the recommendation probability of relevant content without key negative factors, thereby improving the user experience. When determining the gain recommendation coefficient, the target content under test and other related content can be classified and calculated to further reduce the recommendation coefficient of the target content under test that contains false advertising. The recommendation coefficient of each relevant content can also be adjusted based on multiple factors, such as the quality of the relevant content, the presence of false advertising, the presence of key negative factors, and the user's resistance to key negative factors. This can prioritize the recommendation of relevant content with high quality, no false advertising, and no negative key factors, thereby improving the user experience.

[0158] Figure 6 A block diagram of a content recommendation system based on traffic monitoring according to an embodiment of the present invention is exemplarily shown. The system includes:

[0159] A related content information module is configured to, upon receiving a user's search keyword, determine a plurality of related content information related to the search keyword;

[0160] A first recommendation list module is configured to determine a first recommendation list of multiple related content information based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information;

[0161] A historical interaction information module is configured to determine a first preset number of pieces of content information to be tested in the first recommendation list, and obtain historical interaction information of the content information to be tested, wherein the historical interaction information includes historical evaluation information of the content information to be tested;

[0162] a key negative factor module, configured to determine key negative factors based on the historical interaction information;

[0163] A second recommendation list module is used to determine a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors;

[0164] A display module is used to display multiple related content information according to the second recommendation list.

[0165] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0166] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

[0167] Finally, 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A content recommendation method based on traffic monitoring, characterized in that: include: Upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword; Determining a first recommendation list of the plurality of related content information based on the relevance of the plurality of related content information to the search keyword, the historical page views and the historical interaction counts of the plurality of related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information; In the first recommendation list, a first preset number of content information to be tested is determined, and historical interaction information of the content information to be tested is obtained, wherein the historical interaction information includes content of historical evaluation information of the content information to be tested; Determining key negative factors based on the historical interaction information; Determining a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors; According to the second recommendation list, multiple related content information is displayed.

2. The content recommendation method based on traffic monitoring according to claim 1, characterized in that: Upon receiving a user's search keyword, determining a plurality of relevant content information related to the search keyword includes: Acquire the title, introduction information and attribute tag of each content information, wherein the attribute tag includes tag information added by the author of the content information to the content information for describing the attributes of the content information; Using a semantic recognition model, obtain the title semantic information of the title, obtain the introduction semantic information of the introduction information, and obtain the label semantic information of the attribute label; Obtain keyword semantic information of search keywords through semantic recognition model; For each piece of content information, determining a first similarity between the keyword semantic information and the title semantic information, a second similarity between the keyword semantic information and the introduction semantic information, and a third similarity between the keyword semantic information and the tag semantic information; Obtaining a maximum similarity among the first similarity, the second similarity, and the third similarity; If the maximum similarity value corresponding to the content information is greater than or equal to a preset first similarity threshold, the content information is determined to be relevant content information.

3. The content recommendation method based on traffic monitoring according to claim 2, characterized in that: Determining a first recommendation list of the plurality of related content information based on the relevance of the plurality of related content information to the search keyword, the historical page views and the historical interaction counts of the plurality of related content information includes: Determining the maximum similarity value corresponding to the relevant content information as the relevance between the relevant content information and the search keyword; Determine the interaction attraction coefficient of the relevant content information based on the ratio of the historical interaction number and the historical pageview number of the relevant content information; Determining a first recommendation coefficient for relevant content information based on the interaction attraction coefficient, the number of historical interactions, the historical page views, and the relevance; A first recommendation list of multiple related content information is determined according to the first recommendation coefficient.

4. The content recommendation method based on traffic monitoring according to claim 1, characterized in that: Based on the historical interaction information, key negative factors are determined, including: Determine the sentiment classification information of each historical evaluation information of the content information to be tested through the sentiment classification model; Filtering negative evaluation information from multiple historical evaluation information based on the sentiment classification information; Determine the negative semantic information of negative evaluation information through semantic recognition model; performing clustering processing on the negative semantic information to obtain a plurality of negative semantic information clusters; The cluster centers of multiple negative semantic information clusters are identified as key negative factors.

5. The content recommendation method based on traffic monitoring according to claim 1, characterized in that: Determining a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors, includes: Determine, based on the user's historical browsing history, the progress of playback of historical content information browsed by the user, the user's interaction history with the historical content information, and the interaction history of the historical content information; determining the user's resistance coefficient to the key negative factor based on the playback progress, the user interaction record, the interaction information record, and the key negative factor; Determining a publicity negation coefficient of the content information to be tested based on the introduction information of the content information to be tested and the key negative factors; Obtain historical interaction information of multiple related content information in the first recommendation list; Determining a supplementary recommendation coefficient for each piece of relevant content information based on historical interaction information of the plurality of relevant content information and the key negative factors; A second recommendation list of multiple related content information is determined based on the resistance coefficient, the publicity negation coefficient and the supplementary recommendation coefficient.

