Multi-scenario content recommendation methods, devices, electronic devices, and storage media

By acquiring user profiles and scene sequence data, and utilizing multi-head attention mechanisms and fully connected neural networks to extract feature vectors, the accuracy problem of content recommendation in different scenarios was solved, achieving accurate recommendations across scenarios, reducing system resource consumption and R&D costs, and improving user experience.

CN117633360BActive Publication Date: 2026-11-17BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311684700.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-08
Publication Date
2026-11-17
Estimated Expiration
2043-12-08

AI Technical Summary

Technical Problem

When recommending content to users in different scenarios, existing technologies ignore the general and hidden information of the scenarios, resulting in low accuracy of recommendation models. In addition, each scenario requires independent model maintenance, which increases the consumption of system resources and R&D costs.

Method used

By acquiring user profile data and scene sequence data, and utilizing multi-head attention mechanisms and fully connected neural networks, feature vectors are extracted for different scenarios. Combined with user behavior data in multiple scenarios, recommended content is determined to achieve cross-scenario content recommendation.

Benefits of technology

It improves the accuracy of content recommendations, meets users' real needs in different scenarios, reduces system resource consumption and R&D costs, and enhances user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117633360B_ABST
    Figure CN117633360B_ABST
Patent Text Reader

Abstract

This disclosure provides a content recommendation method, apparatus, electronic device, and storage medium based on multiple scenarios, relating to the field of artificial intelligence, and particularly to the field of big data. The specific implementation scheme is as follows: User profile data and scenario sequence data are acquired; wherein the profile data represents the user's historical behavior, and the scenario sequence data represents multiple scenarios experienced by the user within a preset time period; the historical behavior represents the user's actions on previously recommended content within those scenarios; based on the profile data and the scenario sequence data, recommended content is determined from a pre-selected pool of content, and the recommended content is recommended to the user. By combining profile data from different scenarios, the content to be recommended to the user at any given time is determined, improving the accuracy of content recommendation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of big data in the field of artificial intelligence, and in particular to a content recommendation method, device, electronic device and storage medium based on multiple scenarios. Background Technology

[0002] With the rapid development of the mobile internet, artificial intelligence products need to meet users' refined needs in different scenarios and recommend different content to them. However, users and their needs vary greatly in different scenarios, and the content recommended to users also differs.

[0003] If the content recommended to users is not of interest to them, it will directly affect their experience of using AI products. Therefore, providing accurate content recommendations is a significant challenge we currently face. Summary of the Invention

[0004] This disclosure provides a content recommendation method, apparatus, electronic device, and storage medium based on multiple scenarios.

[0005] According to a first aspect of this disclosure, a multi-scenario content recommendation method is provided, including:

[0006] The system acquires user profile data and scene sequence data; wherein the profile data represents the user's historical behavior, and the scene sequence data represents multiple scenes experienced by the user within a preset time period; the historical behavior represents the user's actions on recommended historical content within the preset time period.

[0007] Based on the profile data and the scene sequence data, recommended content is determined from the pre-selected content and then recommended to the user.

[0008] According to a second aspect of this disclosure, a content recommendation device based on multiple scenarios is provided, comprising:

[0009] The acquisition unit is used to acquire user profile data and scene sequence data; wherein, the profile data represents the user's historical behavior, the scene sequence data represents multiple scenes experienced by the user within a preset time period, and the historical behavior represents the user's behavior towards recommended historical content within the preset time period.

[0010] The recommendation unit is used to determine recommended content from the pre-selected content based on the profile data and the scene sequence data, and recommend the recommended content to the user.

[0011] According to a third aspect of this disclosure, an electronic device is provided, comprising:

[0012] At least one processor; and

[0013] A memory that is communicatively connected to the at least one processor;

[0014] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0015] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the method described in the first aspect.

[0016] According to a fifth aspect of this disclosure, a computer program product is provided, the computer program product comprising: a computer program that, when executed by a processor, implements the method described in the first aspect.

[0017] The technology disclosed herein improves the accuracy of content recommendation.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0019] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0020] Figure 1 This is a flowchart illustrating a content recommendation method based on multiple scenarios provided in an embodiment of this disclosure;

[0021] Figure 2 This is a flowchart illustrating a content recommendation method based on multiple scenarios provided in an embodiment of this disclosure;

[0022] Figure 3 This is a flowchart illustrating a content recommendation method based on multiple scenarios provided in an embodiment of this disclosure;

[0023] Figure 4 This is a schematic diagram of the connection relationship of the network structure provided according to the embodiments of this disclosure;

[0024] Figure 5 This is a structural block diagram of a content recommendation device based on multiple scenarios provided in an embodiment of this disclosure;

[0025] Figure 6 This is a structural block diagram of a content recommendation device based on multiple scenarios provided in an embodiment of this disclosure;

[0026] Figure 7This is a block diagram of an electronic device used to implement the multi-scenario content recommendation method of the embodiments of this disclosure;

[0027] Figure 8 This is a block diagram of an electronic device used to implement the multi-scenario content recommendation method of the embodiments of this disclosure. Detailed Implementation

[0028] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0029] With the rapid development of the mobile internet, artificial intelligence (AI) products need to meet users' refined needs in different scenarios. For example, AI products can be application products such as map software, and the scenarios can be local, remote, while traveling, or not traveling. When users are in different scenarios, different content needs to be recommended to them. However, the content, users, and needs vary greatly in different scenarios, making accurate recommendations a core challenge for developers.

[0030] Currently, content recommendation for users in different scenarios typically involves training a dedicated model for each scenario. That is, data samples are collected independently for each scenario, and a separate recommendation model is trained. Specifically, samples are collected and dedicated models are trained separately for different cities, different devices, and different user groups. However, this approach causes the recommendation model to ignore general and hidden information specific to each scenario, affecting the accuracy of content recommendations. Furthermore, maintaining a separate model for each scenario significantly increases system resource consumption and R&D costs.

[0031] This disclosure provides a content recommendation method, apparatus, electronic device, and storage medium based on multiple scenarios, which are applied to the big data field within the field of artificial intelligence to improve the accuracy of content recommendation.

[0032] It should be noted that the model in this embodiment is not designed for any specific user and does not reflect the personal information of any particular user. It should also be noted that the data in this embodiment comes from a publicly available dataset.

[0033] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0034] To help readers gain a deeper understanding of the implementation principles of this disclosure, the following will be discussed in conjunction with... Figures 1-8 The embodiments are further refined.

[0035] Figure 1 This is a flowchart illustrating a multi-scenario content recommendation method provided according to an embodiment of this disclosure. This method can be executed by a multi-scenario content recommendation device. Figure 1 As shown, the method includes the following steps:

[0036] S101. Obtain user profile data and scene sequence data; wherein, profile data represents the user's historical behavior, scene sequence data represents multiple scenes experienced by the user within a preset time period, and historical behavior represents the user's behavior towards recommended historical content within the preset time period.

