Content recommendation method and apparatus, computer device, and computer-readable storage medium
By performing attention feature mining and feature fusion processing on the content attribute information sequence, object feature information of the target object is generated, which solves the shortcomings of I2I and U2U recall in the existing technology and achieves more accurate content recommendation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAOHONGSHU TECH CO LTD
- Filing Date
- 2023-08-15
- Publication Date
- 2026-07-14
AI Technical Summary
In existing content recommendation algorithms, I2I recall cannot take into account user behavior features, resulting in low accuracy of recommendation results, while U2U recall performs poorly in recommending new content.
Based on multiple historical interaction contents of the target object, the content attribute information sequence under various attributes is obtained. Through attention feature mining and feature fusion processing, the object feature information of the target object is generated, and then content is recommended to the target object.
It improves the accuracy of content recommendation, enabling more precise recommendations of relevant and useful content to the target audience.
Smart Images
Figure CN117725299B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of communication technology, specifically to a content recommendation method, apparatus, computer device, and computer-readable storage medium. Background Technology
[0002] In content distribution scenarios, accurately recommending content to users can improve their content browsing experience. Current content recommendation algorithms mainly include recall and ranking. Recall can include I2I recall (recalling materials based on materials) and U2U recall (recalling users based on users). Among them, I2I only recommends content based on the similarity between materials, without considering the user's behavioral characteristics, and cannot accurately predict the content that users are interested in, resulting in low accuracy of recommendation results. U2U recommends based on the content that other users like, and its recommendation effect is poor for new content. Summary of the Invention
[0003] This application provides a content recommendation method, apparatus, computer device, and computer-readable storage medium, which can improve the accuracy of content recommendation.
[0004] This application provides a content recommendation method, including:
[0005] Based on multiple historical interaction contents of the target object, obtain content attribute information sequences under multiple attributes, each content attribute information sequence including attribute information of the multiple historical interaction contents under the same attribute;
[0006] The content attribute information sequence is subjected to attention feature mining processing to obtain the attention feature information sequence corresponding to each attribute;
[0007] The attention feature information sequence is subjected to a first feature fusion process to obtain a fused attention feature information sequence;
[0008] Based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence, generate the behavioral feature information sequence of the target object;
[0009] A second feature fusion process is performed on the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object;
[0010] Content is recommended to the target object based on the object's feature information.
[0011] Accordingly, this application also provides a content recommendation device, comprising:
[0012] The acquisition unit is used to acquire a sequence of content attribute information under multiple attributes based on multiple historical interaction contents of the target object. Each sequence of content attribute information includes the attribute information of the multiple historical interaction contents under the same attribute.
[0013] The mining unit is used to perform attention feature mining processing on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute.
[0014] The first fusion unit is used to perform a first feature fusion process on the attention feature information sequence to obtain a fused attention feature information sequence.
[0015] A sequence generation unit is used to generate a sequence of behavioral feature information of the target object based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence;
[0016] The second fusion unit is used to perform a second feature fusion process on the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object;
[0017] The recommendation unit is used to recommend content to the target object based on the object's feature information.
[0018] In one embodiment, the mining unit includes:
[0019] The extraction subunit is used to extract attention features from the content attribute information sequence under each attribute, so as to obtain the attribute feature information of each historical interaction content on the attribute, and the correlation between the historical interaction content and the associated historical interaction content with respect to the attribute.
[0020] The weighted sub-unit is used to weight the attribute feature information of each historical interaction content according to the similarity between each historical interaction content and the associated historical interaction content for each attribute, so as to obtain the attention feature information sequence corresponding to each attribute.
[0021] In one embodiment, the second fusion unit includes:
[0022] The calculation subunit is used to calculate the similarity between every two historical interaction contents based on the fused attention feature information corresponding to each historical interaction content in the fused attention feature information sequence.
[0023] The weighted fusion subunit is used to perform weighted fusion processing on the behavioral feature information sequence according to the similarity to obtain the object feature information of the target object.
[0024] In one embodiment, the mining unit is further configured to:
[0025] The attention feature mining process is performed on the content attribute information sequence by the trained feature extraction model to obtain the attention feature information sequence corresponding to each attribute.
[0026] In one embodiment, the content recommendation device further includes:
[0027] The information acquisition unit is used to acquire a sequence of content attribute information under multiple attributes based on multiple historical interaction contents of an object sample, and to acquire content samples pushed to the object sample, wherein the content samples have corresponding sample tags;
[0028] The feature extraction unit is used to extract features from the content attribute information of multiple historical interaction contents of the object sample with respect to different attributes through a feature extraction model, so as to obtain the object feature information of the object sample.
