User interest profile generation method and apparatus, device, storage medium, and product

Abstract

Embodiments of the present application provide a user interest profile generation method and apparatus, a device, a storage medium, and a product. The method comprises: obtaining room text information corresponding to a voice room most recently accessed by a user, a statistical time range of the room text information being determined on the basis of a trigger time point of an active behavior event of the user; determining a short-term interest profile of the user on the basis of the room text information and a trained first text classification model; obtaining a recorded historical short-term profile of the user and a room interest profile of the concerned voice room, and determining a long-term interest profile of the user on the basis of the short-term interest profile, the historical short-term profile, and the room interest profile; and combining the short-term interest profile and the long-term interest profile to obtain a user interest profile. According to the present solution, reasonable construction of the short-term interest profile of the user is implemented, long-term interest characteristics of the user are accurately mastered, a comprehensive user interest profile is obtained, and related content conforming to user interests is recommended, thus a personalized experience is customized.

Description

User interest profile generation method, device, equipment, storage medium and product

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on February 7, 2024, with application number 202410175392.9, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The embodiments of the present application relate to the field of computer technology, and in particular to a method, apparatus, device, storage medium, and product for generating a user interest profile. Background Art

[0003] With the increasing popularity of social media and real-time communication tools, voice chat rooms are rapidly emerging as a new form of social interaction. Voice chat rooms provide a platform for live streamers to share their experiences and for users to interact. Users can listen to live streamers' voices or directly participate in multi-topic discussions and share insights, gaining timely access to information and building stronger social connections. To enhance the user experience, voice chat room platforms need to better understand user interests in order to recommend relevant content and customize the interactive experience. Because the data generated during active voice chat rooms is vast and complex, and users discuss a wide variety of topics in voice chat rooms, with discussions covering a wide range of depth and breadth, the interests of users in voice chat rooms can vary greatly. Understanding and adapting to these diverse interests and contexts is crucial to accurately capture user preferences and ensure that the content delivered meets user expectations.

[0004] Related technologies use rule-based or traditional machine learning methods to generate user portraits, but lack personalized analysis of user interests, cannot capture real-time changes in user interests, and cannot fully and accurately grasp user interests, which needs to be improved. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, device, storage medium and product for generating a user interest portrait, which solves the problem in related technologies that it is impossible to capture the real-time changes of user interests and it is impossible to fully and accurately grasp the user's interests. It effectively captures the real-time changes of user interests, reasonably constructs the user's short-term interest portrait, accurately grasps the user's long-term interest characteristics, and effectively constructs the user's long-term interest portrait, thereby obtaining a comprehensive and accurate user interest portrait, which is conducive to recommending relevant content that meets the user's interests and customizing a personalized interactive experience.

[0006] In a first aspect, an embodiment of the present application provides a method for generating a user interest profile, the method comprising:

[0007] Obtaining room text information corresponding to the voice room most recently visited by the user, where the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event;

[0008] Determining the user's short-term interest profile based on the room text information and the trained first text classification model;

[0009] Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait;

[0010] The short-term interest portrait and the long-term interest portrait are combined to obtain a user interest portrait.

[0011] In a second aspect, an embodiment of the present application further provides a device for generating a user interest profile, the device comprising:

[0012] An acquisition module is configured to acquire room text information corresponding to a voice room that a user has visited most recently, wherein a statistical time range of the room text information is determined based on a triggering time point of an active behavior event of the user;

[0013] a short-term portrait generation module, configured to determine the user's short-term interest portrait based on the room text information and the trained first text classification model;

[0014] a long-term portrait generation module configured to obtain the recorded historical short-term portraits of the user and the room interest portraits of the voice rooms of interest, and determine the long-term interest portrait of the user based on the short-term interest portraits, the historical short-term portraits, and the room interest portraits;

[0015] The user portrait generation module is configured to combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0016] In a third aspect, an embodiment of the present application further provides a user interest profile generation device, the device comprising:

[0017] one or more processors;

[0018] a storage device configured to store one or more programs,

[0019] When the one or more programs are executed by the one or more processors, the one or more processors implement the user interest portrait generation method described in the embodiment of the present application.

[0020] In a fourth aspect, an embodiment of the present application further provides a non-volatile storage medium storing computer-executable instructions, which, when executed by a computer processor, are configured to execute the user interest portrait generation method described in an embodiment of the present application.

[0021] In the fifth aspect, an embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor of the device reads and executes the computer program from the computer-readable storage medium, so that the device executes the user interest portrait generation method described in the embodiment of the present application.

[0022] In an embodiment of the present application, by obtaining the room text information corresponding to the voice room that the user last visited, the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event, and based on the room text information and the trained first text classification model, the user's short-term interest profile is determined, the recorded historical short-term profile of the user and the room interest portrait of the voice room that the user is following are obtained, and the user's long-term interest profile is determined based on the short-term interest portrait, the historical short-term portrait, and the room interest portrait; and the short-term interest portrait and the long-term interest portrait are combined to obtain the user interest portrait. In the above scheme, by obtaining the room text information of the voice room that the user last visited, the reference information of the user's interest is effectively determined based on the triggering time point of the user's active behavior event, and by utilizing the room text information that meets the multi-source information and the corresponding type of text classification model, the real-time changes of the user's interests are effectively captured, and the user's short-term interest profile is reasonably constructed. By integrating multiple recorded short-term interest portraits and the room interest portraits corresponding to the voice room that the user is following, the user's long-term interest characteristics are accurately grasped, and the user's long-term interest profile is effectively constructed, thereby obtaining a comprehensive and accurate user interest profile, which is conducive to recommending relevant content that meets the user's interests and customizing a personalized interactive experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] FIG1 is a flow chart of a method for generating a user interest profile according to an embodiment of the present application;

[0024] FIG2 is a flow chart of a method for generating a user interest profile including a short-term interest profile generation process provided by an embodiment of the present application;

[0025] FIG3 is a flowchart of a method for generating a user interest profile including a weight value calculation process provided by an embodiment of the present application;

[0026] FIG4 is a flowchart of another method for generating a user interest portrait including a short-term interest portrait generation process provided by an embodiment of the present application;

[0027] FIG5 is a flow chart of a method for generating a user interest portrait including a long-term interest portrait generation process provided by an embodiment of the present application;

[0028] FIG6 is a flowchart of a method for generating a user interest portrait including a room interest portrait generation process provided by an embodiment of the present application;

[0029] FIG7 is a structural block diagram of a user interest profile generating device provided in an embodiment of the present application;

[0030] FIG8 is a schematic structural diagram of a user interest portrait generation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following is a further detailed description of the embodiments of the present application in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the embodiments of the present application, and are not intended to limit the embodiments of the present application. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions of the embodiments of the present application, rather than all structures.