[0037] For example, when users use map software or other application products, the product can recommend content to the user. For instance, based on the user's frequently visited destinations, it can recommend nearby restaurants, parking lots, gas stations, etc., or recommend ways to get to the destination. The recommended content can be displayed on the interface with preset content icons. After receiving the recommended content, the user can choose to close the icon (i.e., not open the recommended content), or they can click the icon to view the details of the recommended content.

[0038] Users may be in different scenarios at different times. These scenarios can include local, out-of-town, on-the-go, and off-the-go scenarios. A local scenario refers to the user's current location being their usual residence; an out-of-town scenario refers to the user's current location not being their usual residence; an on-the-go scenario refers to the user currently traveling; and an off-the-go scenario refers to the user not currently traveling. The content recommended to the user may differ across these scenarios. For example, when the user is in a local scenario, local shops can be recommended; when the user is in an out-of-town scenario, out-of-town shops can be recommended. The recommended content can also be consistent across different scenarios. For example, if a user loves food, food recommendations can be given regardless of whether the user is in a local or out-of-town scenario. In other words, the content recommended to the user can have both consistency and differences across different scenarios.

[0039] When recommending content to users, user profile data and scenario sequence data over a period of time can be obtained. Profile data represents a user's historical behavior, which in turn represents the user's actions towards previously recommended content within a preset time period. For example, a user's action towards previously recommended content could be either closing the recommendation or opening and viewing it. The preset time period can be a preset length of time preceding the current moment, such as one month prior to the current moment, i.e., obtaining the user's historical behavior within one month. Scenario sequence data represents multiple scenarios experienced by the user within the preset time period, and can be represented by a set of scenarios experienced within that time period. The order of scenarios in the scenario sequence data can be determined based on the time the user experienced each scenario. The last scenario in the scenario sequence data can be the user's current scenario. For example, if within a month, a user first went to their local city A, then to another city B, and then to another city C, the scenario sequence data could be defined as [local, other city, other city].

[0040] Before acquiring user profile data and scene sequence data, a prompt message can be sent to the user asking whether they allow access to the data. For example, a pop-up window can be displayed on the interface showing permission. The user can choose to allow or deny access. If the user grants permission, the user's profile data and scene sequence data can be acquired; if the user denies permission, the data cannot be acquired.

[0041] The acquired profile data can correspond to scenes in the scene sequence data. Each profile data entry can be labeled with the scene it corresponds to. For example, if the profile data is generated in a local scene, it can be labeled "0"; if the profile data is generated in a remote scene, it can be labeled "1". In this embodiment, basic user information can also be obtained, such as the user's gender and age. Profile data and scene sequence data can be obtained once each time the user opens the application, or they can be obtained in real time or periodically.

[0042] S102. Based on the profile data and scene sequence data, determine the recommended content from the pre-selected content and recommend the recommended content to the user.

[0043] For example, multiple alternative content options can be pre-set, which are content that might be recommended to the user. For instance, historical content previously pushed to the user can be used as alternatives, or content other than historical content can be obtained from big data as alternatives. For example, all shops and parking lots in the user's current location can be identified as alternatives.

[0044] After obtaining the user profile data and scene sequence data, these data are processed. Based on the processing results, one piece of content is selected from the candidate content as the recommended content. For example, the user profile data can be used to determine the content that the user might need in different scenarios. This potential content could be content that the user has previously viewed. In other words, by identifying the user's past viewing history in different scenarios, the differences in user needs across different scenarios can be identified. Based on the content that the user might need in different scenarios, the content that the user might need in the current scenario is determined from the pre-selected content. This selected content is then designated as the recommended content and pushed to the user.

[0045] For example, the most frequently opened historical content by a user in different scenarios can be identified as recommended content. Alternatively, the attribute information of the most frequently opened historical content can be determined, and content with that attribute information can be selected from a preset pool of candidates as recommended content. Attribute information can represent content categories, such as food, parking lots, and modes of transportation. If the most frequently opened content by a user across all scenarios is parking lots, then parking lot-related information can be recommended to the user regardless of their current scenario. In this embodiment, determining the content that users may need for different scenarios allows us to understand the differences in user needs across different scenarios. Combining the content that users may need for different scenarios effectively captures the commonalities in user needs across different scenarios, significantly improving the accuracy of content recommendations and enhancing the user experience.

[0046] In this embodiment, when a user is in a certain scenario, user profile data and scenario sequence data over a period of time can be acquired. Profile data from multiple scenarios can be analyzed and processed, rather than being limited to the profile data of the current scenario. This identifies the commonalities and differences in user needs across different scenarios, accurately meeting the user's real needs in different scenarios. It solves the problem of content recommendation errors caused by analyzing user needs through a single scenario in existing technologies, improving the accuracy of content recommendations and thus enhancing the user experience.

[0047] Figure 2 This is a flowchart illustrating a content recommendation method based on multiple scenarios provided in an embodiment of this disclosure.

[0048] In this embodiment, determining recommended content from pre-selected content based on profile data and scene sequence data can be further refined as follows: determining the feature vectors corresponding to users in multiple scenarios based on profile data and scene sequence data; wherein, the feature vectors are used to represent one scenario in multiple scenarios and the data in the profile data corresponding to that scenario; and determining recommended content from pre-selected content based on the user's feature vectors in all scenarios.

[0049] like Figure 2 The method includes the following steps:

[0050] S201. Obtain user profile data and scene sequence data; wherein, profile data represents the user's historical behavior, scene sequence data represents multiple scenes experienced by the user within a preset time period, and historical behavior represents the user's behavior towards recommended historical content within the preset time period.

[0051] S202. Based on the profile data and scene sequence data, determine the feature vectors corresponding to the user in multiple scenarios; wherein, the feature vectors are used to represent one scenario in the multiple scenarios and the data in the profile data corresponding to that scenario.

[0052] For example, all acquired profile data can be categorized according to scenarios to obtain data corresponding to each scenario. For different scenarios, feature extraction processing is performed on the data corresponding to that scenario in the profile data and on the scenario sequence data, extracting vector data in matrix form. This extracted data is then used as the user's feature vector for that scenario. That is, the feature vector can be used to represent the scenario and the corresponding data in the profile data. The number of scenarios in the scenario sequence data corresponds to the number of feature vectors obtained. For example, if the scenario sequence data is [local, remote, remote], then there are two scenarios: local and remote, resulting in two feature vectors. For each scenario, a corresponding feature vector can be determined separately. That is, when determining the feature vectors, the scenarios do not affect each other, ensuring that the differences between scenarios can be obtained.

[0053] In this embodiment, feature vectors can be determined using a pre-defined feature extraction network, which can consist of fully connected layers or other network layers. This embodiment does not impose specific limitations on the network structure of the feature extraction network.

[0054] In this embodiment, the feature vectors corresponding to users in multiple scenarios are determined based on profile data and scene sequence data, including: determining the data for each scenario from the profile data as user scenario data corresponding to each scenario; and determining the feature vectors of users in each scenario based on each user scenario data and scene sequence data.

[0055] Specifically, the user profile data includes data corresponding to different scenarios. The profile data is segmented to obtain data for each scenario. The data corresponding to each scenario within the profile data is then identified as the user scenario data for that scenario. For example, the time period in which the user is in each scenario can be determined, and the data generated within that time period can be extracted from the profile data as the user scenario data for that scenario.

