Content recommendation method and device

Abstract

The invention provides a content recommendation method and device, and relates to the field of data processing, in particular to the field of artificial intelligence. According to the specific implementation scheme, a plurality of candidate recommended languages corresponding to current content browsed by a user are obtained; obtaining a click rate prediction value of the user for each candidate recommendation language to determine a target recommendation language from the plurality of candidate recommendation languages, and generating a recommendation language card based on the target recommendation language and displaying the recommendation language card on a first display page; in response to the monitored confirmation operation of the user on the recommended language card, recalling multiple target recommended contents from the candidate recommended contents according to the target recommended language; and generating a recommendation list according to the multiple target recommendation contents, and displaying the target recommendation contents on a second display page according to the sequence of the recommendation list.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, in particular to the field of artificial intelligence, and specifically to a content recommendation method and device thereof. Background Art

[0002] In the current recommendation system, for example, when a user is watching a video, a fixed recommendation will be displayed on the video page. Users usually interact by clicking on the fixed recommendation to obtain resources related to the recommendation. For the same video, all users may see the same fixed recommendation, and the recommended content they see on the second-jump page after the recommendation is fixed, resulting in a lack of diversity and freshness in the user experience. Summary of the invention

[0003] The present disclosure provides a content recommendation method, apparatus, device and storage medium.

[0004] According to one aspect of the present disclosure, a content recommendation method is provided, including: obtaining multiple candidate recommendation terms corresponding to current content browsed by a user; obtaining a predicted click rate value of the user for each candidate recommendation term to determine a target recommendation term from the multiple candidate recommendation terms, and generating a recommendation term card based on the target recommendation term to display on a first display page; in response to monitoring a user's confirmation operation on the recommendation term card, recalling multiple target recommendation contents from the candidate recommendation contents according to the target recommendation term; generating a recommendation list based on the multiple target recommendation contents, and displaying the target recommendation contents on a second display page in the order of the recommendation list.

[0005] This application obtains multiple candidate recommendation terms for the content currently browsed by the user and predicts the click-through rate of each candidate recommendation term. It can accurately select the most appropriate recommendation term based on the user's interests and behaviors, ensuring the relevance and attractiveness of the recommendation term, thereby improving the user's click-through rate and engagement; based on the target recommendation term, multiple target recommended content is recalled, and a recommendation list is generated and displayed on the second display page, so that the user can browse more content related to the target recommendation term, reducing the user's search time cost and enhancing the user experience.

[0006] According to an embodiment of the present application, obtaining multiple candidate recommendation words corresponding to current content browsed by a user includes: obtaining multiple recommendation word prompts corresponding to the current content; and generating multiple candidate recommendation words corresponding to the current content according to the multiple recommendation word prompts.

[0007] According to an embodiment of the present application, multiple candidate recommendation words corresponding to the current content are generated based on multiple recommendation word prompts, including: obtaining multiple related content similar to the current content; merging the current content and the related content to generate a content set; and generating multiple candidate recommendation words corresponding to the current content based on the content set and the multiple recommendation word prompts.

[0008] According to an embodiment of the present application, acquiring multiple related contents similar to the current content includes: acquiring multiple related contents belonging to the same content dimension as the current content; or acquiring multiple related contents belonging to the same cluster as the current content.

[0009] According to an embodiment of the present application, obtaining multiple candidate recommendation terms corresponding to the current content browsed by the user includes: obtaining a target identifier corresponding to the current content; obtaining a mapping relationship between the identifier and the recommendation terms; and querying the mapping relationship based on the target identifier to determine the multiple candidate recommendation terms corresponding to the current content.

[0010] According to one embodiment of the present application, the user's click rate prediction value for each candidate recommendation term is obtained, including: performing semantic feature extraction on each candidate recommendation term to obtain multiple first semantic vectors; performing semantic feature extraction on user information corresponding to the user to obtain a second semantic vector; performing semantic feature extraction on the current content to obtain a third semantic vector; for any candidate recommendation term, performing click rate prediction on the candidate recommendation term based on the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation term to obtain the click rate prediction value of the candidate recommendation term.

[0011] According to one embodiment of the present application, a recommendation list is generated based on multiple target recommended contents, including: performing semantic feature extraction on the target recommendation words to obtain a fourth semantic vector; performing semantic feature extraction on each target recommended content to obtain multiple fifth semantic vectors; respectively calculating the vector similarity / vector distance between the fourth semantic vector and each fifth semantic vector; and determining the arrangement order of the multiple target recommended contents based on the vector similarity / vector distance to generate a recommendation list.