[0032] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0033] The user interest profile generation method provided in the embodiments of this application can be used to combine multi-source text from voice rooms visited by a user, voice text during the user's visit, and other information to construct a user interest profile and effectively optimize personalized recommendation services. Relevant application scenarios may include: live voice broadcasting, voice social networking, voice teaching, etc. The several application scenarios listed above are merely exemplary and explanatory. In actual applications, the user interest profile generation method can also be used in real-time voice communication in other scenarios, and the embodiments of this application are not limited to this.

[0034] The related technologies that use rule-based or traditional machine learning methods to generate user profiles may have some shortcomings when actually solving the problem of interest understanding and recommendation in real-time voice scenarios, such as insufficient accuracy, real-time nature, and personalization. First, when processing voice data, the related technologies have difficulty efficiently identifying, analyzing, and understanding users' real-time voice input, and are unable to fully utilize the rich information generated by users in real-time voice communication in the voice room, thus failing to provide users with an accurate experience. Second, the related system platforms have certain limitations in responding to the diversity and complexity of user interest expressions. Due to the wide range of topics in the voice room, it is often difficult to accurately capture the changes in users' interests in different topics, resulting in insufficient personalization of recommended content. In addition, the data sources used by the related technologies are relatively single or one-sided, and they cannot provide an in-depth and comprehensive portrayal of users' interests. Based on this, in order to better meet the service needs of real-time voice scenarios, this application aims to provide a user interest portrait generation method, device, equipment, storage medium and product to establish a comprehensive and accurate user interest portrait. The user interest portrait can be used to provide a more accurate reference signal for the recommendation system, so that it can accurately capture the user's interest points, thereby providing customized and personalized room and content recommendations, which helps to improve the user's experience of discovering content of interest, and also injects higher intelligence factors into the platform's content recommendation algorithm. It can also intelligently recommend users with similar interests, promote the establishment of social relationships, and help users expand their social circles more targetedly, enhancing the depth and breadth of the social experience. Furthermore, the user interest portrait can help the operation and product teams to have a deeper understanding of the user's preferences and tendencies, thereby implementing more refined operation strategies, which is conducive to improving the platform's retention rate and user engagement.

[0035] In the user interest portrait generation method provided in the embodiment of the present application, the execution entity of each step can be a computer device, which refers to any electronic device with data calculation, processing and storage capabilities, such as mobile phones, PCs (Personal Computers), tablet computers and other terminal devices, or servers and other devices. The embodiment of the present application does not limit this.

[0036] FIG1 is a flow chart of a method for generating a user interest profile according to an embodiment of the present application. As shown in FIG1 , the method includes the following steps:

[0037] Step S101: Acquire room text information corresponding to the voice room that the user visited most recently, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event.

[0038] The room text information can be text content that reflects the characteristics of the voice content in the voice room, such as the topic of communication and focus, and can be derived from the real-time voice stream, public screen stream, or public information in the voice room. The room text information can include voice recognition text, public screen text, announcement text, and title text. The statistical time range can be based on the trigger time of the user's active behavior event to determine the time period for obtaining room text information. For example, an active behavior event may be a user giving a gift. The triggering time point of the user giving the gift can be used as a reference time, the host's voice stream data within 10 minutes before and after the reference time can be intercepted, and the speech recognition text of the voice stream data can be extracted. For another example, an active behavior event may be a user speaking and interacting on a public screen. The triggering time point of the user speaking and interacting on the public screen can be used as a reference time, the voice stream data and public screen stream data of the room within 5 minutes before and after the reference time can be intercepted, and the speech recognition text of the voice stream data and the public screen text of the public screen stream data can be extracted. Optionally, the voice stream data or public screen stream data can be input into a set sensitive word model or emotion recognition model to eliminate negative emotion data to ensure that the voice stream data and public screen stream data can effectively reflect the user's interest trends. For another example, if the user does not perform any voice or public screen interaction after entering the voice room, but stays for a period of time that meets the preset range, it can be considered that the user is very likely to be interested in the voice room. The voice stream data and public screen stream data of the voice room during the period in the room can be extracted, and the voice recognition text of the voice stream data and the public screen text of the public screen stream data can be extracted. The preset range can be used to eliminate data with too short or too long time in the room. For example, if the time in the room is as long as 2 to 3 hours, it can be regarded as an abnormal situation where the user may hang up. If the time in the room is only 1 to 2 minutes, it can be regarded as the user quickly browsing and filtering the content of interest. Of course, the aforementioned room text information acquisition setting is only an exemplary description. The selection of active behavior events and statistical time ranges can be adaptively adjusted by developers based on the online time of the voice room in the actual scenario and the actual interest prediction effect. This application does not limit this.

[0039] Step S102: Determine the user's short-term interest profile based on the room text information and the trained first text classification model.

