Virtual scene recommendation method and device, storage medium and electronic equipment

Abstract

The invention discloses a virtual scene recommendation method and device, a storage medium and electronic equipment. Relates to the field of artificial intelligence and the field of meta universe, and comprises the following steps: obtaining input data of N modals of a target account, the N modals being composition modals required by a virtual scene, and N being a positive integer greater than or equal to 2; based on the input data of the N modalities, determining fusion features of the N modalities; and determining a target virtual scene recommended to the target account in M candidate virtual scenes based on the fusion feature, M being a positive integer greater than N. By means of the method and device, the problem that the accuracy of the user virtual scene recommendation result is not ideal in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and metaverse, and specifically, to a virtual scene recommendation method, device, storage medium and electronic device. Background Art

[0002] With the development of digital technology, virtual scenes are increasingly used in the real world, such as online games, e-commerce, virtual reality applications, social networks, virtual offices, virtual meetings, etc. The user experience of these virtual scenes is obtained by users in specific situations. Therefore, how to accurately understand the diverse needs of users and map them to appropriate virtual scenes, and then provide users with personalized virtual scene recommendations to improve user experience, is one of the goals of virtual scene recommendation. In related technologies, virtual scene recommendations often rely on manual rules or simple statistical models and a single scoring model, which is difficult to handle complex and changeable user inputs and dynamically changing scene requirements, resulting in poor diversity of recommended virtual scenes, inability to achieve personalized recommendations for users on virtual scenes, and the problem of unsatisfactory accuracy of user virtual scene recommendation results.

[0003] With regard to the problem that the accuracy of user virtual scene recommendation results in related technologies is not ideal, no effective solution has been proposed so far. Summary of the invention

[0004] The main purpose of the present application is to provide a virtual scene recommendation method, device, storage medium and electronic device to solve the problem of unsatisfactory accuracy of user virtual scene recommendation results in the related art.

[0005] In order to achieve the above purpose, according to one aspect of the present application, a virtual scene recommendation method is provided. The method includes: obtaining input data of N modes of a target account, wherein the N modes are respectively the component modes required for the virtual scene, and N is a positive integer greater than or equal to 2; based on the input data of the N modes, determining the fusion features of the N modes; based on the fusion features, determining the target virtual scene recommended to the target account from among the M candidate virtual scenes, wherein M is a positive integer greater than N.

[0006] Optionally, based on the fusion features, a target virtual scene recommended to the target account is determined among M candidate virtual scenes, including: determining the scene scores corresponding to the M candidate virtual scenes respectively according to the fusion features; when the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold, the candidate virtual scene corresponding to the highest score is used as the target virtual scene.

[0007] Optionally, the method also includes: determining the login time of the target account; determining the scene sets corresponding to P time periods, wherein the scenes included in the scene sets corresponding to the P time periods are different, and P is a positive integer; determining the target time period corresponding to the login time among the P time periods; and obtaining M candidate virtual scenes based on the scene sets corresponding to the target time period.

[0008] Optionally, the method further includes: when the target account responds to the target virtual scene, obtaining interaction information of the target account in the target virtual scene; based on the interaction information, determining recommended correction information, wherein the recommended correction information is used to correct the virtual scene recommended to the target account.

[0009] Optionally, based on the fusion features, a target virtual scene recommended to the target account is determined among M candidate virtual scenes, including: in a case where there is historical usage data of the target account in the M candidate virtual scenes, obtaining the scene resource requirements corresponding to the M candidate virtual scenes respectively; determining the historical loading times corresponding to the M candidate virtual scenes respectively based on the historical usage data corresponding to the M candidate virtual scenes respectively; obtaining the delay tolerance of the target account; and determining the target virtual scene based on the fusion features, the delay tolerance, and the scene resource requirements and historical loading times corresponding to the M candidate virtual scenes respectively.

[0010] Optionally, based on the input data of N modalities, the fusion features of the N modalities are determined, including: determining feature extraction models corresponding to the N modalities, wherein the feature extraction models corresponding to the N modalities have different feature extraction capabilities in different modalities; using the feature extraction models corresponding to the N modalities, performing feature extraction based on the input data of the N modalities, and obtaining single modal features corresponding to the N modalities; performing feature fusion based on the single modal features corresponding to the N modalities, and obtaining fused features.