[0012] According to one embodiment of the present application, in response to the current content being text content, semantic feature extraction is performed on the current content to obtain a third semantic vector, and semantic feature extraction is performed on each target recommended content to obtain multiple fifth semantic vectors, including: for any content in the current content and the target recommended content, any content is input into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0013] According to one embodiment of the present application, in response to the current content being video content, semantic feature extraction is performed on the current content to obtain a third semantic vector, and semantic feature extraction is performed on each target recommended content to obtain multiple fifth semantic vectors, including: for any content in the current content and the target recommended content, obtaining text information corresponding to any content; inputting the text information into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0014] According to an embodiment of the present application, multiple candidate recommendation words corresponding to the current content are generated based on multiple recommendation word prompts, including: obtaining preset format parameters corresponding to the recommendation word card; determining recommendation word constraint parameters based on the preset format parameters; and generating multiple candidate recommendation words corresponding to the current content based on the recommendation word constraint parameters and multiple recommendation word prompts.

[0015] According to another aspect of the present disclosure, a content recommendation device is provided, including: an acquisition module, used to acquire multiple candidate recommendation terms corresponding to the current content browsed by a user; a first display module, used to acquire the user's click-through rate prediction value for each candidate recommendation term to determine a target recommendation term from the multiple candidate recommendation terms, and generate a recommendation term card based on the target recommendation term to display on a first display page; a recall module, used to recall multiple target recommendation contents from the candidate recommendation contents according to the target recommendation term in response to monitoring the user's confirmation operation on the recommendation term card; and a second display module, used to generate a recommendation list based on the multiple target recommendation contents, and display the target recommendation contents on a second display page in the order of the recommendation list.

[0016] According to an embodiment of the present application, the acquisition module is further used to: acquire multiple recommendation prompt words corresponding to the current content; and generate multiple candidate recommendation words corresponding to the current content according to the multiple recommendation prompt words.

[0017] According to an embodiment of the present application, the acquisition module is further used to: acquire multiple related contents similar to the current content; merge the current content and the related contents to generate a content set; and generate multiple candidate recommendation words corresponding to the current content according to the content set and the multiple recommendation word prompts.

[0018] According to an embodiment of the present application, the acquisition module is further used to: acquire multiple related contents belonging to the same content dimension as the current content; or acquire multiple related contents belonging to the same cluster as the current content.

[0019] According to an embodiment of the present application, the acquisition module is further used to: acquire a target identifier corresponding to the current content; acquire a mapping relationship between the identifier and the recommendation word; and query the mapping relationship according to the target identifier to determine multiple candidate recommendation words corresponding to the current content.

[0020] According to one embodiment of the present application, the first display module is also used to: perform semantic feature extraction on each candidate recommendation term to obtain multiple first semantic vectors; perform semantic feature extraction on user information corresponding to the user to obtain a second semantic vector; perform semantic feature extraction on the current content to obtain a third semantic vector; for any candidate recommendation term, perform click-through rate prediction on the candidate recommendation term based on the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation term to obtain a click-through rate prediction value of the candidate recommendation term.

[0021] According to one embodiment of the present application, the second display module is also used to: perform semantic feature extraction on the target recommendation term to obtain a fourth semantic vector; perform semantic feature extraction on each target recommended content to obtain multiple fifth semantic vectors; respectively calculate the vector similarity / vector distance between the fourth semantic vector and each fifth semantic vector; determine the arrangement order of multiple target recommended contents according to the vector similarity / vector distance to generate a recommendation list.

[0022] According to one embodiment of the present application, in response to the current content being text content, the first display module and the second display module are also used to: for any content in the current content and the target recommended content, input any content into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0023] According to one embodiment of the present application, in response to the current content being video content, the first display module and the second display module are also used to: obtain text information corresponding to any content in the current content and the target recommended content; input the text information into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0024] According to an embodiment of the present application, the acquisition module is further used to: acquire preset format parameters corresponding to the recommendation word card; determine the recommendation word constraint parameters according to the preset format parameters; and generate multiple candidate recommendation words corresponding to the current content according to the recommendation word constraint parameters and multiple recommendation word prompts.

[0025] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned content recommendation method.

[0026] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the above-mentioned content recommendation method.

[0027] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program implements the above content recommendation method when executed by a processor.

[0028] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0030] Figure 1 It is a schematic diagram of an exemplary implementation of a content recommendation method according to an exemplary embodiment of the present disclosure.

[0031] Figure 2 The present invention is a schematic diagram of generating a recommendation card and recommending content after clicking the recommendation card according to an exemplary embodiment of the present disclosure.