[0040] The first text classification model can be used to generate interest tags based on the provided text content. This first text classification model can be a commonly used neural network, such as a CNN or RNN. The model can be trained using the historically recorded room text data and the corresponding interest tag data. This allows the model to have excellent interest tagging capabilities and accurately tag the acquired room text information. Optionally, the room text information can include multiple different types of text content. A trained first text classification model can be set for each type of text content, allowing the model to effectively adapt to the textual characteristics of that type of text information and accurately tag the interests of the user. For example, the room text information can be speech recognition text. A first text classification model for interest tagging speech recognition text can be trained based on the historically recorded speech recognition data and the corresponding interest tag data. The short-term interest profile can be feature information describing the real-time interests and preferences of users visiting the room. Optionally, the short-term interest profile can be generated by combining all interest tags obtained by inputting the room text information into the trained first text classification model. Alternatively, the room text information may be input into a trained first text classification model to obtain a plurality of interest tags, and a predetermined number of interest tags may be screened from high to low frequencies to obtain partial interest tags, and the partial interest tags may be sorted and combined according to their frequencies to obtain a short-term interest profile. Furthermore, the different types of text content of the room text information may be classified and input into the corresponding trained first text classification model to obtain a plurality of interest tags, and the frequency or weighted calculation of the plurality of interest tags may be performed on the classified types, and a predetermined number of interest tags from high to low frequencies or weight values ​​in each type may be selected, and the partial interest tags may be sorted and combined according to their frequencies or weight values ​​to obtain a short-term interest profile. Alternatively, the frequency or weight value of each selected interest tag may be weighted differently according to the degree of direct correlation between each type of interest tag and the user's interest, and the weighted interest tags may be sorted and combined according to their frequencies or weight values ​​to obtain a short-term interest profile. Of course, the aforementioned method of generating a short-term interest profile is merely an exemplary description, and can be adaptively adjusted by the developer according to the different needs of user interest portrayal in actual application scenarios, and this application is not limited thereto.

[0041] Step S103: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait.

[0042] The short-term interest profile generated for each visit of a user to a voice room can be recorded to obtain multiple historical short-term profiles. Obtaining the recorded historical short-term profile of a user can involve obtaining multiple historical short-term profiles within a preset time range. For example, if the preset time range is six months, multiple historical short-term profiles can be obtained within six months, in addition to the short-term interest profile generated for the user's most recent visit to the voice room. The number of historical short-term profiles obtained can be adaptively selected by the developer based on the degree to which user interests change over time in different application scenarios. This is not limited in this application. For example, a long-term interest profile may refer to the short-term interest profiles generated for 30 visits to the voice room within six months. This reference interest profile information may consist of 29 historical interest profiles and the most recently generated short-term interest profile. Since a user may frequently visit a voice room of interest, the room interest profile of the voice room of interest can be used to determine the user's long-term interest profile. The room interest profile can be feature information describing the content interest preferences of the voice room. Furthermore, the long-term interest profile can be feature information describing the user's real-time interest preferences when visiting the voice room. This long-term interest profile can be obtained by fusing the short-term interest profile, historical short-term profile, and room interest profile. Optionally, a preset number of partial tags can be extracted from all tags of short-term interest portraits, historical short-term portraits, and room interest portraits according to the frequency or weight value from high to low to obtain a long-term interest portrait. Optionally, the tag selection can also be carried out according to different types of interest portraits. For example, a preset number of first tags can be extracted from all tags of short-term interest portraits and historical short-term portraits according to the frequency or weight value from high to low, and then a preset number of second tags can be extracted from all tags of room interest portraits according to the frequency or weight value from high to low. The first and second tags are sorted and combined according to the frequency or weight value to obtain a long-term interest portrait. Optionally, the frequency or weight value of each selected interest tag can be weighted according to the degree of direct correlation between each type of interest portrait and the user's long-term interest, and then the weighted interest tags after the weight adjustment can be sorted and combined according to the frequency or weight value to obtain a long-term interest portrait. Of course, the above-mentioned method of generating long-term interest portraits is only an exemplary description, and the developer can make adaptive adjustments to the different needs of user interest portrayal according to actual application scenarios, and this application is not limited to this.

[0043] Step S104: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0044] Among them, the components of user interest portraits can include short-term interest portraits and long-term interest portraits, which comprehensively depict the user's interest changes from different statistical dimensions. Short-term interest portraits can reflect the user's short-term interest trends, and long-term interest portraits can reflect the user's long-term interest characteristics. In this way, diversified push strategies can be formulated corresponding to different types of interest portraits to improve push quality.

[0045] In the above scheme, by obtaining the room text information corresponding to the voice room that the user most recently visited, the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event; determining the user's short-term interest profile based on the room text information and the trained first text classification model; obtaining the recorded historical short-term profile of the user and the room interest portrait of the voice room that the user is following, and determining the user's long-term interest profile based on the short-term interest portrait, the historical short-term portrait, and the room interest portrait; and combining the short-term interest portrait and the long-term interest portrait to obtain the user interest profile. In the above scheme, by obtaining the room text information of the voice room that the user most recently visited, the reference information of the user's interest is effectively determined based on the triggering time point of the user's active behavior event; by utilizing the room text information that meets the multi-source information and the corresponding type of text classification model, the real-time changes of the user's interests are effectively captured, and the user's short-term interest profile is reasonably constructed; by integrating multiple recorded short-term interest portraits and the room interest portraits corresponding to the voice rooms that the user is following, the user's long-term interest characteristics are accurately grasped, and the user's long-term interest profile is effectively constructed, thereby obtaining a comprehensive and accurate user interest profile, which is conducive to recommending relevant content that meets the user's interests and customizing a personalized interactive experience.

[0046] FIG2 is a flow chart of a method for generating a user interest profile including a short-term interest profile generation process provided by an embodiment of the present application. In which, the room text information includes different types of first text content, as shown in FIG2 , including the following steps:

[0047] Step S201: Obtain room text information corresponding to the voice room most recently visited by the user, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event;

[0048] Step S202: input different types of first text contents into the trained first text classification models of corresponding types respectively to obtain a plurality of first interest tags of different types.