[0011] Optionally, feature fusion is performed based on the unimodal features corresponding to the N modalities to obtain fused features, including: determining the active frequencies corresponding to the N modalities based on the historical behavior data of the target account; determining the fusion weights corresponding to the N modalities according to the active frequencies corresponding to the N modalities; and performing feature fusion based on the fusion weights corresponding to the N modalities and the unimodal features to obtain fused features.

[0012] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a virtual scene recommendation device is provided, which includes: an input data acquisition module, used to obtain input data of N modalities of a target account, wherein the N modalities are respectively the component modalities required for the virtual scene, and N is a positive integer greater than or equal to 2; a fusion feature determination module, used to determine the fusion features of the N modalities based on the input data of the N modalities; a target virtual scene determination module, used to determine the target virtual scene recommended to the target account from among M candidate virtual scenes based on the fusion features, wherein M is a positive integer greater than N.

[0013] Optionally, the target virtual scene determination module includes: a scene score determination module, used to determine the scene scores corresponding to M candidate virtual scenes according to the fusion features; a first scene determination module, used to select the candidate virtual scene corresponding to the highest score as the target virtual scene when the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold.

[0014] Optionally, the device also includes: a login time determination module, used to determine the login time of the target account; a scene set determination module, used to determine the scene sets corresponding to P time periods, wherein the scenes included in the scene sets corresponding to the P time periods are different, and P is a positive integer; a target time period determination module, used to determine the target time period corresponding to the login time in the P time periods; and a candidate scene determination module, used to obtain M candidate virtual scenes based on the scene sets corresponding to the target time periods.

[0015] Optionally, the device also includes: an interaction information determination module, used to obtain the interaction information of the target account in the target virtual scene when the target account responds to the target virtual scene; a correction information determination module, used to determine recommended correction information based on the interaction information, wherein the recommended correction information is used to correct the virtual scene recommended to the target account.

[0016] Optionally, the target virtual scene determination module includes: a resource requirement determination module, which is used to obtain the scene resource requirements corresponding to the M candidate virtual scenes respectively when there is historical usage data of the target account in the M candidate virtual scenes; a loading time determination module, which is used to determine the historical loading times corresponding to the M candidate virtual scenes respectively based on the historical usage data corresponding to the M candidate virtual scenes respectively; a delay tolerance acquisition module, which is used to obtain the delay tolerance of the target account; and a second scene determination module, which is used to determine the target virtual scene based on the fusion features, delay tolerance, and the scene resource requirements and historical loading times corresponding to the M candidate virtual scenes respectively.

[0017] Optionally, the fusion feature determination module includes: a feature extraction model determination module, used to determine the feature extraction models corresponding to N modalities, wherein the feature extraction models corresponding to the N modalities have different feature extraction capabilities in different modalities; a single modal feature determination module, used to use the feature extraction models corresponding to the N modalities, perform feature extraction based on the input data of the N modalities, and obtain single modal features corresponding to the N modalities; a first feature determination module, used to perform feature fusion based on the single modal features corresponding to the N modalities, to obtain a fused feature.

[0018] Optionally, the first feature determination module includes: an active frequency determination module, which is used to determine the active frequencies corresponding to N modalities based on the historical behavior data of the target account; a fusion weight determination module, which is used to determine the fusion weights corresponding to the N modalities according to the active frequencies corresponding to the N modalities; and a second feature determination module, which is used to perform feature fusion based on the fusion weights corresponding to the N modalities and the single modal features to obtain a fused feature.

[0019] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the virtual scene recommendation methods.

[0020] In order to achieve the above-mentioned purpose, according to another aspect of the present application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein any one of the virtual scene recommendation methods is executed when the program is running.

[0021] In the embodiment of the present application, a multi-source heterogeneous data fusion method is adopted to extract and fuse the user's multimodal preference data to achieve the purpose of capturing the mapping relationship between the user input and the virtual scene, thereby achieving the technical effect of providing the user with more accurate and personalized virtual scene recommendations based on the user's multimodal preference input data, thereby solving the technical problem of unsatisfactory accuracy of user virtual scene recommendation results caused by complex and changeable user input and dynamically changing scene requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0023] Figure 1 A hardware structure block diagram of a computer terminal for implementing a virtual scene recommendation method is shown;

[0024] Figure 2 is a flowchart of a virtual scene recommendation method provided according to an embodiment of the present application;

[0025] Figure 3 is a schematic diagram of a virtual scene recommendation device provided according to an embodiment of the present application;

[0026] Figure 4 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0030] NLP (Natural Language Processing) is an important field of artificial intelligence that enables computers to understand and generate human natural language. NLP technology combines statistics, machine learning, deep learning, and theories of computer science and linguistics to parse, understand, interpret, generate, and manipulate text or voice data.