[0032] Figure 3 It is a schematic diagram of an exemplary implementation of a content recommendation method according to an exemplary embodiment of the present disclosure.

[0033] Figure 4 is a flowchart of an exemplary implementation of a content recommendation method according to an exemplary embodiment of the present disclosure.

[0034] Figure 5 is a schematic diagram of a content recommendation device according to an exemplary embodiment of the present disclosure.

[0035] Figure 6 is a schematic diagram of an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0036] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0037] Artificial Intelligence (AI) is a discipline that studies how computers can simulate certain thought processes and intelligent behaviors of human beings (such as learning, reasoning, thinking, planning, etc.). It includes both hardware-level technologies and software-level technologies. AI hardware technologies generally include computer vision technology, speech recognition technology, natural language processing technology, as well as learning / deep learning, big data processing technology, knowledge graph technology, and other aspects.

[0038] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0039] Figure 1 is a schematic diagram of an exemplary implementation of a content recommendation method shown in the present application, such as Figure 1 As shown, the content recommendation method includes the following steps:

[0040] S101, obtaining a plurality of candidate recommendation words corresponding to the current content browsed by the user.

[0041] Among them, the present application can be applied to a variety of recommendation scenarios, such as a user browsing scenario for articles or a user browsing scenario for videos.

[0042] The candidate recommendation words refer to the recommendation words related to the content currently being browsed that guide the user to click. The current content corresponds to multiple candidate recommendation words for subsequent selection of the target recommendation words most relevant to the current user from the multiple candidate recommendation words.

[0043] For example, if the user is currently browsing a video related to social security, the candidate recommendation phrases may be "What are the consequences of not paying social security", "What is social security", "What is the relationship between social security and medical insurance", etc.

[0044] S102, obtaining a predicted click rate value of each candidate recommendation word by the user to determine a target recommendation word from a plurality of candidate recommendation words, and generating a recommendation word card based on the target recommendation word to display on a first display page.

[0045] Optionally, the features of each candidate recommendation, the features of the current content, and the features of the user are input into a pre-trained click rate prediction model to obtain a click rate prediction value output by the model, and the candidate recommendation with the largest click rate prediction value is used as the target recommendation.

[0046] Figure 2 is a schematic diagram of generating a recommendation card and performing content recommendation after clicking the recommendation card, as shown in the present application. Figure 2 As shown, taking the target recommendation displayed on the recommendation card as "What to pay attention to about social security" as an example, the display page where the current content is located is used as the first display page, and the recommendation card can be located in the lower left corner or lower right corner of the first display page, trying not to block the current content.

[0047] S103 , in response to monitoring a confirmation operation of the user on the recommendation word card, recalling a plurality of target recommendation contents from the candidate recommendation contents according to the target recommendation word.

[0048] In the present application, there is no restriction on the user's confirmation operation on the recommendation card. The user can click on the recommendation card to confirm the recommendation card, or confirm the recommendation card through voice confirmation, that is, instruct to obtain multiple target recommendation contents related to the target recommendation on the recommendation card.

[0049] Among them, when recalling multiple target recommended contents from candidate recommended contents, the target recommendation words, the user behavior characteristics corresponding to the current user and multiple candidate recommended contents can be input into a pre-trained recall model together, and the result output by the recall model can be used as the target recommended content.

[0050] S104: Generate a recommendation list according to the multiple target recommended contents, and display the target recommended contents on the second display page according to the order of the recommendation list.

[0051] After the above-mentioned recall of multiple target recommended contents, the multiple target recommended contents are finely ranked. For example, a fine ranking model can be constructed according to multiple indicators (user actual click rate, user satisfaction, user stay time, content relevance, etc.), so as to sort the multiple target recommended contents based on the trained fine ranking model to generate a recommendation list, and display the target recommended contents on the second display page in the order of the recommendation list.

[0052] Optionally, the refined ranking model may adopt a linear regression model, a support vector machine model, a deep neural network model, or the like.

[0053] like Figure 2 As shown, the second display page is the jump page after the user clicks on the recommendation card, which displays target recommended content 1, target recommended content 2, target recommended content 3, etc. in sequence. The user can click on the displayed target recommended content to browse the complete content of the target recommended content, or click the return button to return to the first display page to continue browsing the page content of the first display page.