[0049] Among them, the room text information may include multiple types of first text content corresponding to different information dimensions in the voice room, for example, the voice recognition text corresponding to the voice stream data in the voice room, the public screen text corresponding to the public screen stream data in the voice room, and the announcement text and title text corresponding to the content introduction of the voice room. For user behavior, interest characterization can be performed with reference to multiple types of first text content respectively. For example, for user gift-giving behavior, the gift-giving interest characterization can be performed through the host's voice recognition text, for user public screen interaction behavior, the public screen interaction interest characterization can be performed through the room's voice recognition text and the public screen text, and for user room-entry behavior, the public screen interaction interest characterization can be performed through the room's voice recognition text and the public screen text. Among them, each type of first text content can be marked with interest corresponding to a trained first text classification model. For example, the voice recognition text can be input into the corresponding first text classification model to obtain the interest tag for each sentence in the voice recognition text. Thus, each type of first text content will generate a corresponding first interest tag for each text sentence or text word, and ultimately obtain multiple first interest tags of different types.

[0050] Step S203: Perform frequency statistics on each first interest tag in all first interest tags of the same type, and calculate the weight value of each first interest tag in each type based on the frequency statistics result and the interest tag information of all voice rooms obtained.

[0051] Each user behavior interest profile may correspond to multiple types of first text content. Since each type of first text content generates a corresponding first interest tag for each text sentence or text word, each type of first text content may correspond to multiple duplicate first interest tags. Frequency statistics for each first interest tag can be performed within the full set of first interest tags of the same type to eliminate duplicates while preserving the quantitative distribution characteristics of the first interest tags. Furthermore, the weight value can reflect the likelihood of each first interest tag being matched: a higher weight value indicates a greater likelihood that the first interest tag matches the user's interests, while a lower weight value indicates a lower likelihood that the first interest tag matches the user's interests. Optionally, the frequency of each first interest tag can be calculated based on the frequency statistics, and an adjustment coefficient for each type can be set based on the capacity of different types of text content. The frequency of each first interest tag and the adjustment coefficient corresponding to its type can be multiplied to obtain its weight. Optionally, the frequency of each first interest tag can be calculated based on the frequency statistics, and the inverse frequency of each first interest tag relative to the full set of interest tag information in the voice room can be calculated. The frequency of each first interest tag and the inverse frequency can be multiplied to obtain its weight. Of course, the calculation method of the aforementioned weight value is only an exemplary description, and can be adaptively adjusted by the developer according to the accuracy of interest characterization and the amount of calculation, and this application does not limit it here.

[0052] Step S204: adjust the weight value of each first interest tag to a first preset ratio, and sort and integrate the multiple first interest tags after the weight value adjustment to obtain a short-term interest portrait of the user.

[0053] Among them, the first preset ratio can be set according to the importance of the interest characterization corresponding to different user behaviors. The importance of each interest characterization can be determined by the degree of correlation between different user behaviors and interest orientations. For example, user behaviors can include user gift-giving behaviors, user public screen interaction behaviors, and user room entry behaviors. The importance ranking can be user gift-giving interests> user public screen interaction interests> user room entry interests. The corresponding first preset ratio can be set to 3:2:1. Accordingly, the weight value of the first interest tag corresponding to the user gift-giving interest can be multiplied by 3 / 6, the weight value of the first interest tag corresponding to the user public screen interaction interest can be multiplied by 2 / 6, and the weight value of the first interest tag corresponding to the user room entry interest can be multiplied by 1 / 6. The specific values ​​of the above-mentioned specific user behaviors and the first preset ratio are only exemplary descriptions. The selection of user behaviors and the first preset ratio can be adaptively adjusted by the developer according to the actual interest prediction effect, and this application does not limit it. Among them, the sorting integration can be to sort out multiple first interest tags to obtain the user's short-term interest portrait. Optionally, for the multiple first interest tags after the weight value adjustment, they can be directly sorted and combined according to their weight values ​​to obtain the user's short-term interest portrait. Optionally, a preset number of first interest tags corresponding to the interest profile of each user behavior can be selected based on weight values ​​from high to low, and the first interest tags selected corresponding to different user behaviors can be sorted and combined according to their weight values ​​to obtain the user's short-term interest profile. Of course, the aforementioned method of generating short-term interest profiles is only an example description, and the developer can adjust the integration method based on the interest profile granularity requirements of the actual application scenario, and this application does not limit it here.

[0054] Step S205: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait.

[0055] Step S206: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0056] In the above, by referring to different types of first text content to generate the first interest tag, the multi-source information of the voice room is fully utilized, the comprehensiveness and comprehensiveness of the interest tags are improved, and the generated interest tags more comprehensively reflect the characteristics of the room; by calculating the weight value of each interest tag, the reference value of different interest tags in reflecting the user's interest orientation is effectively distinguished, and a reasonable reference basis is provided for sorting out multiple interest tags to generate the user's short-term interest portrait; by adjusting the weight value of each first interest tag in proportion, the weight value can be made more in line with the importance of the interest characterization corresponding to different user behaviors, and the user's short-term interest portrait can be accurately integrated.

[0057] FIG3 is a flowchart of a method for generating a user interest profile including a weight value calculation process provided by an embodiment of the present application. As shown in FIG3 , the method includes the following steps:

[0058] Step S301: Obtain room text information corresponding to the voice room that the user visited most recently, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event.

[0059] Step S302: Different types of first text contents are respectively input into the trained first text classification models of corresponding types to obtain a plurality of first interest tags of different types.

[0060] Step S303: Perform frequency statistics on each first interest tag among all first interest tags of the same type.

[0061] Step S304: Calculate the frequency corresponding to each first interest tag based on the frequency statistics result and the total amount of each type of first interest tag.

[0062] Among them, the frequency corresponding to each first interest tag can be calculated by dividing the statistically obtained frequency by the total number of first interest tags of the same type. The frequency can be used to reflect the importance of each first interest tag among all first interest tags of the same type. As the frequency increases, the importance is higher.

[0063] Step S305: From the acquired interest tag information of all voice rooms, count the number of voice rooms where each first interest tag appears, and calculate the inverse frequency corresponding to each first interest tag based on the number and the total number of all voice rooms.