[0056] Based on user scenario data and unified scenario sequence data, the user's feature vector for that scenario is determined. That is, the user scenario data differs while the scenario sequence data remains the same when determining feature vectors for different scenarios. The user scenario data and scenario sequence data can be combined into a matrix-like vector, which is then input into a pre-defined feature extraction network to obtain the feature vector. In this embodiment, multiple feature extraction networks can be set up for different scenarios to achieve parallel processing of different scenarios. Feature extraction networks can be pre-defined for each scenario. When feature extraction for a user is required, the scenarios included in the scenario sequence data are determined, thus deciding which feature extraction networks to use. For example, the pre-defined feature extraction networks may include networks corresponding to local scenarios, networks corresponding to remote scenarios, networks corresponding to in-line scenarios, and networks corresponding to non-in-line scenarios. If the scenario sequence data only contains local and remote scenarios, the networks corresponding to in-line scenarios and non-in-line scenarios are not used.

[0057] The advantage of this setup is that it identifies data for each scenario from the overall profile data, thereby obtaining feature vectors for each scenario and enabling separate processing for different scenarios, thus improving the accuracy of content recommendation.

[0058] In this embodiment, determining the data in each scenario from the profile data as the user scenario data corresponding to each scenario includes: obtaining the scenario identifier corresponding to each scenario from the profile data; wherein the scenario identifier is used to represent the scenario; determining the data in the profile data corresponding to each scenario based on the scenario identifier corresponding to each scenario, and using the data corresponding to each scenario as the user scenario data corresponding to each scenario.

[0059] Specifically, each data entry in the profile data can be labeled with a scene identifier, which represents a scene. The scene identifiers on each data entry can be the same or different. The process involves retrieving the scene identifiers from each data entry in the profile data and identifying the data entries labeled with the same scene identifier. This data with the same scene identifier is then identified as the data in the profile data corresponding to that scene. In other words, based on the scene identifier, the data in the profile data corresponding to each scene is determined. The data in the profile data corresponding to each scene is then identified as the user scene data for that scene. For example, if the profile data contains 1000 data entries, 300 of them might be user scene data for a local scene, and 700 might be user scene data for a remote scene.

[0060] The advantage of this setup is that, based on scene identifiers, the profile data can be accurately distinguished by scene, thereby enabling precise feature extraction for different scenes. This avoids confusion in profile data across different scenes, improves the accuracy of feature extraction, and ultimately enhances the accuracy of content recommendation.

[0061] In this embodiment, the method further includes: determining the amount of user scenario data corresponding to each scenario in the scenario sequence data; determining the scenario to be adjusted based on the difference in the amount of data between different scenarios; wherein the scenario to be adjusted represents the amount of user scenario data that needs to be adjusted; adjusting the amount of data of the scenario to be adjusted to obtain the adjusted user scenario data corresponding to the scenario to be adjusted.

[0062] Specifically, the amount of user scenario data may vary depending on the scenario. For example, a user might have 10,000 user scenario data entries in a local scenario, but only 100 in a remote scenario. This significant difference in data volume can affect the processing efficiency of user scenario data in different scenarios. Therefore, the amount of user scenario data can be adjusted to balance the data volume across different scenarios.

[0063] Before performing feature extraction on user scenario data, the data volume of user scenario data corresponding to each scenario is determined. The difference in data volume between different scenarios is then determined. Based on this difference, the scenario to be adjusted is identified, and the scenario to be adjusted represents the user scenario data volume required for that scenario. For example, if there is a large difference in data volume between two scenarios, both scenarios can be identified as scenarios to be adjusted, or one of these two scenarios can be identified as the scenario to be adjusted.

[0064] Adjust the amount of data for the scenario to be adjusted. For example, reduce the amount of data for user scenarios with a large amount of data, or increase the amount of data for user scenarios with a small amount of data, to obtain the adjusted user scenario data for the scenario to be adjusted, so that the difference in the amount of data between scenarios is reduced after adjustment.

[0065] The advantage of this setup is that it adjusts the amount of user scenario data for different scenarios, balances the differences in data volume between different scenarios, avoids asynchronous generation of feature vectors corresponding to different scenarios, and improves the efficiency of feature extraction.

[0066] In this embodiment, determining the scenario to be adjusted based on the difference in data volume between different scenarios includes: determining the difference in data volume between different scenarios; wherein, the different scenarios include a first scenario and a second scenario, the data volume of user scenario data in the first scenario is greater than the data volume of user scenario data in the second scenario; if the difference is greater than a preset difference threshold, then the first scenario is determined as the scenario to be adjusted.

[0067] Specifically, for multiple scenarios in the scenario sequence data, the difference in the amount of user scenario data between any two scenarios can be determined. If this difference is greater than a preset difference threshold, the scenario to be adjusted is determined from these two scenarios, and the amount of user scenario data in the scenario to be adjusted is adjusted until the difference in the amount of user scenario data between any two scenarios is no greater than the preset difference threshold.

[0068] When determining the difference in data volume between different scenarios, one scenario can be designated as the first scenario and the other as the second scenario. The user scenario data volume in the first scenario is greater than that in the second scenario; that is, the scenario with more data volume is designated as the first scenario. If the difference in data volume exceeds a preset difference threshold, then the first scenario is designated as the scenario to be adjusted; that is, the scenario with more data volume is designated as the first scenario.

[0069] The advantage of this setting is that if the difference in data volume is large, the scenario with the larger data volume is identified as the scenario to be adjusted, which makes it easier to adjust the data volume of the scenario with the larger data volume and improve the consistency of user scenario data volume between different scenarios.

[0070] In this embodiment, the data volume of the scenario to be adjusted is adjusted to obtain the adjusted user scenario data corresponding to the scenario to be adjusted, including: reducing the user scenario data of the first scenario according to the data volume of the user scenario data of the second scenario; and determining the user scenario data retained after the reduction as the adjusted user scenario data corresponding to the first scenario.

[0071] Specifically, after identifying the first scenario as the scenario to be adjusted, the amount of user scenario data in the first scenario can be processed based on the amount of user scenario data in the second scenario. For example, the amount of user scenario data in the first scenario can be reduced to make it closer to the amount of user scenario data in the second scenario.

[0072] After reducing the user scenario data for the first scenario, the remaining user scenario data is designated as the adjusted user scenario data for the first scenario. The amount of user scenario data for the second scenario remains unchanged. For example, if the first scenario has 10,000 user scenario data entries and the second scenario has 100, after reduction, the first scenario can have 1,000 user scenario data entries and the second scenario can have 100.

[0073] The advantage of this setup is that it reduces the existing user scenario data in the first scenario without requiring the addition of user scenario data in the second scenario, simplifying data processing, improving the efficiency of data adjustment, and thus improving the efficiency of content recommendation.

[0074] In this embodiment, the user scenario data of the first scenario is reduced based on the amount of user scenario data of the second scenario, including: determining the average value between the amount of user scenario data of the first scenario and the amount of user scenario data of the second scenario; and reducing the user scenario data of the first scenario to the average value based on the average value.