[0054] The embodiment of the present application proposes a content recommendation method, including: obtaining multiple candidate recommendation terms corresponding to the current content browsed by the user; obtaining the predicted click rate value of each candidate recommendation term by the user to determine the target recommendation term from the multiple candidate recommendation terms, and generating a recommendation term card based on the target recommendation term to display on the first display page; in response to monitoring the user's confirmation operation on the recommendation term card, recalling multiple target recommendation contents from the candidate recommendation contents according to the target recommendation term; generating a recommendation list according to the multiple target recommendation contents, and displaying the target recommendation contents on the second display page in the order of the recommendation list. By obtaining multiple candidate recommendation terms for the content currently browsed by the user and predicting the click rate of each candidate recommendation term, the present application can accurately select the most appropriate recommendation term according to the user's interests and behaviors, ensure the relevance and attractiveness of the recommendation term, and thus improve the user's click rate and engagement; recalling multiple target recommendation contents based on the target recommendation term, and generating a recommendation list to display on the second display page, so that the user can browse more content related to the target recommendation term, reducing the time cost of the user's search, and enhancing the user experience.

[0055] Figure 3is a schematic diagram of an exemplary implementation of a content recommendation method shown in the present application, such as Figure 3 As shown, the content recommendation method includes the following steps:

[0056] S301, obtaining multiple recommendation prompt words corresponding to the current content.

[0057] The recommendation prompt words are keywords or phrases related to the current content, which are intended to guide the generation process of the recommendation words. The recommendation prompt words usually reflect the main elements of the current content or aspects that the user may be interested in.

[0058] S302: Generate multiple candidate recommendation words corresponding to the current content according to multiple recommendation word prompt words.

[0059] As a feasible method, the content information of the current content (such as the text content of the video) and multiple recommendation prompts are directly input into the pre-trained large language model to obtain multiple candidate recommendation words corresponding to the current content. For example, for the content of "healthy diet", the generated candidate recommendation words may include "explore the most effective diet plan", "get advice from nutrition experts", "discover healthy recipes suitable for you" and other candidate recommendation words.

[0060] As another feasible method, when generating multiple candidate recommendation words corresponding to the current content, first obtain multiple related contents similar to the current content. For example, if the current content is a food-related video, another 100 food-related videos can be obtained from the video library as related contents similar to the current content. The current content and the related content are merged to generate a content set, and multiple candidate recommendation words corresponding to the current content are generated based on the content set and multiple recommendation word prompts. For example, the relevant information of the content set and multiple recommendation word prompts can be input into a pre-trained large language model to obtain multiple candidate recommendation words corresponding to the current content. This ensures that the generated candidate recommendation words are consistent with the current content in theme, and enhances the logic and naturalness of the candidate recommendation words.

[0061] Optionally, multiple contents belonging to the same content dimension as the current content can be obtained as related content. For example, each content can carry a content tag, such as food, car, travel, etc., to mark the content dimension corresponding to the content. Content tags can be selected by users when uploading content, or can be marked in the background according to content information.

[0062] Optionally, multiple contents belonging to the same cluster as the current content may be obtained as related contents. After the user uploads the content, the backend may classify the content into corresponding clusters according to the content information.

[0063] Among them, S301 to S302 can be performed in an offline stage to reduce online computing resources. When the user browses the current content, multiple candidate recommendation words corresponding to the current content generated in the offline stage can be directly called and subsequent steps can be performed.

[0064] S303, obtaining a predicted click rate value of each candidate recommendation word by the user to determine a target recommendation word from a plurality of candidate recommendation words, and generating a recommendation word card based on the target recommendation word to display on the first display page.

[0065] First, semantic features are extracted for each candidate recommendation word to obtain a first semantic vector corresponding to each candidate recommendation word.

[0066] Semantic feature extraction is performed on user information corresponding to the user (user's interest preference, user's social attributes such as age, gender, etc.) to obtain a second semantic vector corresponding to the user.

[0067] Semantic features are extracted for the current content to obtain a third semantic vector corresponding to the current content.

[0068] If the current content is text content, the current content is input into the semantic vector generation model to obtain a third semantic vector corresponding to the current content output by the semantic vector generation model.

[0069] If the current content is video content, obtain text information corresponding to the current content; input the text information into a semantic vector generation model to obtain a third semantic vector corresponding to the current content output by the semantic vector generation model.

[0070] For any candidate recommendation word, the click rate of the candidate recommendation word is predicted according to the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation word to obtain the click rate prediction value of the candidate recommendation word, and the candidate recommendation word with the largest click rate prediction value is used as the target recommendation word.

[0071] In this way, by extracting semantic features from candidate recommendation terms, user information, and current content, we can more accurately match user interests and improve the accuracy of recommendations. By predicting click-through rates, we can select the recommendation terms that are most likely to attract users and improve user experience.