[0031] The BERT (Bidirectional Encoder Representations from Transformers) model is a deep learning model for natural language processing (NLP) tasks. The model uses the Transformer architecture pre-training model. By introducing a bidirectional encoding mechanism and using unsupervised learning, it can pre-train a large amount of text data, thereby generating word embedding representations with rich contextual information.

[0032] MLP (Multilayer Perceptron) is an artificial neural network model used to handle complex classification and regression problems. MLP consists of an input layer, an output layer, and one or more hidden layers. Each layer consists of multiple neurons or nodes. The neurons in each layer are fully connected to the neurons in the next layer to form a network structure.

[0033] One-Hot Encoding is a data preprocessing technique that is mainly used to convert categorical variables or label data into numerical vector representations that can be used by machine learning models. In one-hot encoding, each category is represented as a binary vector whose length is equal to the number of categories, where only the position corresponding to the category is 1 and the rest are 0.

[0034] Fusion Neural Network refers to a network architecture that integrates features or information flows from multiple different sources into a unified model in machine learning and deep learning. This network design is usually used to process multimodal data, that is, to fuse multiple types of data such as images, text, audio, etc. for analysis and prediction, or to fuse multiple different features in a single modality, such as fusing user behavior data, text descriptions, and preference information in a recommendation system.

[0035] It should be noted that the collected information involved in this application (including but not limited to user device information, user personal information, user conversation content preference information, user clothing selection preference information, user music preference information, candidate virtual scene set and user delay tolerance, etc.) and data (including but not limited to data for display, analysis data, user interaction information data, user scene resource demand data, user historical loading time data, etc.) are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0036] Example 1

[0037] According to an embodiment of the present application, a method embodiment of a virtual scene recommendation method is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a virtual scene recommendation method. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0039] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0040] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the virtual scene recommendation method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned virtual scene recommendation method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0041] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0042] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0043] Under the above operating environment, this application provides Figure 2 The recommended method for the virtual scene shown. Figure 2 It is a flowchart of a virtual scene recommendation method provided according to an embodiment of the present application.

[0044] Step S201, obtaining input data of N modes of the target account, wherein the N modes are component modes required for the virtual scene, and N is a positive integer greater than or equal to 2;

[0045] It can be understood that the component modalities required to determine the virtual scene, such as conversation content, clothing selection, music preference, etc., each modality reflects the user's needs or interests in different aspects. Obtain the input data of N modalities of the target account. By obtaining and integrating the input data of N modalities, the performance of the virtual scene recommendation system can be significantly improved, making it more in line with the user's real needs and interests, thereby providing a more personalized and immersive user experience in the fields of virtual reality (VR) and augmented reality (AR).

[0046] Step S202, determining fusion features of the N modalities based on the input data of the N modalities;

[0047] It can be understood that the input data of N modes of the target account is obtained and preprocessed, and the preprocessed input data of N modes of the target account is input into the fusion model to obtain the fusion features of N modes, wherein the above fusion model is used to integrate the features of N modes. By fusing the features of N modes, the user's interest preferences can be more comprehensively understood, and the accuracy and personalization of virtual scene recommendations can be improved.

[0048] Optionally, for conversation content data, NLP technology is used for text preprocessing, including word segmentation, stop word removal, word embedding, etc. For clothing selection data, the preprocessing method is to encode clothing features (such as category, color, style) into one-hot encoding or embedding vectors (if the number of categories is small, one-hot encoding is used; if the number of categories is large, an embedding layer is used). For music preference data, similar to clothing selection, the preprocessing method is to encode music features (such as type, composer, genre) into one-hot encoding or embedding vectors.