[0075] Specifically, if it is determined that the amount of user scenario data in the first scenario needs to be adjusted, the average value between the amount of user scenario data in the first scenario and the amount of user scenario data in the second scenario is determined. Based on this average value, the user scenario data in the first scenario is then reduced; for example, it can be reduced to the average value to obtain the reduced user scenario data for the first scenario.

[0076] The advantage of this setup is that by determining the average value, the amount of data in the first scenario can be quickly reduced, making the amount of user scenario data in the first scenario close to the amount of user scenario data in the second scenario, thus achieving a balance in the amount of data between the first and second scenarios.

[0077] In this embodiment, multiple attention networks are pre-set, with one attention network corresponding to one scene. The attention network is a neural network structure based on the attention mechanism. Based on each user scene data and scene sequence data, the feature vector of the user in each scene is determined, including: determining the attention network corresponding to each user scene data according to the scene represented by each user scene data; and inputting all user scene data and scene sequence data as N sets of input data into the corresponding attention networks to obtain the feature vector of the user in each scene, where N is a positive integer and each set of input data includes scene sequence data and one user scene data.

[0078] Specifically, this embodiment can use a pre-defined feature extraction network model for feature extraction. This network model can employ a multi-head attention mechanism, which means there are multiple attention networks. The attention network is a neural network structure based on the attention mechanism and can be used for feature extraction. One attention network corresponds to one scene. An attention network for each scene is pre-constructed, and the attention network can include fully connected layers and other network layers.

[0079] The process involves identifying the user scenario data corresponding to each scenario in the scenario sequence data, and then determining the attention network corresponding to that user scenario data. This means identifying the attention network for each scenario within the scenario sequence data. The scenario sequence data and the corresponding user scenario data are input into the attention network for that scenario. The network layers within the attention network perform feature extraction to obtain the user's feature vector for that scenario. Each attention network can extract features from its own input data without affecting others. User scenario data and scenario sequence data for a single scenario can be used as a set of input data. If there are N scenarios, all user scenario data and scenario sequence data can be used as N sets of input data, each input to its corresponding attention network. Each attention network outputs a feature vector for its own scenario.

[0080] The advantage of this setup is that the multi-head attention mechanism can better adapt to various scenarios, enabling feature extraction for different scenarios within a single model. This intelligently and accurately meets users' diverse content needs and improves the model's generalization performance. The multi-head attention mechanism also has the advantage of improving scenario discrimination, allowing for the recommendation of corresponding high-quality content based on the scenario, thus improving the accuracy of content recommendation.

[0081] In this embodiment, all user scene data and scene sequence data are used as N sets of input data, which are respectively input into the corresponding attention network to obtain the user's feature vector in each scene. This includes: performing feature extraction processing on the user scene data corresponding to the attention network based on the attention network to obtain the matrix vector of each user scene data; determining the weight of each scene according to the number of scenes in the scene sequence data; wherein, the weight of the scene represents the importance of the scene; and obtaining the user's feature vector in each scene based on the matrix vector of each user scene data and the weight of each scene.

[0082] Specifically, for each scene's corresponding attention network, the input data consists of scene sequence data and user scene data. Through the network layers in the attention network, features can be extracted from the user scene data, and the result is a matrix vector corresponding to the user scene data.

[0083] Feature extraction can also be performed on scene sequence data, which consists of all the scenes experienced by the user within a certain period. The scene sequence data obtained by each attention network is consistent. Using attention networks can delve deeper into the hidden information in the scene sequence data. In this embodiment, the hidden information can refer to the weights of each scene in the scene sequence data. The weights can represent the importance of the scene to the user. The larger the weight, the more important the scene is to the user, and the more reliable the matrix vector corresponding to that scene is.

[0084] The attention network determines the weight of its corresponding scene based on the scene sequence data. For example, if the scene sequence data is [local, remote, remote], then it is necessary to use an attention network for the local scene and an attention network for the remote scene. The weight of the local scene is determined by the attention network for the local scene, and the weight of the remote scene is determined by the attention network for the remote scene. This results in a situation where the weight of the local scene is smaller and the weight of the remote scene is larger.

[0085] After obtaining the matrix vector and scene weights through the attention network, the matrix vector and scene weights can be combined to obtain the user's feature vector in that scene. For example, the matrix vector and scene weights can be concatenated to obtain a matrix-form vector data, which can be used as the feature vector.

[0086] The advantage of this setup is that by using a multi-head attention mechanism to obtain matrix vectors for different scenarios and deeply mining the hidden information in the scenario sequence data, it can effectively capture the commonalities and differences in user interests and needs in different scenarios, efficiently understand and accurately meet the real needs of users in different scenarios, and improve the accuracy of content recommendation.

[0087] S203. Based on the user's feature vectors in all scenarios, determine the recommended content from the pre-selected content and recommend the recommended content to the user.

[0088] For example, based on profile data and scene sequence data, feature vectors can be determined for each scene in the scene sequence data. Based on the feature vectors for all scenes in the scene sequence data, the content most likely to be needed by the user is determined from the pre-selected content and used as recommended content, which is then pushed to the user.

[0089] Based on the feature vector, the recommendation probability of each candidate content can be determined. The recommendation probability represents the user's level of demand for the candidate content. The higher the recommendation probability of candidate content, the more likely the user is to need that candidate content. The candidate content with the highest recommendation probability can be identified as the recommended content. For example, if the candidate content consists of previously recommended historical content, the recommendation probability of each historical content can be determined based on the historical content represented in the feature vector, and the historical content with the highest recommendation probability can be identified as the recommended content.

[0090] For each scenario, a feature vector can be obtained, enabling targeted processing for different scenarios and identifying the differences between them. By combining the feature vectors from all scenarios, the final recommended content is obtained, identifying the commonalities between different scenarios, achieving efficient understanding and accurate fulfillment of users' real needs, and improving the accuracy of content recommendation.

[0091] In this embodiment, when a user is in a certain scenario, user profile data and scenario sequence data over a period of time can be acquired. Profile data from multiple scenarios can be analyzed and processed, rather than being limited to the profile data of the current scenario. This identifies the commonalities and differences in user needs across different scenarios, accurately meeting the user's real needs in different scenarios. It solves the problem of content recommendation errors caused by analyzing user needs through a single scenario in existing technologies, improving the accuracy of content recommendations and thus enhancing the user experience.

[0092] Figure 3 This is a flowchart illustrating a content recommendation method based on multiple scenarios provided in an embodiment of this disclosure.

[0093] In this embodiment, recommended content is determined from the pre-selected content based on the user's feature vectors in all scenarios. This is further refined as follows: based on the user's feature vectors in each scenario, the recommendation probability of historical content corresponding to that scenario is determined; and based on the recommendation probability of historical content corresponding to all scenarios, recommended content is determined from the pre-selected content.

[0094] like Figure 3 The method includes the following steps:

[0095] S301. Obtain user profile data and scene sequence data; wherein, profile data represents the user's historical behavior, scene sequence data represents multiple scenes experienced by the user within a preset time period, and historical behavior represents the user's behavior towards recommended historical content within the preset time period.