[0072] S304 : In response to monitoring a confirmation operation of the user on the recommendation word card, recall a plurality of target recommendation contents from the candidate recommendation contents according to the target recommendation word.

[0073] If a confirmation operation of the user on the recommendation card is detected, the target recommendation and multiple candidate recommendation contents in the resource pool are input into the recall model, and the multiple candidate recommendation contents recalled by the recall model are obtained as the target recommendation contents.

[0074] S305: Generate a recommendation list according to the multiple target recommended contents, and display the target recommended contents on the second display page according to the order of the recommendation list.

[0075] After recalling the multiple target recommended contents, the multiple target recommended contents need to be finely sorted. First, semantic feature extraction is performed on the target recommended words to obtain a fourth semantic vector corresponding to the target recommended words, and semantic feature extraction is performed on each target recommended content to obtain a fifth semantic vector corresponding to each target recommended content.

[0076] If the current content is text content, the target recommended content is input into the semantic vector generation model to obtain a fifth semantic vector corresponding to the target recommended content output by the semantic vector generation model.

[0077] If the current content is video content, obtain text information corresponding to the target recommended content; input the text information into the semantic vector generation model to obtain a fifth semantic vector corresponding to the target recommended content output by the semantic vector generation model.

[0078] Then, the vector similarity / vector distance between the fourth semantic vector and each fifth semantic vector is calculated respectively; and the arrangement order of the plurality of target recommended contents is determined according to the vector similarity / vector distance to generate a recommendation list.

[0079] Exemplarily, the plurality of target recommended contents are sorted in descending order according to vector similarity to generate a recommendation list.

[0080] Exemplarily, a plurality of target recommended contents are sorted in ascending order according to the vector distance to generate a recommendation list.

[0081] Among them, the fourth semantic vector corresponding to each candidate recommendation word and the fifth semantic vector corresponding to each candidate recommendation content can be generated in advance in the offline stage and stored in the vector table. After the target recommendation word and target recommendation content are determined in the online stage, they are directly queried and called from the vector table.

[0082] In the training phase of the semantic vector generation model, training may be performed based on a sequence of click behaviors of sample users on displayed sample recommendation contents corresponding to sample recommendation terms.

[0083] The embodiment of the present application obtains multiple candidate recommendation terms for the content currently browsed by the user and predicts the click-through rate of each candidate recommendation term, thereby accurately selecting the most appropriate recommendation term based on the user's interests and behaviors, ensuring the relevance and attractiveness of the recommendation term, thereby improving the user's click-through rate and engagement; recalls multiple target recommendation contents based on the target recommendation term, and performs fine sorting to generate a recommendation list for display on the second display page, so that the user can browse more content related to the target recommendation term, reducing the user's search time cost and enhancing the user experience.

[0084] It is not difficult to understand that in order to make the recommendation card more pleasing to the eye, when generating candidate recommendations, it is generally necessary to obtain the preset format parameters corresponding to the recommendation card (such as font, color, layout and other elements); determine the recommendation constraint parameters (such as font size, word limit) according to the preset format parameters; and generate multiple candidate recommendations corresponding to the current content according to the recommendation constraint parameters and multiple recommendation prompts. This ensures that the recommendation card is visually consistent with the overall design and improves the user experience.

[0085] Among them, in this application, each content (article or video) corresponds to a unique content identifier, and multiple candidate recommendation words corresponding to each content can be set in advance at the present stage, and the mapping relationship between the identifier and the recommendation word can be generated and stored. In this way, when the user browses the content online, the target identifier corresponding to the current content can be obtained, and the mapping relationship between the identifier and the recommendation word can be obtained. According to the target identifier, the mapping relationship is queried to determine the multiple candidate recommendation words corresponding to the current content. In this way, the recommendation words can be quickly obtained when the user browses the content, without the need for real-time generation, thereby reducing the response time and the time the user spends waiting for the recommendation generation process, thereby providing a smoother user experience.

[0086] Figure 4 is a flowchart of an exemplary implementation of a content recommendation method shown in the present application, such as Figure 4 As shown, the content recommendation method includes the following steps:

[0087] Get multiple recommended prompt words corresponding to the current content ( Figure 4 Taking 3 recommendation prompt words as an example, it can be set according to actual conditions), obtain multiple related contents similar to the current content, merge the current content and the related content to generate a content set, and generate multiple candidate recommendation words corresponding to the current content according to the content set and the multiple recommendation prompt words.

[0088] Semantic features are extracted for each candidate recommendation word to obtain a first semantic vector corresponding to each candidate recommendation word.