[0029] A prediction unit is used to predict the predicted label of the content sample based on the object feature information of the object sample;
[0030] The training unit is used to train the feature extraction model based on the predicted label and the sample label to obtain the trained feature extraction model.
[0031] In one embodiment, the information acquisition unit includes:
[0032] The time acquisition subunit is used to acquire the content sample pushed to the object sample within a preset time and the display time of the content sample during the push process;
[0033] The tag generation subunit is used to generate sample tags for the content sample based on the display time.
[0034] In one embodiment, the recommendation unit includes:
[0035] The content acquisition subunit is used to acquire candidate recommended content;
[0036] The content feature extraction subunit is used to extract features from the candidate recommended content to obtain the content feature information of the candidate recommended content;
[0037] The score calculation subunit is used to calculate the recommendation score of the candidate recommendation content based on the object feature information of the target object and the content feature information of the candidate recommendation content.
[0038] The recommendation subunit is used to recommend content to the target object based on the recommendation score and the candidate recommendation content.
[0039] Accordingly, this application also provides a computer device including a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to execute any of the content recommendation methods provided in this application.
[0040] Accordingly, embodiments of this application also provide a computer-readable storage medium for storing a computer program, which is loaded by a processor to execute any of the content recommendation methods provided in embodiments of this application.
[0041] This application embodiment obtains content attribute information sequences under multiple attributes based on multiple historical interaction contents of the target object. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute. Attention feature mining processing is performed on the content attribute information sequences to obtain attention feature information sequences corresponding to each attribute. A first feature fusion processing is performed on the attention feature information sequences to obtain fused attention feature information sequences. Based on the target content attribute information sequences belonging to the content identifier attribute in the content attribute information sequences, a behavioral feature information sequence of the target object is generated. A second feature fusion processing is performed on the fused attention feature information sequence and the behavioral feature information sequence to obtain object feature information of the target object. Content is recommended to the target object based on the object feature information.
[0042] This application embodiment performs attention feature mining processing on the content attribute information sequence corresponding to each attribute, which can obtain an attention feature information sequence that focuses on the information most relevant to the target object in each attribute and is most useful for recommending content to the target object. This makes the object feature information obtained based on the attention feature information sequence highly expressive, and can accurately recommend content to the target object based on the object feature information, thereby improving the content recommendation effect. Attached Figure Description
[0043] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0044] Figure 1 This is a flowchart of the content recommendation method provided in the embodiments of this application;
[0045] Figure 2 This is a schematic diagram of the model structure provided in the embodiments of this application;
[0046] Figure 3 This is a schematic diagram of the content recommendation device provided in the embodiments of this application;
[0047] Figure 4 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation
[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0049] This application provides a content recommendation method, apparatus, computer device, and computer-readable storage medium. The content recommendation apparatus can be integrated into a computer device, which may be a server or a terminal, etc.
[0050] The terminal may include mobile phones, wearable smart devices, tablets, laptops, personal computers (PCs), and in-vehicle computers, etc.
[0051] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.
[0052] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.
[0053] This embodiment will be described from the perspective of a content recommendation device, which can be integrated into a computer device, such as a server or a terminal.
[0054] This application provides a content recommendation method, such as... Figure 1 As shown, the specific process of this content recommendation method can be summarized as follows:
[0055] 101. Based on multiple historical interaction contents of the target object, obtain content attribute information sequences under multiple attributes. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute.
[0056] The content can include media content such as articles, pictures, audio and video, and can also include commercial media content such as advertisements and products; the attributes of the content can be items used to describe the characteristics or features of the content, and the attribute information corresponding to the attribute can be considered as the value of the content under the attribute. The content attribute information sequence can include the attribute information of multiple historical interactive content under the same attribute.
[0057] For example, if the topic of content A is "summer," then the topic can be an attribute, and "summer" is the attribute information under the topic attribute. The sequence of content attribute information under the topic attribute can include the topic content of content A, content B, content C, content D, and content D, etc.
[0058] The target audience can be users or businesses that need to be recommended content to, and the historical interaction content can include content that the target audience has interacted with, such as content that has been saved, liked, viewed, or commented on.
[0059] For example, in the process of pushing content to users, respond to users' interactive operations on the content and obtain multiple historical interaction contents.