[0096] S302. Based on the profile data and scene sequence data, determine the user's feature vector in the scene; wherein, the feature vector is used to represent the data in the scene and the profile data corresponding to the scene.

[0097] S303. Based on the user's feature vector in each scenario, determine the recommendation probability of historical content corresponding to each scenario.

[0098] For example, after determining the user's feature vector in each scenario, the recommendation probability of each piece of historical content in that scenario can be determined. Here, each scenario refers to a scenario contained in the scenario sequence data, and the historical content in that scenario refers to the content recommended to the user when the user was in that scenario within a preset time period.

[0099] The feature vector can include a matrix vector corresponding to user scenario data and scenario weights. That is, the recommendation probability of historical content can be determined based on the matrix vector of user scenario data and the scenario weights. For example, data located in a preset row and column can be extracted from the matrix vector, and the extracted data can be calculated with the weights in a preset manner. The result is the recommendation probability. The higher the recommendation probability, the more interested the user is in the historical content. In this embodiment, the calculation method for the recommendation probability is not specifically limited.

[0100] In this embodiment, the recommendation probability of historical content corresponding to each scenario is determined based on the user's feature vector in each scenario. This includes: inputting the feature vector of each scenario into a preset fully connected neural network; and performing feature extraction processing on the feature vector of each scenario through the preset fully connected neural network to obtain the recommendation probability of historical content corresponding to each scenario recommended to the user.

[0101] Specifically, a fully connected network (FCN) structure is pre-configured, and the feature vectors obtained in step S302 for each scene are input into the FCN network. That is, the input to the FCN can be the output of a multi-head attention mechanism. The number of scene sequences corresponds to the number of scene vectors in the FCN input. The FCN network structure can include fully connected layers and other network layers for feature extraction. In this embodiment, the specific network structure of the FCN is not limited.

[0102] After inputting multiple feature vectors into the FCN, the FCN can extract features from the feature vectors and output the recommendation probability of each historical content. For a given scenario, there can be multiple historical content entries, and the recommendation probabilities of these entries for each scenario can be obtained; that is, for each historical content entry, a recommendation probability can be obtained.

[0103] The advantage of this setup is that it inputs the data output from the attention networks of all scenarios into the FCN, enabling the processing of feature vectors from different scenarios through a single overall FCN network, thus meeting the recommendation needs of different scenarios.

[0104] In this embodiment, a preset fully connected neural network is used to perform feature extraction processing on the feature vector of each scenario to obtain the recommendation probability of historical content corresponding to the scenario recommended to the user. This includes: performing convolution processing on the user's matrix vector in each scenario using a preset fully connected neural network to obtain the initial probability of historical content corresponding to each scenario recommended to the user; and determining the recommendation probability of historical content corresponding to each scenario recommended to the user based on the weight of each scenario and each initial probability.

[0105] Specifically, feature extraction processing using FCN can involve convolutional processing of the user scenario data matrix vector within the feature vector. This matrix vector can include previously recommended content. By performing convolution, a probability can be calculated for each piece of historical content, and this calculated probability is used as the initial probability. The initial probability represents the likelihood of recommending a feature to the user without considering scenario weights.

[0106] After obtaining the initial probabilities of historical content in each scenario, the recommendation probability of the historical content corresponding to that scenario can be calculated based on the initial probability and the weight of that scenario. For example, the initial probability can be multiplied by the weight of the scenario to obtain the recommendation probability.

[0107] The advantage of this setup is that by combining the hidden information in the scene sequence data, namely the scene weight, the recommendation probability of each historical content can be determined, thereby achieving accurate determination of the recommendation probability in multiple scenes and improving the accuracy of content recommendation.

[0108] S304. Based on the recommendation probability of historical content corresponding to all scenarios, determine the recommended content from the pre-selected content and recommend the recommended content to the user.

[0109] For example, multiple candidate content options are pre-set. Based on the recommendation probability of historical content corresponding to all scenarios, a recommended content option is determined from these options. For instance, if historical content is included among the candidate options, the historical content with the highest recommendation probability can be selected as the recommended content. If historical content is not included among the candidate options, the historical content with the highest recommendation probability can be determined first, and then the content with the strongest relevance to that historical content can be selected from the candidate options. This selected content is then chosen as the content with the highest recommendation probability among the candidate options, i.e., the recommended content is selected as the recommended content. After obtaining the recommended content, it can be presented to the user, for example, displayed on the application's interface.

[0110] By combining the recommendation probability of historical content in different scenarios, the final recommended content for users is determined, breaking down the barriers between different scenarios. No matter what scenario the user is in, content can be recommended to the user, avoiding overly simplistic recommendations, meeting the recommendation needs of different scenarios, improving the accuracy of content recommendations, and enhancing the user experience.

[0111] In this embodiment, the pre-selected content consists of recommended historical content; the recommended content is determined from the pre-selected content based on the recommendation probability of the historical content corresponding to all scenarios, including: sorting the recommendation probabilities of the historical content corresponding to all scenarios; and determining the historical content corresponding to the recommendation probability located at a preset sorting position as the recommended content.

[0112] Specifically, the preset selection content can be previously recommended historical content. After obtaining the recommendation probability of each historical content, the recommendation probabilities can be sorted. For example, they can be sorted in descending order. Based on the sorting result, the historical content corresponding to the recommendation probability at the preset sorting position is determined, and that historical content is designated as the recommended content. For example, the historical content ranked first can be designated as the recommended content, that is, the historical content with the highest recommendation probability is designated as the recommended content.

[0113] The advantage of this setup is that, by determining the probability of recommendation, recommended content can be quickly identified from historical content, ensuring that the recommended content is something the user has previously clicked on and viewed. This guarantees that the user is interested in the recommended content and effectively improves the user experience.

[0114] In this embodiment, the pre-selected content consists of all content in a preset database. Recommended content is determined from the pre-selected content based on the recommendation probabilities of historical content corresponding to all scenarios. This includes: sorting the recommendation probabilities of historical content corresponding to all scenarios; identifying the historical content corresponding to the recommendation probability at a preset sorting position as content of interest; determining the content category of the content of interest; searching for content of the content category in the pre-selected content; and identifying the content of the content category as recommended content.

[0115] Specifically, the candidate content can be content other than historical content. After obtaining the recommendation probability of each piece of historical content, the recommendation probabilities of all historical content are sorted, for example, in descending order. The historical content corresponding to the recommendation probability at a preset sorting position is determined; for example, the historical content corresponding to the highest recommendation probability can be determined, and this historical content is identified as the content of interest. The content of interest is the content that the user is most interested in among the historical content, not necessarily the recommended content. Recommended content can then be determined from the candidate content based on the content of interest.

[0116] The system identifies content categories of interest, searches for content in that category from a pre-selected pool, and selects content from that category as recommended content. If multiple items in the same category are available as options, one item can be randomly selected as the recommended content. Alternatively, a recommended item can be determined from multiple items in the same category based on the user's current location. For example, if the content category is parking information, the system can retrieve the user's current location and recommend nearby parking information to the user.