[0089] Semantic feature extraction is performed on user information corresponding to the user (user's interest preference, user's social attributes such as age, gender, etc.) to obtain a second semantic vector corresponding to the user.

[0090] Semantic features are extracted for the current content to obtain a third semantic vector corresponding to the current content.

[0091] For any candidate recommendation word, the click rate of the candidate recommendation word is predicted according to the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation word to obtain the click rate prediction value of the candidate recommendation word, the candidate recommendation word with the largest click rate prediction value is taken as the target recommendation word, and a recommendation word card is generated based on the target recommendation word and displayed on the first display page.

[0092] If a confirmation operation of the user on the recommendation card is detected, multiple target recommendation contents are recalled from the candidate recommendation contents according to the target recommendation words.

[0093] Semantic features are extracted from the target recommendation word to obtain a fourth semantic vector.

[0094] Semantic feature extraction is performed on each target recommendation content to obtain multiple fifth semantic vectors.

[0095] The vector similarity / vector distance between the fourth semantic vector and each fifth semantic vector is calculated respectively.

[0096] The order of arrangement of multiple target recommendation contents is determined according to vector similarity / vector distance to generate a recommendation list.

[0097] The target recommended content is displayed on the second display page in the order of the recommendation list.

[0098] The embodiment of the present application obtains multiple candidate recommendation terms for the content currently browsed by the user and predicts the click-through rate of each candidate recommendation term, thereby accurately selecting the most appropriate recommendation term based on the user's interests and behaviors, ensuring the relevance and attractiveness of the recommendation term, thereby improving the user's click-through rate and engagement; recalls multiple target recommended contents based on the target recommendation term, and generates a recommendation list for display on the second display page, so that the user can browse more content related to the target recommendation term, reducing the user's search time cost and enhancing the user experience.

[0099] Figure 5 is a schematic diagram of a content recommendation device shown in the present application, such as Figure 5 As shown, the content recommendation device 500 includes an acquisition module 501, a first display module 502, a recall module 503 and a second display module 504, wherein:

[0100] The acquisition module 501 is used to acquire multiple candidate recommendation words corresponding to the current content browsed by the user.

[0101] The first display module 502 is used to obtain the predicted click rate value of each candidate recommendation word of the user to determine a target recommendation word from multiple candidate recommendation words, and generate a recommendation word card based on the target recommendation word to display on the first display page.

[0102] The recall module 503 is used to recall multiple target recommended contents from the candidate recommended contents according to the target recommended words in response to monitoring the user's confirmation operation on the recommended word card.

[0103] The second display module 504 is used to generate a recommendation list according to the multiple target recommended contents, and display the target recommended contents on the second display page according to the order of the recommendation list.

[0104] By acquiring multiple candidate recommendation terms for the content currently browsed by the user and predicting the click-through rate of each candidate recommendation term, the device can accurately select the most appropriate recommendation term according to the user's interests and behaviors, thereby ensuring the relevance and attractiveness of the recommendation term, thereby improving the user's click-through rate and engagement; based on the target recommendation term, multiple target recommendation contents are recalled, and a recommendation list is generated and displayed on the second display page, so that the user can browse more content related to the target recommendation term, reducing the user's search time cost and enhancing the user experience.

[0105] According to an embodiment of the present application, the acquisition module 501 is further used to: acquire multiple recommendation prompt words corresponding to the current content; and generate multiple candidate recommendation words corresponding to the current content according to the multiple recommendation prompt words.

[0106] According to an embodiment of the present application, the acquisition module 501 is further used to: acquire multiple related contents similar to the current content; merge the current content and the related contents to generate a content set; and generate multiple candidate recommendation words corresponding to the current content according to the content set and the multiple recommendation word prompts.

[0107] According to an embodiment of the present application, the acquisition module 501 is further used to: acquire multiple related contents belonging to the same content dimension as the current content; or acquire multiple related contents belonging to the same cluster as the current content.

[0108] According to an embodiment of the present application, the acquisition module 501 is further used to: acquire a target identifier corresponding to the current content; acquire a mapping relationship between the identifier and the recommendation word; and query the mapping relationship according to the target identifier to determine multiple candidate recommendation words corresponding to the current content.

[0109] According to one embodiment of the present application, the first display module 502 is also used to: perform semantic feature extraction on each candidate recommendation term to obtain multiple first semantic vectors; perform semantic feature extraction on user information corresponding to the user to obtain a second semantic vector; perform semantic feature extraction on the current content to obtain a third semantic vector; for any candidate recommendation term, perform click-through rate prediction on the candidate recommendation term based on the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation term to obtain a click-through rate prediction value of the candidate recommendation term.