[0064] Among them, if a first interest tag appears frequently in other voice rooms, it means that the first interest tag is also applicable to other users and does not have good discrimination. Therefore, the lower the frequency of the first interest tag appearing in other voice rooms, the better. Therefore, as the frequency of the first interest tag appearing in other voice rooms increases, the inverse frequency should be lower. For example, the formula for calculating the inverse frequency corresponding to each first interest tag is as follows:

[0065] Among them, +1 can avoid the situation where the denominator appears to be 0. Of course, other subtraction functions or functions that can reflect the negative correlation between the function value and the independent variable can also be used to perform inverse frequency calculation, and this application does not limit this.

[0066] Step S306: Multiply the frequency and inverse frequency corresponding to each first interest tag to obtain a corresponding weight value.

[0067] Among them, the frequency becomes larger as the frequency of occurrence of the first interest tag in all first interest tags of the same type increases; the inverse frequency becomes smaller as the frequency of occurrence of the first interest tag in other voice rooms increases. Therefore, by multiplying the frequency by the inverse frequency, a highly distinctive first interest tag belonging to the user can be obtained to meet personalized needs.

[0068] Step S307: Adjust the weight value of each first interest tag to a first preset ratio, and sort and integrate the multiple first interest tags after the weight value adjustment to obtain a short-term interest portrait of the user.

[0069] Step S308: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait.

[0070] Step S309: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0071] As described above, by obtaining the corresponding weight value according to the frequency and inverse frequency of the first interest tag, a highly distinguishable user-specific interest tag is effectively obtained, providing the user with personalized interest portrayal and improving the personalization of the user interest portrait.

[0072] FIG4 is a flowchart of another method for generating a user interest profile including a short-term interest profile generation process provided by an embodiment of the present application. As shown in FIG4 , the method includes the following steps:

[0073] Step S401: Acquire room text information corresponding to the voice room most recently visited by the user, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event;

[0074] Step S402: Different types of first text contents are respectively input into the trained first text classification models of corresponding types to obtain a plurality of first interest tags of different types.

[0075] Step S403: Perform frequency statistics on each first interest tag in all first interest tags of the same type, and calculate the weight value of each first interest tag in each type based on the frequency statistics result and the interest tag information of all voice rooms obtained.

[0076] Step S404: Obtain the user voice text corresponding to the voice room that the user visited most recently, input the user voice text into the trained second text classification model, and obtain multiple second interest tags.

[0077] Among them, user behavior can also include user voice behavior. The user voice text can be obtained by recognizing the voice content of the user in the voice room. The user voice text can be input into the corresponding second text classification model to obtain the second interest tag for each sentence in the user voice text. The multiple second interest tags can be used to characterize the user's voice interest.

[0078] Step S405: Perform frequency statistics on each second interest tag in all second interest tags, and calculate the weight value of each second interest tag based on the frequency statistics result and the obtained interest tag information corresponding to all users.

[0079] The second text classification model generates a corresponding second interest tag for each text sentence or text word in the user's voice text. Therefore, the user's voice text may correspond to multiple duplicate second interest tags. The frequency of each second interest tag can be counted among all second interest tags of the same type to achieve the effect of deduplication while preserving the quantitative distribution characteristics of the second interest tags. In addition, the weight value can reflect the hit probability of each second interest tag. The higher the weight value, the greater the probability that the second interest tag hits the user's interest orientation, and the lower the weight value, the lower the probability that the second interest tag hits the user's interest orientation. Optionally, the frequency of each second interest tag can be calculated based on the frequency statistics results, and the frequency of each second interest tag can be used as its weight value. Optionally, the frequency of each second interest tag can be calculated based on the frequency statistics results, and the inverse frequency of each second interest tag relative to the interest tag information of the entire user can be calculated. The frequency and inverse frequency of each second interest tag can be multiplied to obtain its weight value. Of course, the aforementioned weight value calculation method is only an example description and can be adaptively adjusted by the developer based on the accuracy of interest characterization and the computational complexity. This application is not limited to this.

[0080] Step S406: adjust the weight value of each first interest tag and each second interest tag corresponding to the second preset ratio, sort and integrate the multiple first interest tags and multiple second interest tags after the weight value adjustment to obtain the user's short-term interest portrait.

[0081] Among them, the second preset ratio can be set according to the importance of the interest characterization corresponding to different user behaviors, and the importance of each interest characterization can be determined by the degree of correlation between different user behaviors and interest orientations. For example, user behaviors can include user gift-giving behavior, user public screen interaction behavior, user room entry behavior, and user voice behavior. The importance ranking can be user gift-giving interest>user voice interest>user public screen interaction interest>user room entry interest. The corresponding second preset ratio can be set to 4:3:2:1. Accordingly, the weight value of the first interest tag corresponding to the user gift-giving interest can be multiplied by 4 / 10, the weight value of the second interest tag corresponding to the user voice interest can be multiplied by 3 / 10, the weight value of the first interest tag corresponding to the user public screen interaction interest can be multiplied by 2 / 10, and the weight value of the first interest tag corresponding to the user room entry interest can be multiplied by 1 / 10. The specific values ​​of the aforementioned specific user behaviors and the second preset ratio are only exemplary descriptions. The selection of user behaviors and the second preset ratio can be adaptively adjusted by the developer according to the actual interest prediction effect, and this application does not limit them.

[0082] Step S407: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait.

[0083] Step S408: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0084] As described above, by adding the interest characterization of the user's voice behavior, the multi-source information of the voice room can be more fully and comprehensively utilized, which improves the comprehensiveness and comprehensiveness of the interest tags, so that the generated interest tags can synchronously reflect the interest characteristics of the user during the time in the room; by adjusting the weight value of each first interest tag and each second interest tag in proportion, the weight value can be made more in line with the importance of the interest characterization corresponding to different user behaviors, and the user's short-term interest portrait can be accurately integrated.

[0085] FIG5 is a flow chart of a method for generating a user interest profile including a long-term interest profile generation process provided by an embodiment of the present application. As shown in FIG5 , the method includes the following steps:

[0086] Step S501: Acquire room text information corresponding to the voice room most recently visited by the user, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event;

[0087] Step S502: Different types of first text contents are respectively input into the trained first text classification models of corresponding types to obtain a plurality of first interest tags of different types.