[0060] Optionally, when a request to push content to the target object is received, the target object's recent historical interaction content can be obtained. For example, the 50 most recently interacted content items of the target object can be obtained to get multiple historical interaction content items.
[0061] Various attributes can include content identifiers (ID), keywords (which can be extracted from keywords), topics, entities (such as locations, games, and products in the content), categories (such as sports, entertainment, music, and health), and author attributes.
[0062] For each attribute, obtain the attribute information of each historical interaction content under that attribute, and concatenate the attribute information to obtain the content attribute information sequence under that attribute.
[0063] 102. Perform attention feature mining on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute.
[0064] For example, attention feature mining can be performed on the content attribute information sequence corresponding to each attribute. Specifically, for each attribute, since the content attribute information sequence contains the attribute information of each historical interaction content on that attribute, the attribute information of each historical interaction content can be mapped to the same feature space. Based on the distance of the attribute information in the feature space, multiple historical interaction contents are clustered to obtain at least one cluster. The number of historical interaction contents contained in each cluster is obtained. Based on the proportion of historical interaction contents in the cluster to the number of multiple historical interaction contents interacted with by the target object, the weight of each historical interaction content is obtained. Then, the historical interaction contents are weighted according to the weight to obtain the attention feature information sequence under that attribute. Similarly, the attention feature information sequences of other attributes can be obtained.
[0065] The weight of each historical interaction is obtained by taking the proportion of the historical interaction content corresponding to the cluster relative to the number of historical interaction content that the target object has interacted with. Alternatively, the weight of each historical interaction content in the cluster can be taken as the proportion of the historical interaction content in the cluster relative to the number of historical interaction content that the target object has interacted with.
[0066] Optionally, based on the distance between the historical interaction content corresponding to the cluster center of the cluster and other historical interaction content in the cluster, the similarity between the historical interaction content of the cluster center and other historical interaction content in the cluster is calculated. Then, the proportion of the number of the cluster is multiplied by each historical interaction content to obtain the weight of each historical interaction content.
[0067] Optionally, attention feature information sequences for each attribute can be obtained through attention feature mining. Specifically, in one embodiment, the step "perform attention feature mining on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute" may include:
[0068] For each attribute, attention features are extracted from the content attribute information sequence under the attribute to obtain the attribute feature information of each historical interaction content on the attribute, as well as the correlation between the historical interaction content and the attribute.
[0069] For each attribute, the attribute feature information of each historical interaction is weighted based on the similarity between each historical interaction and its associated historical interaction, resulting in a sequence of attention feature information corresponding to each attribute.
[0070] The associated historical interaction content can include each historical interaction content.
[0071] For example, specifically, for each attribute, attention features are extracted from the content attribute information sequence to obtain the query feature information (Q), key feature information (K), and value feature information (V) of each historical interaction content under that attribute. The value feature vector is the attribute feature information of the historical interaction content.
[0072] The query feature information of the target historical interaction content (one of multiple historical interaction content) is multiplied by the key feature information of each of the multiple historical interaction content to obtain the correlation between the target historical interaction content and each of the historical interaction content. The correlation can be similarity. The attribute feature information of each historical interaction content is then weighted according to the correlation between each historical interaction content and the target historical interaction content to obtain the attention feature information of the target historical interaction content. Optionally, the correlation can be normalized before weighting.
[0073] It is understandable that the target historical interaction content is any one of multiple historical interaction content. By taking each historical interaction content as the target historical interaction content, we can obtain the attention feature information of each historical interaction content under this attribute, and then obtain the attention feature information sequence.
[0074] Optionally, the content attribute information sequence can be input into an attention mechanism network. The attention mechanism network extracts attention features from the content attribute information sequence and performs weighted processing.
[0075] In one embodiment, the object feature information of the target object can be obtained by extracting it through a post-trained feature extraction model. The attention mechanism network can be the network in the post-trained feature extraction model. That is, in one embodiment, the step "perform attention feature mining processing on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute" can specifically include:
[0076] The attention feature mining process is performed on the content attribute information sequence by the trained feature extraction model to obtain the attention feature information sequence corresponding to each attribute.
[0077] Post-trained feature extraction models can include, for example: Figure 2 The model structure shown can be described as follows: the feature extraction model can include an attention mechanism layer, a first fusion processing layer, and a second fusion processing layer. The attention mechanism layer can include multiple attention mechanism networks. The first fusion processing layer is used to perform a first fusion processing on the attention feature information sequence. The second fusion processing layer is used to perform a second fusion processing on the fused attention feature information sequence and the object feature information.