[0049] Optionally, in the virtual scene recommendation method provided in the embodiment of the present application, based on the input data of N modalities, the fusion features of N modalities are determined, including: determining the feature extraction models corresponding to the N modalities, wherein the feature extraction models corresponding to the N modalities have different feature extraction capabilities in different modalities; using the feature extraction models corresponding to the N modalities, performing feature extraction based on the input data of the N modalities, and obtaining single modal features corresponding to the N modalities; performing feature fusion based on the single modal features corresponding to the N modalities, and obtaining fused features.

[0050] It can be understood that for each of the N modalities, based on the expressiveness and characteristics of the modal data, a feature extraction model suitable for the modal data type is selected for each modality. The feature extraction model corresponding to each of the above modalities is used to extract features from the preprocessed input data of each modality to obtain the single modal features corresponding to the N modalities. Based on the single modal features corresponding to the above N modalities, a fusion model is used to perform feature fusion to obtain fused features. By using feature extraction models designed specifically for each modality, complex information about user interests and preferences can be captured from different angles, making the generated fused feature vector richer and more comprehensive, and able to more accurately reflect the user's real needs.

[0051] Optionally, for the conversation content data of the target account, the BERT model can be used as a feature extraction model to convert the preprocessed conversation content into a fixed-length vector representation. Then the vector output by BERT is used as input, and the output of BERT is mapped to the classification space of the virtual scene through a fully connected layer, and the relevance of the conversation content to the virtual scene or the classification label (i.e., unimodal feature) is output.

[0052] Optionally, select and load a pre-trained BERT model as a feature extractor. Input the pre-processed conversation content into the BERT model, and extract the embedded representation of the CLS tag (Class Token, a special tag used to summarize the information of the entire sequence in sequence classification tasks for the final classification decision) as a fixed-length vector of the conversation. Add one or more fully connected layers after the BERT output, map the BERT output to the classification space of the virtual scene, and output the relevance or classification label of the conversation content to the virtual scene. In the training phase, use the annotated data of the conversation content to train the conversation content model.

[0053] Optionally, for the clothing selection data of the target account, an MLP model can be used as a feature extraction model. The music preference model includes an MLP layer, a hidden layer, and an output layer. The preprocessed clothing selection data is input into the MLP model. The MLP layer is used to process the embedding vector of the clothing selection, and the feature transformation is performed through the hidden layer to obtain the correlation between the clothing selection and the virtual scene. The output layer is used to output the correlation between the clothing and the virtual scene (i.e., unimodal features). In the training stage, the clothing selection model is trained using the labeled data of the clothing and the virtual scene.

[0054] Optionally, for the music preference data of the target account, an MLP model can be used as a feature extraction model. The music preference model includes an MLP layer, a hidden layer, and an output layer. The preprocessed music preference data is input into the MLP model. The MLP layer is used to process the embedding vector of the music preference, and the feature transformation is performed through the hidden layer to obtain the correlation between the music preference and the virtual scene. The output layer is used to output the correlation between the music and the virtual scene (i.e., unimodal features). In the training stage, the music preference model is trained using the labeled data of the music preference and the virtual scene.

[0055] Optionally, after obtaining the unimodal features corresponding to the conversation content data, clothing selection data, and music preference data, respectively, the fusion model is used to obtain the fusion features of the target account. A fusion neural network is designed, whose input layer receives the outputs of the conversation content model, clothing selection model, and music preference model, that is, the unimodal features corresponding to the conversation content data, clothing selection data, and music preference data, respectively. The fusion neural network adopts a multi-layer fully connected network structure, combines these inputs through nonlinear transformations and activation functions, and finally outputs the virtual scene area that the user may be interested in.

[0056] Optionally, the fusion neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the output of the dialogue content model, the clothing selection model, and the music preference model as input. The hidden layer is used to add multiple fully connected layers and use nonlinear activation functions such as ReLU (Rectified Linear Unit) to perform feature fusion and transformation. The output layer is used to output the scene score of the virtual scene area that the user may be interested in. In the training stage, the fusion neural network is trained using a comprehensive data set containing all model outputs to learn the interactive relationship between different features.

[0057] Optionally, in the virtual scene recommendation method provided in the embodiment of the present application, feature fusion is performed based on the single modal features corresponding to the N modalities to obtain a fused feature, including: determining the active frequencies corresponding to the N modalities based on the historical behavior data of the target account; determining the fusion weights corresponding to the N modalities according to the active frequencies corresponding to the N modalities; and performing feature fusion based on the fusion weights corresponding to the N modalities and the single modal features to obtain a fused feature.