[0117] The advantage of this setup is that it allows for the selection of content that best meets user needs from a large pool of information, expanding the range of available content, increasing the breadth of content recommendations, avoiding overly limited recommendations, and ultimately helping to meet user needs and improve user experience.

[0118] Figure 4 This diagram illustrates the connection relationships of the network structure provided in this embodiment. The network structure of the model used in determining recommended content may include a multi-head attention mechanism and an FCN, where the multi-head attention mechanism may include multiple attention networks. Figure 4In this implementation, the input data for Scenario 1 consists of scenario sequence data and user scenario data for Scenario 1; the input data for Scenario 2 consists of scenario sequence data and user scenario data for Scenario 2; and the input data for Scenario 3 consists of scenario sequence data and user scenario data for Scenario 3. The multi-head attention mechanism uses n attention networks. Since the scenario sequence data only includes scenarios 1, 2, and 3, only the attention networks for scenarios 1, 2, and 3 are used. The input data for scenario 1 is input into the attention network for scenario 1 to obtain the first feature vector; the input data for scenario 2 is input into the attention network for scenario 2 to obtain the second feature vector; and the input data for scenario 3 is input into the attention network for scenario 3 to obtain the third feature vector. All attention networks share a single Functional Network (FCN), meaning the first, second, and third feature vectors are all input into the FCN to ultimately obtain the recommended content. In this embodiment, by using a multi-head attention mechanism and an FCN, content recommendation for different scenarios can be achieved using a single model, effectively saving system resources.

[0119] In this embodiment, when a user is in a certain scenario, user profile data and scenario sequence data over a period of time can be acquired. Profile data from multiple scenarios can be analyzed and processed, rather than being limited to the profile data of the current scenario. This identifies the commonalities and differences in user needs across different scenarios, accurately meeting the user's real needs in different scenarios. It solves the problem of content recommendation errors caused by analyzing user needs through a single scenario in existing technologies, improving the accuracy of content recommendations and thus enhancing the user experience.

[0120] Figure 5 This is a structural block diagram of a content recommendation device based on multiple scenarios, provided in an embodiment of this disclosure. For ease of explanation, only the parts relevant to the embodiments of this disclosure are shown. (Refer to...) Figure 5 The content recommendation device 500 based on multiple scenarios includes: an acquisition unit 501 and a recommendation unit 502.

[0121] The acquisition unit 501 is used to acquire user profile data and scene sequence data; wherein, the profile data represents the user's historical behavior, the scene sequence data represents multiple scenes experienced by the user within a preset time period, and the historical behavior represents the user's behavior towards recommended historical content in the scenarios within the preset time period;

[0122] The recommendation unit 502 is used to determine recommended content from the pre-selected content based on the profile data and the scene sequence data, and recommend the recommended content to the user.

[0123] Figure 6A structural block diagram of a content recommendation device based on multiple scenarios provided in this disclosure embodiment is shown below. Figure 6 As shown, the content recommendation device 600 based on multiple scenarios includes an acquisition unit 601 and a recommendation unit 602, wherein the recommendation unit 602 includes a first determination module 6021 and a second determination module 6022.

[0124] The first determining module 6021 is used to determine the feature vector corresponding to the user in multiple scenarios based on the profile data and the scene sequence data; wherein, the feature vector is used to represent one scenario in the multiple scenarios and the data in the profile data corresponding to that scenario;

[0125] The second determining module 6022 is used to determine recommended content from the pre-selected content based on the user's feature vectors in all scenarios.

[0126] In one example, the first determining module 6021 includes:

[0127] The data determination submodule is used to determine the data in each scenario from the profile data, as the user scenario data corresponding to each scenario;

[0128] The feature determination submodule is used to determine the feature vector of the user in each scenario based on the user scenario data and scenario sequence data.

[0129] In one example, the data determination submodule is specifically used for:

[0130] Obtain a scene identifier corresponding to each scene from the portrait data; wherein, the scene identifier is used to represent a scene;

[0131] Based on the scene identifier corresponding to each scene, the data in the profile data corresponding to each scene is determined, and the data corresponding to each scene is used as the user scene data corresponding to each scene.

[0132] One example also includes:

[0133] A data volume determination unit is used to determine the data volume of user scene data corresponding to each scene in the scene sequence data;

[0134] A scenario determination unit is used to determine the scenario to be adjusted based on the difference in data volume between different scenarios; wherein, the scenario to be adjusted represents the data volume of user scenario data that needs to be adjusted;

[0135] The data volume adjustment unit is used to adjust the data volume of the scenario to be adjusted, so as to obtain the adjusted user scenario data corresponding to the scenario to be adjusted.

[0136] In one example, the scene determination unit includes:

[0137] The difference determination module is used to determine the difference in data volume between different scenarios; wherein, the different scenarios include a first scenario and a second scenario, and the data volume of user scenario data in the first scenario is greater than the data volume of user scenario data in the second scenario;

[0138] The difference comparison module is used to determine the first scenario as the scenario to be adjusted if the difference is greater than a preset difference threshold.

[0139] In one example, the data volume adjustment unit includes:

[0140] The deletion submodule is used to delete user scenario data of the first scenario based on the amount of user scenario data of the second scenario.

[0141] The adjustment completion submodule is used to determine the user scenario data retained after data deletion and processing as the adjusted user scenario data corresponding to the first scenario.

[0142] In one example, submodules are removed, specifically for:

[0143] Determine the average value between the data volume of user scenario data in the first scenario and the data volume of user scenario data in the second scenario;

[0144] The user scenario data for the first scenario is reduced based on the average value, reducing the user scenario data for the first scenario to the amount of the average value.

[0145] In one example, multiple attention networks are pre-set, with one attention network corresponding to one scene. The attention network is a neural network structure based on an attention mechanism. The feature determination submodule includes:

[0146] The network determination submodule is used to determine the attention network corresponding to each user scene data according to the scene represented by each user scene data.

[0147] The data input submodule is used to input all the user scene data and scene sequence data as N sets of input data into the corresponding attention network to obtain the feature vector of the user in each scene, where N is a positive integer and each set of input data includes the scene sequence data and one user scene data.

[0148] In one example, the data input submodule is specifically used for:

[0149] Based on the attention network, feature extraction processing is performed on the user scene data corresponding to the attention network to obtain a matrix vector of each user scene data.

[0150] The weight of each scene is determined based on the number of scenes in the scene sequence data; wherein the weight of a scene represents the importance of the scene.

[0151] Based on the matrix vector of each user scenario data and the weight of each scenario, the feature vector of the user in each scenario is obtained.

[0152] In one example, the second determining module 6022 includes:

[0153] The probability determination submodule is used to determine the recommendation probability of historical content corresponding to each scenario based on the user's feature vector in each scenario.

[0154] The content determination submodule is used to determine the recommended content from the pre-selected content based on the recommendation probability of historical content corresponding to all the scenarios.

[0155] In one example, the probability determination submodule includes:

[0156] The vector input submodule is used to input the feature vector of each scenario into a preset fully connected neural network;

[0157] The probability output submodule is used to perform feature extraction processing on the feature vector of each scenario through the preset fully connected neural network to obtain the recommendation probability of historical content corresponding to each scenario recommended to the user.