[0078] The feature extraction model can be pre-trained to accurately extract the user's object feature information, thereby accurately recommending content to the user and improving the user's efficiency in obtaining information. Specifically, in one embodiment, before the step of "obtaining a sequence of content attribute information under multiple attributes based on multiple historical interaction contents of the target object", the content recommendation method provided in this application embodiment may further include:
[0079] Based on multiple historical interaction contents of object samples, obtain content attribute information sequences under various attributes, and obtain content samples pushed to object samples, with each content sample having a corresponding sample tag;
[0080] By using a feature extraction model, feature information about the content attributes of multiple historical interaction contents of an object sample with respect to different attributes is extracted, thus obtaining the object feature information of the object sample.
[0081] Predict the predicted label of the content sample based on the object feature information of the object sample;
[0082] The feature extraction model is trained based on the predicted labels and sample labels to obtain the trained feature extraction model.
[0083] Among them, the object sample can be a sample used to train the feature extraction model, or an object that meets preset conditions, such as an object whose browsing time meets a preset time.
[0084] The content sample can be content pushed to the target object's terminal and displayed to the target object through the terminal. The sample tag can mark the content sample as a positive sample or a negative sample. A positive sample is content that the object likes, and a negative sample is content that the object dislikes. The content sample can be marked according to the user's interaction behavior, and it can also be marked according to the exposure time (display time) of the content sample.
[0085] For example, it could be based on multiple historical interaction contents of an object sample, obtaining a sequence of content attribute information under various attributes, and obtaining content samples pushed to the object sample.
[0086] The feature extraction model extracts the content attribute information of multiple historical interaction contents of the object sample with respect to different attributes, and obtains the object feature information of the object sample. For the specific process, please refer to steps 102-104, which will not be elaborated here.
[0087] Feature extraction of content samples can be performed to obtain content feature information. This can be done by extracting attribute information corresponding to different attributes from the content sample, and then performing feature extraction on the attribute information to obtain content feature information. Attributes can include content feature attributes such as content identifier (ID), keywords, topics, entities, categories, and authors, as well as statistical feature attributes such as click-through rate and exposure duration.
[0088] The predicted label of the content sample is predicted based on the object feature information of the object sample and the content feature information of the content sample. For example, the object sample and the content sample can be mapped to the same feature space based on the object feature information and the content sample. The predicted label of the content sample is determined based on the distance between the object sample and the content sample in the feature space. For example, similarity can be calculated based on the distance and the similarity can be used as the predicted label. Alternatively, if the distance is less than a preset distance threshold, the predicted label of the content sample is a label indicating that the content sample is a positive sample. Otherwise, the predicted label is a label indicating that the content sample is a negative sample.
[0089] The feature extraction model is trained based on the predicted labels and sample labels to obtain the trained feature extraction model.
[0090] The sample tags can be automatically generated based on the display time of the content sample. In one embodiment, the step "obtaining the content sample pushed to the object sample" may specifically include:
[0091] Get the content sample pushed to the object sample within a preset time and the display time of the content sample during the push process;
[0092] Generate sample tags for the content samples based on the display time.
[0093] For example, the feature extraction model can be trained on a daily or hourly basis. On a daily basis, the content samples pushed to the target sample in the past day can be obtained, and the exposure duration of the content samples can be obtained. Content samples with an exposure duration exceeding a preset time threshold (e.g., 10 seconds) are taken as positive samples, and sample labels corresponding to the positive samples are generated. Content samples with an exposure duration less than the preset time threshold are taken as negative samples, and sample labels corresponding to the negative samples are generated.
[0094] 103. Perform first feature fusion processing on the attention feature information sequence to obtain the fused attention feature information sequence.
[0095] For example, it can be done by fusing attention feature information sequences corresponding to different attributes to obtain a fused attention feature sequence that represents multiple historical interaction contents.
[0096] The first feature fusion process can be to fuse multiple attention feature information sequences by using concatenate (also known as concat) fusion.
[0097] This application embodiment performs attention feature mining processing on the content attribute information sequence corresponding to each attribute, which can obtain an attention feature information sequence that focuses on the information most relevant to the target object in each attribute and is most useful for recommending content to the target object. Then, the attention feature information sequences are fused. The fused attention feature information sequence contains more information. The object feature information obtained based on the fused attention feature information sequence has strong expressive power. Therefore, based on the object feature information, content can be accurately recommended to the target object, improving the content recommendation effect.