[0088] Step S503: Perform frequency statistics on each first interest tag in all first interest tags of the same type, and calculate the weight value of each first interest tag in each type based on the frequency statistics result and the interest tag information of all voice rooms obtained.

[0089] Step S504: adjust the weight value of each first interest tag to a first preset ratio, and sort and integrate the multiple first interest tags after the weight value adjustment to obtain a short-term interest portrait of the user.

[0090] Step S505: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest.

[0091] Step S506: extract a first preset number of interest tags from the short-term interest portrait and the historical short-term portrait according to the weight values ​​from high to low.

[0092] Among them, the first preset number of interest tags are extracted in order from high to low according to the weight value, and the first few interest tags with high importance in the short-term interest portrait can be selected, and redundant interest tags with low importance can be removed, so as to reasonably simplify the number of interest tags in the short-term interest portrait.

[0093] Step S507: extract a second preset number of interest tags from the room interest portrait in descending order of weight values.

[0094] Among them, a second preset number of interest tags are extracted in sequence from high to low according to the weight value, and the first few interest tags with high importance in the room interest portrait can be selected, and redundant interest tags with low importance can be removed to reasonably simplify the number of interest tags in the room interest portrait.

[0095] Step S508: Adjust the weight values ​​of the extracted interest tags corresponding to the third preset ratio, and sort and integrate the multiple interest tags after the weight value adjustment to obtain the user's long-term interest portrait.

[0096] Among them, the third preset ratio can be set according to the importance of interest portraits in different dimensions. Since the short-term interest portrait is related to the user's behavior in the room, and the room interest portrait is only related to the voice room that the user sets to pay attention to, the user may not enter the relevant voice room. Therefore, multiple short-term interest portraits within the statistical time range are closer to the user's interest orientation than the room interest portrait. Therefore, for example, the importance ranking can be short-term interest portrait > room interest portrait, and the corresponding third preset ratio can be set to 4:1. Accordingly, the weight value of the interest tag corresponding to the short-term interest portrait can be multiplied by 4 / 5, and the weight value of the interest tag corresponding to the room interest portrait can be multiplied by 1 / 5. The specific value of the aforementioned third preset ratio is only an exemplary description. The selection of the third preset ratio can be adaptively adjusted by the developer according to the actual interest prediction effect, and this application does not limit it here.

[0097] Step S509: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0098] As mentioned above, by combining short-term interest portraits and room interest portraits of the voice rooms that users pay attention to, we can not only effectively grasp the user's current interests, but also consider the characteristics of the user's long-term interests, and achieve a comprehensive portrayal of the user's interest orientation, which is conducive to improving the accuracy of personalized recommendations and user satisfaction.

[0099] FIG6 is a flowchart of a method for generating a user interest profile including a room interest profile generation process provided by an embodiment of the present application. As shown in FIG6 , the method includes the following steps:

[0100] Step S601: Acquire room text information corresponding to the voice room that the user visited most recently, wherein the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event.

[0101] Step S602: Determine the user's short-term interest profile based on the room text information and the trained first text classification model.

[0102] Step S603: Acquire the room text information of the voice room that the user is following when it was last broadcast. The room text information includes different types of second text content.

[0103] Among them, the room text information may include multiple types of second text content corresponding to different information dimensions in the voice room, and the room text information may include the voice recognition text, public screen text, announcement text and title text corresponding to the most recent broadcast.

[0104] Step S604: input the second text content of different types into the trained first text classification model of the corresponding type to obtain a plurality of third interest tags of different types.

[0105] Each type of second text content can be assigned an interest tag using a trained first text classification model. For example, a speech recognition text can be input into the corresponding first text classification model to obtain an interest tag for each sentence in the speech recognition text. Thus, each type of second text content will generate a corresponding third interest tag for each text sentence or text word, ultimately resulting in multiple third interest tags of different types.

[0106] Step S605: Perform frequency statistics on each third interest tag in all third interest tags of the same type, and calculate the weight value of each third interest tag in each type based on the frequency statistics results and the interest tag information of all voice rooms obtained.

[0107] Among them, each type of second text content will generate a corresponding third interest tag for each text sentence or text word, so each type of second text content may correspond to multiple repeated third interest tags. The frequency statistics of each third interest tag can be performed in the full amount of third interest tags of the same type to achieve the effect of deduplication while retaining the quantitative distribution characteristics of the third interest tags. In addition, the weight value can reflect the hit probability of each third interest tag. The higher the weight value, the greater the possibility that the third interest tag hits the interest topic of the voice room, and the lower the weight value, the smaller the possibility that the third interest tag hits the interest topic of the voice room.

[0108] Step S606: Adjust the weight value of each third interest tag corresponding to the fourth preset ratio, and sort and integrate the multiple third interest tags after the weight value adjustment to obtain a room interest portrait of the voice room.

[0109] Among them, the fourth preset ratio can be set according to the importance of different types of second text content, and the importance of the second text content can be determined according to the degree of correlation between the user's behavioral habits of visiting the voice room and the interest orientation. For example, if the room text information includes voice recognition text, public screen text, title text and announcement text, then the importance ranking can be title text and announcement text> voice recognition text> public screen text, and the corresponding fourth preset ratio can be set to 5:3:1. Accordingly, the weight value of the third interest tag corresponding to the title text and announcement text can be multiplied by 5 / 9, the weight value of the third interest tag corresponding to the voice recognition text can be multiplied by 3 / 9, and the weight value of the third interest tag corresponding to the public screen text can be multiplied by 1 / 9. The specific values ​​of the aforementioned specific voice room information and the fourth preset ratio are only exemplary descriptions. The selection of the voice room information and the fourth preset ratio can be adaptively adjusted by the developer according to the actual interest prediction effect, and this application does not limit it.

[0110] Step S607: Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait.