[0158] In one example, the probability output submodule is specifically used for:

[0159] The user's matrix vector in each scenario is convolved using the preset fully connected neural network to obtain the initial probability of the historical content corresponding to each scenario recommended to the user.

[0160] Based on the weight of each scenario and the initial probability of each scenario, the recommendation probability of historical content corresponding to each scenario is determined for the user.

[0161] In one example, the pre-selected content is recommended historical content; the content determination submodule is specifically used for:

[0162] The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked.

[0163] The historical content corresponding to the recommendation probability located at the preset sorting position is determined as the recommended content.

[0164] In one example, the preset selected content is all content in a preset database; the content determination submodule is specifically used for:

[0165] The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked.

[0166] The historical content corresponding to the recommendation probability located at the preset sorting position is identified as content of interest;

[0167] The content category of the content of interest is determined, the content of the content category is searched from the preset selected content, and the content of the content category is determined as the recommended content.

[0168] According to embodiments of this disclosure, this disclosure also provides an electronic device.

[0169] Figure 7 A structural block diagram of an electronic device provided in this disclosure embodiment, such as... Figure 7 As shown, the electronic device 700 includes: at least one processor 702; and a memory 701 communicatively connected to the at least one processor 702; wherein the memory stores instructions executable by the at least one processor 702, the instructions being executed by the at least one processor 702 to enable the at least one processor 702 to execute the multi-scenario-based content recommendation method of this disclosure.

[0170] The electronic device 700 also includes a receiver 703 and a transmitter 704. The receiver 703 is used to receive instructions and data sent by other devices, and the transmitter 704 is used to send instructions and data to external devices.

[0171] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0172] According to embodiments of this disclosure, this disclosure also provides a computer program product comprising: a computer program stored in a readable storage medium, at least one processor of an electronic device being able to read the computer program from the readable storage medium, and the at least one processor executing the computer program causing the electronic device to perform the scheme provided in any of the above embodiments.

[0173] Figure 8A schematic block diagram of an example electronic device 800 that can be used to implement embodiments 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 may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0174] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0175] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0176] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a multi-scene-based content recommendation method. For example, in some embodiments, the multi-scene-based content recommendation method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the multi-scene-based content recommendation method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the multi-scene-based content recommendation method by any other suitable means (e.g., by means of firmware).

[0177] Various embodiments 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-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0178] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0179] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0180] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, 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 sound input, voice input, or tactile input).

[0181] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0182] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0183] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0184] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A content recommendation method based on multiple scenarios, comprising: The system acquires user profile data and scene sequence data; wherein the profile data represents the user's historical behavior, and the scene sequence data represents multiple scenes experienced by the user within a preset time period; the historical behavior represents the user's actions on recommended historical content within the scenes of the preset time period. Based on the profile data and the scene sequence data, recommended content is determined from the pre-selected content and recommended to the user; The step of determining recommended content from pre-selected content based on the profile data and the scene sequence data includes: Based on the profile data and the scene sequence data, feature vectors corresponding to the user in multiple scenarios are determined; wherein, for each of the multiple scenarios, a corresponding feature vector is determined, and the feature vector is used to represent one of the multiple scenarios and the data in the profile data corresponding to that scenario; multiple attention networks are pre-set, one attention network corresponds to one scenario, and the attention network is used to obtain the feature vector of the corresponding scenario; Based on the user's feature vector in each scenario, determine the recommendation probability of historical content corresponding to each scenario; The recommended content is determined from the pre-selected content based on the recommendation probability of historical content corresponding to all the scenarios.

2. The method according to claim 1, wherein, The step of determining the feature vector corresponding to the user in multiple scenarios based on the profile data and the scene sequence data includes: The data for each scenario is determined from the profile data, and used as the user scenario data corresponding to each scenario. Based on each user scenario data and the scenario sequence data, the feature vector of the user in each scenario is determined.

3. The method according to claim 2, wherein, The step of determining the data in each scenario from the profile data, as the user scenario data corresponding to each scenario, includes: Obtain a scene identifier corresponding to each scene from the portrait data; wherein, the scene identifier is used to represent a scene; Based on the scene identifier corresponding to each scene, the data in the profile data corresponding to each scene is determined, and the data corresponding to each scene is used as the user scene data corresponding to each scene.

4. The method according to claim 2 or 3, further comprising: Determine the amount of user scenario data corresponding to each scenario in the scenario sequence data; Based on the difference in data volume between different scenarios, the scenario to be adjusted is determined; wherein, the scenario to be adjusted represents the user scenario data whose data volume needs to be adjusted. The data volume of the scenario to be adjusted is adjusted to obtain the adjusted user scenario data corresponding to the scenario to be adjusted.

5. The method according to claim 4, wherein, The step of determining the scenario to be adjusted based on the difference in data volume between different scenarios includes: Determine the difference in data volume between different scenarios; wherein, the different scenarios include a first scenario and a second scenario, and the data volume of user scenario data in the first scenario is greater than the data volume of user scenario data in the second scenario; If the difference is greater than a preset difference threshold, then the first scenario is determined as the scenario to be adjusted.

6. The method according to claim 5, wherein, The step of adjusting the data volume of the scenario to be adjusted to obtain the adjusted user scenario data corresponding to the scenario to be adjusted includes: Based on the amount of user scenario data in the second scenario, the user scenario data in the first scenario is reduced. The user scenario data retained after deletion is determined as the adjusted user scenario data corresponding to the first scenario.

7. The method according to claim 6, wherein, The step of reducing the user scenario data of the first scenario based on the amount of user scenario data of the second scenario includes: Determine the average value between the data volume of user scenario data in the first scenario and the data volume of user scenario data in the second scenario; The user scenario data for the first scenario is reduced based on the average value, reducing the user scenario data for the first scenario to the amount of the average value.

8. The method according to any one of claims 5-7, wherein, The attention network is a neural network structure based on an attention mechanism; determining the user's feature vector in each scenario based on each user scenario data and the scenario sequence data includes: Based on the scenario represented by each user scenario data, determine the attention network corresponding to each user scenario data; All the user scene data and the scene sequence data are used as N sets of input data and input into the corresponding attention network respectively to obtain the feature vector of the user in each scene, where N is a positive integer and each set of input data includes the scene sequence data and one user scene data.

9. The method according to claim 8, wherein, The step involves taking all the user scene data and the scene sequence data as N sets of input data and inputting them into the corresponding attention network to obtain the feature vector of the user in each scene, including: Based on the attention network, feature extraction processing is performed on the user scene data corresponding to the attention network to obtain a matrix vector of each user scene data. The weight of each scene is determined based on the number of scenes in the scene sequence data; wherein the weight of a scene represents the importance of the scene. Based on the matrix vector of each user scenario data and the weight of each scenario, the feature vector of the user in each scenario is obtained.

10. The method according to claim 9, wherein, The step of determining the recommendation probability of historical content corresponding to each scenario based on the user's feature vector in each scenario includes: The feature vector for each scenario is input into a pre-defined fully connected neural network; The preset fully connected neural network is used to extract features from the feature vectors of each scenario to obtain the recommendation probability of historical content corresponding to each scenario for the user.