[0098] 104. Generate a sequence of behavioral feature information of the target object based on the target content attribute information sequence that belongs to the content identifier attribute.
[0099] Among them, the content identification attribute can be information such as ID that can uniquely represent the content.
[0100] For example, the sequence of target content attribute information on the content identifier attribute of historical interaction content can be used as the sequence of behavioral feature information of the target object.
[0101] Alternatively, the sequence of value features obtained by attention feature mining of the target content attribute information sequence can be used as the sequence of behavioral feature information of the target object.
[0102] 105. Perform a second feature fusion process on the fused attention feature information sequence and behavior feature information sequence to obtain the object feature information of the target object.
[0103] For example, the second feature fusion process can be achieved by using the concatenate (also known as concat) function to fuse the fused attention feature information sequence and the behavior feature information sequence; it can also be achieved by adding the fused attention feature information sequence and the behavior feature information sequence to integrate the features; or it can be achieved by using the dot product algorithm to fuse the fused attention feature information sequence and the behavior feature information sequence.
[0104] Optionally, the similarity between every two historical interaction contents can be calculated using the fused attention feature information sequence, and the behavioral feature information sequence can be weighted and fused based on the similarity to obtain the object feature information of the target object. That is, in one embodiment, the step "performing a second feature fusion process on the fused attention feature information sequence and the behavioral feature information sequence to obtain the object feature information of the target object" includes:
[0105] Based on the fused attention feature information of each historical interaction content in the fused attention feature information sequence, calculate the similarity between each pair of historical interaction content;
[0106] The behavioral feature information sequences are weighted and fused based on similarity to obtain the object feature information of the target object.
[0107] For example, the similarity between any two historical interaction contents can be calculated based on the distance in the feature space between the fused attention feature information corresponding to each historical interaction content in the fused attention feature information sequence.
[0108] For each historical interaction, the behavioral feature information sequence is weighted according to its similarity to other historical interactions to obtain the weighted behavioral feature information sequence corresponding to that historical interaction. Similarly, the weighted behavioral feature information sequence corresponding to each historical interaction can be obtained.
[0109] The object feature information of the target object is obtained by weighting the behavioral feature information sequence corresponding to multiple historical interaction contents.
[0110] 106. Recommend content to target objects based on object feature information.
[0111] For example, based on object feature information, objects similar to the target object can be identified, and content that the target object has interacted with can be recommended to the target object.
[0112] Optionally, content can be recommended to the target object based on the object feature information of the target object and the content feature information of the candidate recommended content. That is, in one embodiment, the step "recommend content to the target object based on object feature information" may specifically include:
[0113] Obtain candidate recommendation content;
[0114] Feature extraction is performed on the candidate recommendation content to obtain the content feature information of the candidate recommendation content;
[0115] Based on the object feature information of the target object and the content feature information of the candidate recommended content, the recommendation score of the candidate recommended content is calculated.
[0116] Content is recommended to the target audience based on the recommendation score and candidate recommendations.
[0117] The candidate recommendations can be content stored in a database.
[0118] For example, it could involve acquiring candidate recommendation content, extracting features from the candidate recommendation content to obtain content feature information, determining the recommendation score of the content to be recommended based on the distance between the content feature information and the object feature information in the feature space, and then determining the content to be recommended to the target object based on the recommendation score and recommending it to the target object.
[0119] As can be seen from the above, the embodiments of this application obtain content attribute information sequences under multiple attributes based on multiple historical interaction contents of the target object. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute. Attention feature mining processing is performed on the content attribute information sequences to obtain attention feature information sequences corresponding to each attribute. A first feature fusion processing is performed on the attention feature information sequences to obtain a fused attention feature information sequence. Based on the target content attribute information sequence belonging to the content identifier attribute in the content attribute information sequence, a behavioral feature information sequence of the target object is generated. A second feature fusion processing is performed on the fused attention feature information sequence and the behavioral feature information sequence to obtain the object feature information of the target object. Content is recommended to the target object based on the object feature information.
[0120] This application embodiment performs attention feature mining processing on the content attribute information sequence corresponding to each attribute, which can obtain an attention feature information sequence that focuses on the information most relevant to the target object in each attribute and is most useful for recommending content to the target object. This makes the object feature information obtained based on the attention feature information sequence highly expressive, and can accurately recommend content to the target object based on the object feature information, thereby improving the content recommendation effect.