[0110] According to one embodiment of the present application, the second display module 504 is also used to: perform semantic feature extraction on the target recommendation term to obtain a fourth semantic vector; perform semantic feature extraction on each target recommendation content to obtain multiple fifth semantic vectors; respectively calculate the vector similarity / vector distance between the fourth semantic vector and each fifth semantic vector; determine the arrangement order of multiple target recommendation contents according to the vector similarity / vector distance to generate a recommendation list.

[0111] According to one embodiment of the present application, in response to the current content being text content, the first display module 502 and the second display module 504 are also used to: for any content in the current content and the target recommended content, input any content into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0112] According to one embodiment of the present application, in response to the current content being video content, the first display module 502 and the second display module 504 are also used to: obtain text information corresponding to any content in the current content and the target recommended content; input the text information into the semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

[0113] According to an embodiment of the present application, the acquisition module 501 is further used to: acquire preset format parameters corresponding to the recommendation word card; determine the recommendation word constraint parameters according to the preset format parameters; and generate multiple candidate recommendation words corresponding to the current content according to the recommendation word constraint parameters and multiple recommendation word prompts.

[0114] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0115] Figure 6A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0116] like Figure 6 As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0117] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0118] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as the content recommendation method. For example, in some embodiments, the content recommendation method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the content recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform the content recommendation method in any other appropriate manner (e.g., by means of firmware).

[0119] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0120] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0121] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0122] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0123] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0124] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0125] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0126] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A content recommendation method, comprising: Obtain multiple candidate recommendation words corresponding to the current content browsed by the user; Obtaining a predicted click rate value of the user for each of the candidate recommendation words to determine a target recommendation word from the plurality of candidate recommendation words, and generating a recommendation word card based on the target recommendation word for display on a first display page; In response to monitoring a confirmation operation of the user on the recommendation word card, recalling a plurality of target recommendation contents from candidate recommendation contents according to the target recommendation word; A recommendation list is generated according to the plurality of target recommended contents, and the target recommended contents are displayed on a second display page in the order of the recommendation list.

2. The method according to claim 1, wherein: The step of obtaining multiple candidate recommendation words corresponding to the current content browsed by the user includes: Acquire multiple recommendation prompt words corresponding to the current content; A plurality of candidate recommendation words corresponding to the current content are generated according to the plurality of recommendation word prompt words.

3. The method according to claim 2, wherein: The step of generating a plurality of candidate recommendation words corresponding to the current content according to the plurality of recommendation word prompt words includes: Acquire multiple related contents similar to the current content; merging the current content and the related content to generate a content collection; A plurality of candidate recommendation words corresponding to the current content are generated according to the content set and the plurality of recommendation word prompt words.

4. The method according to claim 3, wherein: The acquiring of a plurality of related contents similar to the current content includes: Acquire multiple related contents belonging to the same content dimension as the current content; or, A plurality of related contents belonging to the same cluster as the current content is obtained.

5. The method according to claim 1, wherein: The step of obtaining multiple candidate recommendation words corresponding to the current content browsed by the user includes: Obtaining a target identifier corresponding to the current content; Obtaining a mapping relationship between the identifier and the recommendation; The mapping relationship is queried according to the target identifier to determine a plurality of candidate recommendation words corresponding to the current content.

6. The method according to any one of claims 1 to 5, wherein: The obtaining of the predicted click rate value of each candidate recommendation word by the user includes: Extracting semantic features from each of the candidate recommendation words to obtain a plurality of first semantic vectors; Performing semantic feature extraction on the user information corresponding to the user to obtain a second semantic vector; Performing semantic feature extraction on the current content to obtain a third semantic vector; For any of the candidate recommendation words, click rate prediction is performed on the candidate recommendation word according to the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation word to obtain a click rate prediction value of the candidate recommendation word.

7. The method according to claim 6, wherein: The generating a recommendation list according to the plurality of target recommended contents comprises: Extracting semantic features of the target recommendation word to obtain a fourth semantic vector; Performing semantic feature extraction on each of the target recommended contents to obtain a plurality of fifth semantic vectors; respectively calculating the vector similarity / vector distance between the fourth semantic vector and each of the fifth semantic vectors; The arrangement order of the plurality of target recommended contents is determined according to the vector similarity / vector distance to generate a recommendation list.