[0058] It can be understood that, first, the usage records of N modalities are extracted from the historical behavior data of the target account to obtain the active frequencies corresponding to the N modalities. The active frequency is used to indicate the ratio of the number of operations or time of the target account on a certain modality to the total number of operations or total time. Secondly, according to the active frequencies corresponding to the N modalities, the fusion weights corresponding to the N modalities are determined. The fusion weight reflects the degree of influence of each modal feature on user interest and satisfaction under the current behavior mode of the target account. Therefore, the higher the active frequency of the modality, the greater its corresponding fusion weight, so as to ensure that in the feature fusion process, the more frequently used modality contributes more significantly to the final fusion feature. Based on the fusion weights corresponding to the N modalities and the single modal features, the features are fused to obtain the fusion features. By using the active frequency in the historical behavior data to determine the fusion weight, it is ensured that the recommendation model can more accurately reflect the user's preference strength on different modalities, thereby improving the personalization and accuracy of the recommendation, reducing the recommendation error rate, and enhancing the user experience.

[0059] Step S203: Based on the fusion features, determine the target virtual scene recommended to the target account from among the M candidate virtual scenes, where M is a positive integer greater than N.

[0060] It can be understood that based on the preprocessed input data of the N modalities of the target account, the fusion model is used to obtain the fusion features of the N modalities. According to the fusion features, the target virtual scene recommended to the target account is determined from the M candidate virtual scenes. The strategy of determining the target virtual scene from the M candidate virtual scenes based on the fusion features can effectively improve the personalization and diversity of the recommendation system, optimize the user experience, and also improve the efficiency and accuracy of the recommendation system, and promote the accurate matching between virtual scenes and user interests.

[0061] Optionally, in the virtual scene recommendation method provided in the embodiment of the present application, based on the fusion features, a target virtual scene recommended to the target account is determined among M candidate virtual scenes, including: determining the scene scores corresponding to the M candidate virtual scenes respectively according to the fusion features; when the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold, the candidate virtual scene corresponding to the highest score is used as the target virtual scene.

[0062] It can be understood that based on the fusion features of the N modalities of the target account, the scene scores corresponding to the M candidate virtual scenes are obtained. If the highest score among the above M scene scores is greater than the predetermined score threshold, the candidate virtual scene corresponding to the highest score is recommended to the target account as the target virtual scene; if the highest score among the above M scene scores is less than or equal to the predetermined score threshold, the virtual scene is not recommended to the target account. By setting a predetermined score threshold, it is ensured that the recommended scene not only performs best among the candidates, but also reaches a standard that is considered to be attractive enough to users, thereby improving the confidence of the recommendation results and user satisfaction.

[0063] Optionally, in the virtual scene recommendation method provided in the embodiment of the present application, the method also includes: determining the login time of the target account; determining the scene sets corresponding to P time periods respectively, wherein the scenes included in the scene sets corresponding to the P time periods are different, and P is a positive integer; determining the target time period corresponding to the login time among the P time periods; and obtaining M candidate virtual scenes based on the scene sets corresponding to the target time period.

[0064] It can be understood that the login time of the target account is determined, the time of a day is divided into P time periods, and the scene sets corresponding to the P time periods are determined according to the needs and interests of the target account in different time periods. Since the needs and interests of the target account in different time periods may be different, the scenes included in the specific scene set corresponding to each time period are different. According to the login time of the target account, the time period corresponding to the above login time in the P time periods is determined, that is, the same or closest time period, and the corresponding time period is used as the target time period. According to the scene set corresponding to the target time period, M candidate virtual scenes of the target account are obtained. Associating virtual scene recommendations with user login time can provide users with personalized recommendations that are more in line with time preferences, meet the different interests and preferences that users may have at different times, and thus provide users with more accurate and personalized virtual scene recommendations, which not only improves the user experience, but also optimizes the allocation and utilization of scene resources.

[0065] Optionally, in the virtual scene recommendation method provided in an embodiment of the present application, the method also includes: when the target account responds to the target virtual scene, obtaining interaction information of the target account in the target virtual scene; based on the interaction information, determining recommendation correction information, wherein the recommendation correction information is used to correct the virtual scene recommended to the target account.