[0121] To facilitate better implementation of the content recommendation method provided in the embodiments of this application, a content recommendation apparatus is also provided in one embodiment. The meanings of the terms used are the same as in the content recommendation method described above, and specific implementation details can be found in the description of the method embodiments.
[0122] This content recommendation device can be integrated into computer devices, such as... Figure 3 As shown, the content recommendation device may include: an acquisition unit 301, a mining unit 302, a first fusion unit 303, a sequence generation unit 304, a second fusion unit 305, and a recommendation unit 306, as detailed below:
[0123] (1) Acquisition unit 301 is used to acquire content attribute information sequence under multiple attributes based on multiple historical interaction contents of the target object. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute.
[0124] (2) Mining unit 302 is used to perform attention feature mining on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute.
[0125] In one embodiment, the mining unit 302 may include an extraction subunit and a weighting subunit, specifically:
[0126] Extract sub-units, which are used to extract attention features from the content attribute information sequence under each attribute for each attribute, to obtain the attribute feature information of each historical interaction content on the attribute, as well as the correlation between the attribute and the associated historical interaction content;
[0127] The weighted sub-unit is used to weight the attribute feature information of each historical interaction content according to the similarity between each historical interaction content and the associated historical interaction content for each attribute, so as to obtain the attention feature information sequence corresponding to each attribute.
[0128] In one embodiment, the mining unit 302 is further configured to:
[0129] The attention feature mining process is performed on the content attribute information sequence by the trained feature extraction model to obtain the attention feature information sequence corresponding to each attribute.
[0130] In one embodiment, the content recommendation device may further include an information acquisition unit, a feature extraction unit, a prediction unit, and a training unit, specifically:
[0131] The information acquisition unit is used to acquire content attribute information sequences under multiple attributes based on multiple historical interaction contents of object samples, as well as to acquire content samples pushed to object samples, and the content samples have corresponding sample tags.
[0132] The feature extraction unit is used to extract the content attribute information of multiple historical interaction contents of the object sample with respect to different attributes through the feature extraction model, so as to obtain the object feature information of the object sample.
[0133] The prediction unit is used to predict the predicted label of the content sample based on the object feature information of the object sample.
[0134] The training unit is used to train the feature extraction model based on the predicted labels and sample labels, resulting in a trained feature extraction model.
[0135] (3) The first fusion unit 303 is used to perform first feature fusion processing on the attention feature information sequence to obtain the fused attention feature information sequence.
[0136] (4) Sequence generation unit 304 is used to generate a sequence of behavioral feature information of the target object based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence.
[0137] (5) The second fusion unit 305 is used to perform second feature fusion processing on the fused attention feature information sequence and behavior feature information sequence to obtain the object feature information of the target object.
[0138] In one embodiment, the second fusion unit 305 may include a calculation subunit and a weighted fusion subunit, specifically:
[0139] The computational subunit is used to calculate the similarity between every two historical interaction contents based on the fused attention feature information corresponding to each historical interaction content in the fused attention feature information sequence.
[0140] The weighted fusion subunit is used to perform weighted fusion processing on the behavioral feature information sequence based on similarity to obtain the object feature information of the target object.
[0141] In one embodiment, the information acquisition unit may include a time acquisition subunit and a tag generation subunit, specifically:
[0142] The time acquisition subunit is used to acquire the content sample pushed to the object sample within a preset time and the display time of the content sample during the push process;
[0143] The tag generation sub-unit is used to generate sample tags for content samples based on the display time.
[0144] (6) Recommendation unit 306 is used to recommend content to the target object based on object feature information.
[0145] In one embodiment, the recommendation unit 306 may include a content acquisition subunit, a content feature extraction subunit, a score calculation subunit, and a recommendation subunit, specifically:
[0146] The content acquisition subunit is used to acquire candidate recommended content;
[0147] The content feature extraction subunit is used to extract features from candidate recommended content to obtain content feature information of candidate recommended content;
[0148] The score calculation subunit is used to calculate the recommendation score of the candidate recommendation content based on the object feature information of the target object and the content feature information of the candidate recommendation content.
[0149] The recommendation subunit is used to recommend content to the target audience based on the recommendation score and candidate recommendations.