[0111] Step S608: Combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0112] As described above, by using the room text information when the voice room was last broadcast to generate the room's third interest tag, real-time monitoring of interest topics in the voice room is achieved, and the user's interest orientation is captured by the interest characteristics of the voice room of interest, thereby improving the flexibility of user interest portrayal and providing an effective interest tag reference for the user's long-term interest portrait.

[0113] Figure 7 is a block diagram of a user interest profile generation device provided in an embodiment of the present application. The device is configured to execute the user interest profile generation method provided in the above embodiment and has the corresponding functional modules and beneficial effects of the execution method. As shown in Figure 7, the device includes:

[0114] The acquisition module 101 is configured to obtain the room text information corresponding to the voice room that the user visited most recently, where the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event;

[0115] The short-term profile generation module 102 is configured to determine the user's short-term interest profile based on the room text information and the trained first text classification model;

[0116] The long-term profile generation module 103 is configured to obtain the recorded historical short-term profiles of the user and the room interest profiles of the voice room of interest, and determine the long-term interest profile of the user based on the short-term interest profiles, the historical short-term profiles and the room interest profiles;

[0117] The user portrait generation module 104 is configured to combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

[0118] In the above scheme, by obtaining the room text information corresponding to the voice room that the user most recently visited, the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event; determining the user's short-term interest profile based on the room text information and the trained first text classification model; obtaining the recorded historical short-term profile of the user and the room interest portrait of the voice room that the user is following, and determining the user's long-term interest profile based on the short-term interest portrait, the historical short-term portrait, and the room interest portrait; and combining the short-term interest portrait and the long-term interest portrait to obtain the user interest profile. In the above scheme, by obtaining the room text information of the voice room that the user most recently visited, the reference information of the user's interest is effectively determined based on the triggering time point of the user's active behavior event; by utilizing the room text information that meets the multi-source information and the corresponding type of text classification model, the real-time changes of the user's interests are effectively captured, and the user's short-term interest profile is reasonably constructed; by integrating multiple recorded short-term interest portraits and the room interest portraits corresponding to the voice rooms that the user is following, the user's long-term interest characteristics are accurately grasped, and the user's long-term interest profile is effectively constructed, thereby obtaining a comprehensive and accurate user interest profile, which is conducive to recommending relevant content that meets the user's interests and customizing a personalized interactive experience.

[0119] In a possible embodiment, the short-term portrait generation module 102 is further configured to:

[0120] Inputting different types of first text content into the trained first text classification model of the corresponding type, respectively, to obtain a plurality of first interest tags of different types;

[0121] Perform frequency statistics on each first interest tag in all first interest tags of the same type, and calculate the weight value of each first interest tag in each type based on the frequency statistics result and the obtained interest tag information of all voice rooms;

[0122] A weight value corresponding to a first preset ratio is adjusted for each first interest tag, and the plurality of first interest tags after the weight value adjustment are sorted and integrated to obtain a short-term interest portrait of the user.

[0123] In a possible embodiment, the short-term portrait generation module 102 is further configured to:

[0124] Calculate the frequency of each first interest tag based on the frequency statistics and the total amount of each type of first interest tag;

[0125] From the obtained interest tag information of all voice rooms, count the number of voice rooms where each first interest tag appears, and calculate the inverse frequency corresponding to each first interest tag based on the number and the total number of all voice rooms;

[0126] The frequency and inverse frequency corresponding to each first interest tag are multiplied to obtain a corresponding weight value.

[0127] In a possible embodiment, the second short-term portrait generation module is further included, configured to:

[0128] Obtaining the user voice text corresponding to the voice room that the user visited most recently, inputting the user voice text into the trained second text classification model to obtain multiple second interest tags;

[0129] Performing frequency statistics on each second interest tag in all second interest tags, and calculating a weight value of each second interest tag based on the frequency statistics result and the obtained interest tag information corresponding to all users;

[0130] The weight values ​​of each first interest tag and each second interest tag are adjusted to correspond to the second preset ratio, and the multiple first interest tags and multiple second interest tags after the weight values ​​are adjusted are sorted and integrated to obtain the user's short-term interest portrait.

[0131] In one possible embodiment, the long-term portrait generation module is further configured to:

[0132] Extracting a first preset number of interest tags from the short-term interest portrait and the historical short-term portrait in descending order of weight values;

[0133] Extracting a second preset number of interest tags from the room interest portrait in descending order of weight values;

[0134] The extracted interest tags are weighted according to a third preset ratio, and the plurality of interest tags after the weight adjustment are sorted and integrated to obtain a long-term interest portrait of the user.

[0135] In a possible embodiment, a room portrait generation module is further included, configured to:

[0136] Obtain the room text information of the voice room that the user is following during the most recent broadcast, where the room text information includes different types of second text content;

[0137] Inputting different types of second text content into the trained first text classification model of the corresponding type to obtain a plurality of third interest tags of different types;

[0138] Perform frequency statistics on each third interest tag in all third interest tags of the same type, and calculate the weight value of each third interest tag in each type based on the frequency statistics and the interest tag information obtained from all voice rooms;

[0139] A weight value corresponding to a fourth preset ratio is adjusted for each third interest tag, and the multiple third interest tags after the weight value adjustment are sorted and integrated to obtain a room interest portrait of the voice room.

[0140] Figure 8 is a schematic diagram of the structure of a user interest profile generation device provided in an embodiment of the present application. As shown in Figure 8, the device includes a processor 201, a memory 202, an input device 203, and an output device 204. The device may contain one or more processors 201, with one processor 201 being used as an example in Figure 8. The processor 201, memory 202, input device 203, and output device 204 in the device may be connected via a bus or other means, with bus connection being used as an example in Figure 8. Memory 202, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the user interest profile generation method in the embodiment of the present application. Processor 201 executes the software programs, instructions, and modules stored in memory 202 to execute various functional applications and data processing of the device, thereby implementing the user interest profile generation method described above. Input device 203 can be configured to receive input digital or character information and generate key signal input related to user settings and function control of the device. Output device 204 may include a display device such as a display screen.