8. The method according to claim 7, wherein: In response to the current content being text content, extracting semantic features from the current content to obtain a third semantic vector, and extracting semantic features from each of the target recommended contents to obtain a plurality of fifth semantic vectors, include: For any one of the current content and the target recommended content, the any one of the content is input into a semantic vector generation model to obtain a semantic vector corresponding to the any one of the content output by the semantic vector generation model.

9. The method according to claim 7, wherein: In response to the current content being video content, extracting semantic features from the current content to obtain a third semantic vector, and extracting semantic features from each of the target recommended contents to obtain a plurality of fifth semantic vectors, including: For any content of the current content and the target recommended content, obtain text information corresponding to the any content; The text information is input into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

10. The method according to claim 3, wherein: The step of generating a plurality of candidate recommendation words corresponding to the current content according to the plurality of recommendation word prompt words includes: Obtaining preset format parameters corresponding to the recommendation card; Determining a recommendation word constraint parameter according to the preset format parameter; A plurality of candidate recommendation words corresponding to the current content are generated according to the recommendation word constraint parameters and the plurality of recommendation word prompt words.

11. A content recommendation device, comprising: An acquisition module is used to acquire multiple candidate recommendation words corresponding to the current content browsed by the user; A first display module, configured to obtain a predicted click rate value of each candidate recommendation word by the user to determine a target recommendation word from the plurality of candidate recommendation words, and generate a recommendation word card based on the target recommendation word to display on a first display page; a recall module, configured to recall a plurality of target recommended contents from the candidate recommended contents according to the target recommended words in response to monitoring the user's confirmation operation on the recommended word card; The second display module is used to generate a recommendation list according to the multiple target recommended contents, and display the target recommended contents on a second display page according to the order of the recommendation list.

12. The device according to claim 11, wherein The acquisition module is further used for: Acquire multiple recommendation prompt words corresponding to the current content; A plurality of candidate recommendation words corresponding to the current content are generated according to the plurality of recommendation word prompt words.

13. The device according to claim 12, wherein: The acquisition module is further used for: Acquire multiple related contents similar to the current content; merging the current content and the related content to generate a content collection; A plurality of candidate recommendation words corresponding to the current content are generated according to the content set and the plurality of recommendation word prompt words.

14. The device according to claim 13, wherein: The acquisition module is further used for: Acquire multiple related contents belonging to the same content dimension as the current content; or, A plurality of related contents belonging to the same cluster as the current content is obtained.

15. The device according to claim 11, wherein The acquisition module is further used for: Obtaining a target identifier corresponding to the current content; Obtaining a mapping relationship between the identifier and the recommendation; The mapping relationship is queried according to the target identifier to determine a plurality of candidate recommendation words corresponding to the current content.

16. The device according to any one of claims 11 to 15, wherein: The first display module is further used for: Extracting semantic features from each of the candidate recommendation words to obtain a plurality of first semantic vectors; Performing semantic feature extraction on the user information corresponding to the user to obtain a second semantic vector; Extracting semantic features of the current content to obtain a third semantic vector; For any of the candidate recommendation words, click rate prediction is performed on the candidate recommendation word according to the first semantic vector, the second semantic vector and the third semantic vector corresponding to the candidate recommendation word to obtain a click rate prediction value of the candidate recommendation word.

17. The device according to claim 16, wherein: The second display module is further used for: Extracting semantic features of the target recommendation word to obtain a fourth semantic vector; Performing semantic feature extraction on each of the target recommended contents to obtain a plurality of fifth semantic vectors; respectively calculating the vector similarity / vector distance between the fourth semantic vector and each of the fifth semantic vectors; The arrangement order of the plurality of target recommended contents is determined according to the vector similarity / vector distance to generate a recommendation list.

18. The device according to claim 17, wherein: In response to the current content being text content, the first display module and the second display module are further configured to: For any one of the current content and the target recommended content, the any one of the content is input into a semantic vector generation model to obtain a semantic vector corresponding to the any one of the content output by the semantic vector generation model.

19. The device according to claim 17, wherein: In response to the current content being video content, the first display module and the second display module are further configured to: For any content of the current content and the target recommended content, obtain text information corresponding to the any content; The text information is input into a semantic vector generation model to obtain a semantic vector corresponding to any content output by the semantic vector generation model.

20. The device according to claim 13, wherein: The acquisition module is further used for: Obtaining preset format parameters corresponding to the recommendation card; Determining a recommendation word constraint parameter according to the preset format parameter; A plurality of candidate recommendation words corresponding to the current content are generated according to the recommendation word constraint parameters and the plurality of recommendation word prompt words.

21. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-10.

23. A computer program product comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images