[0150] As can be seen from the above, the content recommendation device in this application embodiment obtains content attribute information sequences under multiple attributes based on multiple historical interaction contents of the target object by the acquisition unit 301. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute. The mining unit 302 performs attention feature mining processing on the content attribute information sequence to obtain attention feature information sequences corresponding to each attribute. The first fusion unit 303 performs first feature fusion processing on the attention feature information sequence to obtain a fused attention feature information sequence. The sequence generation unit 304 generates a behavior feature information sequence of the target object based on the target content attribute information sequence belonging to the content identifier attribute in the content attribute information sequence. The second fusion unit 305 performs second feature fusion processing on the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object. The recommendation unit 306 recommends content to the target object based on the object feature information.
[0151] This application embodiment performs attention feature mining processing on the content attribute information sequence corresponding to each attribute, which can obtain an attention feature information sequence that focuses on the information most relevant to the target object in each attribute and is most useful for recommending content to the target object. This makes the object feature information obtained based on the attention feature information sequence highly expressive, and can accurately recommend content to the target object based on the object feature information, thereby improving the content recommendation effect.
[0152] This application also provides a computer device, which can be a terminal or a server, such as... Figure 4 As shown, it illustrates a structural schematic diagram of the computer device involved in the embodiments of this application, specifically:
[0153] The computer device may include components such as a processor 1001 with one or more processing cores, a memory 1002 with one or more computer-readable storage media, a power supply 1003, and an input unit 1004. Those skilled in the art will understand that... Figure 4 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:
[0154] The processor 1001 is the control center of the computer device. It connects various parts of the computer device via various interfaces and lines, and performs various functions and processes data by running or executing software programs and / or modules stored in the memory 1002, and by calling data stored in the memory 1002, thereby providing overall monitoring of the computer device. Optionally, the processor 1001 may include one or more processing cores; preferably, the processor 1001 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and computer programs, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1001.
[0155] The memory 1002 can be used to store software programs and modules. The processor 1001 executes various functional applications and data processing by running the software programs and modules stored in the memory 1002. The memory 1002 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, computer programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 1002 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 1002 may also include a memory controller to provide the processor 1001 with access to the memory 1002.
[0156] The computer equipment also includes a power supply 1003 that supplies power to the various components. Preferably, the power supply 1003 can be logically connected to the processor 1001 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 1003 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.
[0157] The computer device may also include an input unit 1004, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
[0158] Although not shown, the computer device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 1001 in the computer device loads the executable files corresponding to the processes of one or more computer programs into the memory 1002 according to the following instructions, and the processor 1001 runs the computer programs stored in the memory 1002 to realize various functions, as follows:
[0159] Based on multiple historical interaction contents of the target object, obtain content attribute information sequences under multiple attributes. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute.
[0160] Attention feature mining is performed on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute.
[0161] The attention feature information sequence is subjected to the first feature fusion process to obtain the fused attention feature information sequence.
[0162] Based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence, generate the behavioral feature information sequence of the target object;
[0163] A second feature fusion process is performed on the fused attention feature information sequence and behavior feature information sequence to obtain the object feature information of the target object; content is recommended to the target object based on the object feature information.
[0164] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0165] As can be seen from the above, the computer device in this embodiment can obtain content attribute information sequences under multiple attributes based on multiple historical interaction contents of a target object. Each content attribute information sequence includes attribute information of multiple historical interaction contents under the same attribute. Attention feature mining processing is performed on the content attribute information sequences to obtain attention feature information sequences corresponding to each attribute. A first feature fusion processing is performed on the attention feature information sequences to obtain a fused attention feature information sequence. A behavioral feature information sequence of the target object is generated based on the target content attribute information sequence belonging to the content identifier attribute in the content attribute information sequence. A second feature fusion processing is performed on the fused attention feature information sequence and the behavioral feature information sequence to obtain object feature information of the target object. Content is recommended to the target object based on the object feature information.
[0166] This application embodiment performs attention feature mining processing on the content attribute information sequence corresponding to each attribute, which can obtain an attention feature information sequence that focuses on the information most relevant to the target object in each attribute and is most useful for recommending content to the target object. This makes the object feature information obtained based on the attention feature information sequence highly expressive, and can accurately recommend content to the target object based on the object feature information, thereby improving the content recommendation effect.
[0167] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations of the above embodiments.
[0168] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a computer program, or by a computer program controlling related hardware. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0169] Therefore, embodiments of this application provide a computer-readable storage medium storing a computer program that can be loaded by a processor to execute any of the content recommendation methods provided in embodiments of this application.
[0170] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.