11. The method according to claim 10, wherein, The step of extracting features from the feature vectors of each scenario using the preset fully connected neural network to obtain the recommendation probability of historical content corresponding to each scenario for the user includes: The user's matrix vector in each scenario is convolved using the preset fully connected neural network to obtain the initial probability of the historical content corresponding to each scenario recommended to the user. Based on the weight of each scenario and the initial probability of each scenario, the recommendation probability of historical content corresponding to each scenario is determined for the user.

12. The method according to claim 10 or 11, wherein, The pre-selected content consists of recommended historical content; determining recommended content from the pre-selected content based on the recommendation probability of all historical content corresponding to the scenarios includes: The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked. The historical content corresponding to the recommendation probability located at the preset sorting position is determined as the recommended content.

13. The method according to claim 10 or 11, wherein, The pre-selected content consists of all content in a pre-set database; determining recommended content from the pre-selected content based on the recommendation probability of historical content corresponding to all the scenarios includes: The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked. The historical content corresponding to the recommendation probability located at the preset sorting position is identified as content of interest; The content category of the content of interest is determined, the content of the content category is searched from the preset selected content, and the content of the content category is determined as the recommended content.

14. A content recommendation device based on multiple scenarios, comprising: The acquisition unit is used to acquire user profile data and scene sequence data; wherein, the profile data represents the user's historical behavior, and the scene sequence data represents multiple scenes experienced by the user within a preset time period; the historical behavior represents the user's behavior towards recommended historical content in the scenarios within the preset time period. The recommendation unit is used to determine recommended content from the pre-selected content based on the profile data and the scene sequence data, and recommend the recommended content to the user. The recommendation unit includes: The first determining module is used to determine the feature vector corresponding to the user in multiple scenarios based on the profile data and the scene sequence data; wherein, a corresponding feature vector is determined for each scenario in the multiple scenarios, and the feature vector is used to represent one scenario in the multiple scenarios and the data in the profile data corresponding to that scenario; multiple attention networks are pre-set, one attention network corresponds to one scenario, and the attention network is used to obtain the feature vector of the corresponding scenario; The second determining module includes: The probability determination submodule is used to determine the recommendation probability of historical content corresponding to each scenario based on the user's feature vector in each scenario. The content determination submodule is used to determine the recommended content from the pre-selected content based on the recommendation probability of historical content corresponding to all the scenarios.

15. The apparatus according to claim 14, wherein, The first determining module includes: The data determination submodule is used to determine the data in each scenario from the profile data, as the user scenario data corresponding to each scenario; The feature determination submodule is used to determine the feature vector of the user in each scenario based on each user scenario data and the scenario sequence data.

16. The apparatus according to claim 15, wherein, The data determination submodule is specifically used for: Obtain a scene identifier corresponding to each scene from the portrait data; wherein, the scene identifier is used to represent a scene; Based on the scene identifier corresponding to each scene, the data in the profile data corresponding to each scene is determined, and the data corresponding to each scene is used as the user scene data corresponding to each scene.

17. The apparatus according to claim 15 or 16, further comprising: A data volume determination unit is used to determine the data volume of user scene data corresponding to each scene in the scene sequence data; A scenario determination unit is used to determine the scenario to be adjusted based on the difference in data volume between different scenarios; wherein, the scenario to be adjusted represents the data volume of user scenario data that needs to be adjusted; The data volume adjustment unit is used to adjust the data volume of the scenario to be adjusted, so as to obtain the adjusted user scenario data corresponding to the scenario to be adjusted.

18. The apparatus according to claim 17, wherein, The scene determination unit includes: The difference determination module is used to determine the difference in data volume between different scenarios; wherein, the different scenarios include a first scenario and a second scenario, and the data volume of user scenario data in the first scenario is greater than the data volume of user scenario data in the second scenario; The difference comparison module is used to determine the first scenario as the scenario to be adjusted if the difference is greater than a preset difference threshold.

19. The apparatus according to claim 18, wherein, The data volume adjustment unit includes: The deletion submodule is used to delete user scenario data of the first scenario based on the amount of user scenario data of the second scenario. The adjustment completion submodule is used to determine the user scenario data retained after data deletion and processing as the adjusted user scenario data corresponding to the first scenario.

20. The apparatus according to claim 19, wherein, The deletion submodule is specifically used for: Determine the average value between the data volume of user scenario data in the first scenario and the data volume of user scenario data in the second scenario; The user scenario data for the first scenario is reduced based on the average value, reducing the user scenario data for the first scenario to the amount of the average value.

21. The apparatus according to any one of claims 18-20, wherein, The attention network is a neural network structure based on the attention mechanism; The feature determination submodule includes: The network determination submodule is used to determine the attention network corresponding to each user scene data according to the scene represented by each user scene data. The data input submodule is used to input all the user scene data and the scene sequence data as N sets of input data into the corresponding attention network to obtain the feature vector of the user in each scene, where N is a positive integer and each set of input data includes the scene sequence data and one user scene data.

22. The apparatus according to claim 21, wherein, The data input sub-module is specifically used for: Based on the attention network, feature extraction processing is performed on the user scene data corresponding to the attention network to obtain a matrix vector of each user scene data. The weight of each scene is determined based on the number of scenes in the scene sequence data; wherein the weight of a scene represents the importance of the scene. Based on the matrix vector of each user scenario data and the weight of each scenario, the feature vector of the user in each scenario is obtained.

23. The apparatus according to claim 22, wherein, The probability determination submodule includes: The vector input submodule is used to input the feature vector of each scenario into a preset fully connected neural network; The probability output submodule is used to perform feature extraction processing on the feature vector of each scenario through the preset fully connected neural network to obtain the recommendation probability of historical content corresponding to each scenario recommended to the user.

24. The apparatus according to claim 23, wherein, The probability output submodule is specifically used for: The user's matrix vector in each scenario is convolved using the preset fully connected neural network to obtain the initial probability of the historical content corresponding to each scenario recommended to the user. Based on the weight of each scenario and the initial probability of each scenario, the recommendation probability of historical content corresponding to each scenario is determined for the user.

25. The apparatus according to claim 23 or 24, wherein, The pre-selected content consists of recommended historical content; the content determination submodule is specifically used for: The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked. The historical content corresponding to the recommendation probability located at the preset sorting position is determined as the recommended content.

26. The apparatus according to claim 23 or 24, wherein, The pre-selected content consists of all content in the pre-set database; the content determination submodule is specifically used for: The recommendation probabilities of historical content corresponding to all the aforementioned scenarios are ranked. The historical content corresponding to the recommendation probability located at the preset sorting position is identified as content of interest; The content category of the content of interest is determined, the content of the content category is searched from the preset selected content, and the content of the content category is determined as the recommended content.

27. 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 to enable the at least one processor to perform the method of any one of claims 1-13.

28. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-13.

29. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1-13.

Citation Information

Patent Citations

  • User preference evaluation method and device

    CN116957626A

  • Construction method of multi-scene recommendation model, scene recommendation method and related equipment

    CN117011643A