[0141] An embodiment of the present application also provides a non-volatile storage medium containing computer-executable instructions, which, when executed by a computer processor, is configured to execute a user interest portrait generation method described in the above embodiment, which includes: obtaining room text information corresponding to the voice room that the user last visited, and the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event; determining the user's short-term interest portrait based on the room text information and the trained first text classification model; obtaining the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determining the user's long-term interest portrait based on the short-term interest portrait, the historical short-term portrait and the room interest portrait; combining the short-term interest portrait and the long-term interest portrait to obtain the user's interest portrait.

[0142] It is worth noting that in the embodiment of the above-mentioned user interest portrait generation device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the various functional units are only for the convenience of distinguishing each other, and are not configured to limit the scope of protection of the embodiments of this application.

[0143] In some possible implementations, various aspects of the methods provided herein may also be implemented in the form of a program product, which includes program code. When the program product is executed on a computer device, the program code is configured to cause the computer device to execute the steps of the methods according to the various exemplary embodiments of the present application described above in this specification. For example, the computer device may execute the user interest profile generation method described in the embodiments of the present application. The program product may be implemented using any combination of one or more readable media.

Claims

1. A method for generating a user interest profile, wherein: include: Obtaining room text information corresponding to the voice room most recently visited by the user, where the statistical time range of the room text information is determined based on the triggering time point of the user's active behavior event; Determining the user's short-term interest profile based on the room text information and the trained first text classification model; Obtain the recorded historical short-term portrait of the user and the room interest portrait of the voice room of interest, and determine the long-term interest portrait of the user based on the short-term interest portrait, the historical short-term portrait and the room interest portrait; The short-term interest portrait and the long-term interest portrait are combined to obtain a user interest portrait.

2. The method for generating a user interest profile according to claim 1, wherein: The room text information includes different types of first text content, and determining the user's short-term interest profile based on the room text information and the trained first text classification model includes: Inputting the different types of first text content into the trained first text classification model of the corresponding type respectively to obtain a plurality of first interest tags of different types; Performing frequency statistics on each of the first interest tags in all first interest tags of the same type, and calculating a weight value of each of the first interest tags in each type based on the frequency statistics and the obtained interest tag information of all voice rooms; A weight value corresponding to a first preset ratio is adjusted for each of the first interest tags, and the plurality of first interest tags after the weight value adjustment are sorted and integrated to obtain a short-term interest portrait of the user.

3. The method for generating a user interest profile according to claim 2, wherein: The calculating, based on the frequency statistics and the acquired interest tag information of all voice rooms, a weight value of each first interest tag in each type includes: Calculate the frequency corresponding to each first interest tag based on the frequency statistics result and the total amount of each type of first interest tag; Counting the number of voice rooms in which each of the first interest tags appears from the acquired interest tag information of all voice rooms, and calculating the inverse frequency corresponding to each of the first interest tags based on the number and the total number of all voice rooms; The frequency and inverse frequency corresponding to each of the first interest tags are multiplied to obtain a corresponding weight value.

4. The method for generating a user interest profile according to claim 2 or 3, wherein: The method further comprises: Obtaining a user voice text corresponding to a voice room that the user most recently visited, and inputting the user voice text into a trained second text classification model to obtain a plurality of second interest tags; Performing frequency statistics on each second interest tag in all second interest tags, and calculating a weight value of each second interest tag based on the frequency statistics result and the obtained interest tag information corresponding to all users; The weight value of each first interest tag and each second interest tag is adjusted to correspond to a second preset ratio, and the multiple first interest tags and multiple second interest tags after the weight value adjustment are sorted and integrated to obtain the short-term interest portrait of the user.

5. The method for generating a user interest profile according to claim 2 or 3, wherein: The determining the long-term interest profile of the user according to the short-term interest profile, the historical short-term profile, and the room interest profile includes: Extracting a first preset number of interest tags from the short-term interest portrait and the historical short-term portrait in descending order of weight values; Extracting a second preset number of interest tags from the room interest portrait in descending order of weight values; The extracted interest tags are weighted according to a third preset ratio, and the plurality of interest tags after the weight adjustment are sorted and integrated to obtain a long-term interest portrait of the user.

6. The method for generating a user interest profile according to any one of claims 1 to 5, wherein: Before obtaining the recorded historical short-term portrait of the user and the room interest portrait of the concerned voice room, the method further includes: Obtaining room text information of the voice room that the user is following during the most recent broadcast, wherein the room text information includes different types of second text content; Inputting the different types of second text content into a trained first text classification model of the corresponding type to obtain a plurality of third interest tags of different types; Performing frequency statistics on each of the third interest tags in all third interest tags of the same type, and calculating a weight value of each of the third interest tags in each type based on the frequency statistics result and the obtained interest tag information of all voice rooms; A weight value corresponding to a fourth preset ratio is adjusted for each of the third interest tags, and the plurality of third interest tags after the weight value adjustment are sorted and integrated to obtain a room interest portrait of the voice room.

7. A user interest profile generation device, wherein: include: An acquisition module is configured to acquire room text information corresponding to a voice room that a user has visited most recently, wherein a statistical time range of the room text information is determined based on a triggering time point of an active behavior event of the user; a short-term portrait generation module, configured to determine the user's short-term interest portrait based on the room text information and the trained first text classification model; a long-term portrait generation module configured to obtain the recorded historical short-term portraits of the user and the room interest portraits of the voice rooms of interest, and determine the long-term interest portrait of the user based on the short-term interest portraits, the historical short-term portraits, and the room interest portraits; The user portrait generation module is configured to combine the short-term interest portrait and the long-term interest portrait to obtain a user interest portrait.

8. A device for generating a user interest profile, the device comprising: one or more processors; A storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the user interest portrait generation method described in any one of claims 1-6.

9. A non-volatile storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a computer processor, are configured to execute the user interest portrait generation method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, wherein When the computer program is executed by a processor, the method for generating a user interest portrait according to any one of claims 1 to 6 is implemented.

Images