6. The content recommendation method based on traffic monitoring according to claim 5, characterized in that: Determining the user's resistance coefficient to the key negative factor based on the playback progress, the user interaction record, the interaction information record, and the key negative factor, including: According to the interactive information record of the historical content information, select the target historical content information having the i-th key negative factor from the multiple historical content information; According to the formula ， Determine the user's resistance coefficient to the i-th key negative factor ,in, is the playback progress of the jth target historical content information with the i-th key negative factor, is the average playback progress of historical content information, is the semantic similarity between the user's historical evaluation information of the jth target historical content information with the i-th key negative factor and the i-th key negative factor. If the user has not evaluated the jth target historical content information with the i-th key negative factor, then , is the number of target historical content information with the i-th key negative factor, if (*) is the conditional function, j≤ , and i, j and All are positive integers.

7. The content recommendation method based on traffic monitoring according to claim 5, characterized in that: Determine the publicity negation coefficient of the content to be tested based on the introduction information of the content to be tested and the key negative factors, including: Based on the historical interaction information of the content to be tested, the target content to be selected is screened for the presence of the i-th key negative factor; Segmenting the introduction information of the target content information and removing stop words to obtain multiple introduction words of the introduction information; Determine the introductory sentiment classification information of each introductory word through the sentiment classification model; According to the introduction sentiment classification information, positive introduction words are determined among the introduction words; Add negative affixes to positive introductory words to obtain negative introductory words; Obtain negative introduction semantic information of negative introduction words through semantic recognition model; According to the formula ， Determine the negative coefficient of the kth target content information for the ith key negative factor ,in, is the negative introduction semantic information of the sth negative introduction word of the kth target content information to be tested, is the i-th key negative factor, for and where N is the number of key negative factors, max(*) is the maximum value function, i≤N, and s, M, i and N are all positive integers.

8. The content recommendation method based on traffic monitoring according to claim 5, characterized in that: Based on the historical interaction information of the plurality of relevant content information and the key negative factors, a supplementary recommendation coefficient for each relevant content information is determined, including: Determine the interactive sentiment classification information of historical interactive information through the sentiment classification model; According to the interaction sentiment classification information, positive historical interaction information is filtered out from the historical interaction information; Add negative affixes to positive historical interaction information to obtain modified historical interaction information; Obtaining modification interaction semantic information of modification history interaction information through a semantic recognition model; According to the formula ， Determine the supplementary recommendation coefficient of the tth relevant content information for the ith key negative factor ,in, is the modified interaction semantic information of the xth historical interaction information of the tth related content information, is the i-th key negative factor, for and The semantic similarity of is the number of historical interactive information of the t-th related content information, x≤ , and x and All are positive integers.

9. The content recommendation method based on traffic monitoring according to claim 7, characterized in that: According to the resistance coefficient, the publicity negation coefficient, and the supplementary recommendation coefficient, a second recommendation list of multiple related content information is determined, including: If the tth related content information is not the target content information to be tested, then according to the formula ， Determine the gain recommendation coefficient of the tth relevant content information ,in, is the number of positive historical interaction information of the t-th related content information, is the number of historical interactive information of the t-th related content information, is the average positive interaction rate of multiple related content information, is the supplementary recommendation coefficient of the t-th relevant content information for the i-th key negative factor, is the user's resistance coefficient to the i-th key negative factor, N is the number of key negative factors, i≤N, and both i and N are positive integers; If t relevant content information is the target content information to be tested, then according to the formula ， Determine the gain recommendation coefficient of the tth relevant content information ,in, is the publicity negation coefficient of the t-th relevant content information for the i-th key negative factor; A second recommendation list is determined according to the gain recommendation coefficients of the respective related content information.

10. A content recommendation system based on traffic monitoring, characterized in that: include: A related content information module is configured to, upon receiving a user's search keyword, determine a plurality of related content information related to the search keyword; A first recommendation list module is configured to determine a first recommendation list of multiple related content information based on the relevance of the multiple related content information to the search keyword, the historical page views and the historical interaction counts of the multiple related content information, wherein the historical interaction counts include the number of historical evaluation information of the content information; A historical interaction information module is configured to determine a first preset number of pieces of content information to be tested in the first recommendation list, and obtain historical interaction information of the content information to be tested, wherein the historical interaction information includes historical evaluation information of the content information to be tested; a key negative factor module, configured to determine key negative factors based on the historical interaction information; A second recommendation list module is used to determine a second recommendation list of multiple related content information based on the user's historical browsing history, the introduction information of the content information to be tested, and the key negative factors; A display module is used to display multiple related content information according to the second recommendation list.

Images