[0171] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0172] Since the computer program stored in the computer-readable storage medium can execute any of the content recommendation methods provided in the embodiments of this application, it can achieve the beneficial effects that any of the content recommendation methods provided in the embodiments of this application can achieve, as detailed in the preceding embodiments, and will not be repeated here.
[0173] The foregoing has provided a detailed description of a content recommendation method, apparatus, computer device, and computer-readable storage medium provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A content recommendation method, characterized in that, include: Based on multiple historical interaction contents of the target object, obtain content attribute information sequences under multiple attributes. Each content attribute information sequence includes attribute information of the multiple historical interaction contents under the same attribute. The content includes at least one of articles, pictures, audio, video, advertisements and products. The content attribute information sequence is subjected to attention feature mining processing to obtain the attention feature information sequence corresponding to each attribute; The attention feature information sequence is subjected to a first feature fusion process to obtain a fused attention feature information sequence; Based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence, generate the behavioral feature information sequence of the target object; A second feature fusion process is performed on the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object; Content is recommended to the target object based on the object's feature information.
2. The method according to claim 1, characterized in that, The step of performing attention feature mining on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute includes: For each attribute, attention features are extracted from the content attribute information sequence under the attribute to obtain the attribute feature information of each historical interaction content on the attribute, as well as the correlation between the historical interaction content and the attribute. For each attribute, the attribute feature information of each historical interaction is weighted based on the similarity between each historical interaction and its associated historical interaction, resulting in a sequence of attention feature information corresponding to each attribute.
3. The method according to claim 1, characterized in that, The second feature fusion processing of the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object includes: Based on the fused attention feature information of each historical interaction content in the fused attention feature information sequence, the similarity between each two historical interaction contents is calculated; The behavioral feature information sequence is weighted and fused based on the similarity to obtain the object feature information of the target object.
4. The method according to claim 1, characterized in that, The step of performing attention feature mining on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute includes: The attention feature mining process is performed on the content attribute information sequence by the trained feature extraction model to obtain the attention feature information sequence corresponding to each attribute.
5. The method according to claim 4, characterized in that, Before obtaining the content attribute information sequence under multiple attributes based on multiple historical interaction contents of the target object, the method further includes: Based on multiple historical interaction contents of object samples, obtain content attribute information sequences under various attributes, and obtain content samples pushed to the object samples, wherein the content samples correspond to sample tags; The object feature information of the object sample is obtained by extracting the content attribute information of multiple historical interaction contents of the object sample with respect to different attributes through a feature extraction model. Predict the predicted label of the content sample based on the object feature information of the object sample; The feature extraction model is trained based on the predicted label and the sample label to obtain the trained feature extraction model.
6. The method according to claim 5, characterized in that, The step of obtaining the content sample pushed to the object sample includes: Obtain the content samples pushed to the object sample within a preset time period and the display time of the content samples during the push process; Based on the display time, generate sample tags for the content sample.
7. The method according to any one of claims 1-6, characterized in that, The content recommendation to the target object based on the object feature information includes: Obtain candidate recommendation content; Feature extraction is performed on the candidate recommendation content to obtain the content feature information of the candidate recommendation content; Based on the object feature information of the target object and the content feature information of the candidate recommended content, the recommendation score of the candidate recommended content is calculated; Content is recommended to the target object based on the recommendation score and the candidate recommendation content.
8. A content recommendation device, characterized in that, include: The acquisition unit is used to acquire a sequence of content attribute information under multiple attributes based on multiple historical interaction contents of the target object. Each sequence of content attribute information includes the attribute information of the multiple historical interaction contents under the same attribute. The content includes at least one of articles, pictures, audio, video, advertisements and products. The mining unit is used to perform attention feature mining processing on the content attribute information sequence to obtain the attention feature information sequence corresponding to each attribute. The first fusion unit is used to perform a first feature fusion process on the attention feature information sequence to obtain a fused attention feature information sequence. A sequence generation unit is used to generate a sequence of behavioral feature information of the target object based on the target content attribute information sequence that belongs to the content identifier attribute in the content attribute information sequence; The second fusion unit is used to perform a second feature fusion process on the fused attention feature information sequence and the behavior feature information sequence to obtain the object feature information of the target object; The recommendation unit is used to recommend content to the target object based on the object's feature information.
9. A computer device, characterized in that, It includes a memory and a processor; the memory stores a computer program, and the processor is used to run the computer program in the memory to perform the content recommendation method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which is loaded by a processor to perform the content recommendation method according to any one of claims 1 to 7.