[0066] It can be understood that after the target account responds to the target virtual scene, that is, after the user enters the target virtual scene, the target account's interactive information in the target virtual scene is obtained, including dialogue response, clothing changes, music selection changes, user activities, stay time, click-through rate, feedback evaluation, etc. The above interactive information is preprocessed and analyzed, and the recommendation correction information is determined based on the analysis results. The virtual scene recommended to the target account is corrected using the recommendation correction information. By collecting the user's interactive information in the virtual scene and determining the recommendation correction information based on this information, it is not only possible to improve the personalization and accuracy of the recommendation and enhance the user experience, but also to achieve adaptive optimization of the system and enhance the user's immersion and participation in the virtual scene.

[0067] Optionally, the interaction information after the target account enters the target virtual scene is analyzed to obtain an analysis result, which can represent the behavior preference and satisfaction of the target account in the target virtual scene. The above analysis results are used to determine the recommendation correction information. The recommendation correction information is fed back to the recommendation system to adjust the model parameters and optimize the recommendation algorithm. By iterating the above process, the recommendation system can continue to learn and optimize, and gradually improve the personalization level and accuracy of the recommendation.

[0068] Optionally, the evaluation and optimization of the recommendation system includes evaluating performance, optimizing models, and providing user feedback. Evaluating performance refers to evaluating the performance of the recommendation system using a test set, and evaluation indicators include accuracy, recall, F1 score (an indicator for evaluating model performance, commonly used for classification task evaluation in fields such as information retrieval, natural language processing, and medical diagnosis), etc. Optimizing the model refers to adjusting the model structure, hyperparameters, or training process based on the evaluation results to improve system performance. User feedback refers to collecting user feedback to understand whether the recommendation results meet user expectations and further optimize the system.

[0069] Optionally, in the virtual scene recommendation method provided in the embodiment of the present application, based on the fusion features, a target virtual scene recommended to the target account is determined among M candidate virtual scenes, including: in a case where there is historical usage data of the target account in the M candidate virtual scenes, obtaining the scene resource requirements corresponding to the M candidate virtual scenes respectively; based on the historical usage data corresponding to the M candidate virtual scenes respectively, determining the historical loading times corresponding to the M candidate virtual scenes respectively; obtaining the delay tolerance of the target account; and determining the target virtual scene based on the fusion features, the delay tolerance, and the scene resource requirements and historical loading times corresponding to the M candidate virtual scenes respectively.

[0070] It can be understood that when there is historical usage data of the target account in the M candidate virtual scenes, the resource demand information corresponding to each candidate virtual scene is obtained. The above resource demand information may include computing resources, storage resources, network bandwidth, etc. According to the historical usage data of the target account in the M candidate virtual scenes, the historical loading time corresponding to each candidate virtual scene is determined. The delay tolerance of the target account is obtained, and the delay tolerance of the target account can be obtained through user settings, historical behavior analysis or questionnaires. Based on the fusion characteristics of the target account, the delay tolerance, and the scene resource requirements and historical loading times corresponding to the M candidate virtual scenes, the target virtual scene recommended to the target account is determined. By comprehensively considering user preferences, device performance and scene resource requirements, a comprehensive and personalized virtual scene recommendation mechanism is provided to users, which not only improves the personalization and accuracy of the recommendation, but also takes into account the limitations of the actual operating environment, thereby optimizing resource allocation and improving user experience.

[0071] Optionally, the resource requirement information is used to evaluate the hardware conditions required by the target account when loading and running each scenario, ensuring that the recommended scenario can run smoothly on the target account's device. The historical loading duration is used to evaluate the performance requirements of the virtual scenario for the target account's device. The delay tolerance refers to the loading delay time of the virtual scenario that the target account can accept.

[0072] Through the above step S201, the input data of N modalities of the target account are obtained, wherein the N modalities are the component modalities required for the virtual scene, and N is a positive integer greater than or equal to 2; step S202 determines the fusion features of the N modalities based on the input data of the N modalities; step S203 determines the target virtual scene recommended to the target account from the M candidate virtual scenes based on the fusion features, wherein M is a positive integer greater than N. The purpose of capturing the mapping relationship between user input and virtual scenes can be achieved, and the technical effect of providing users with more accurate and personalized virtual scene recommendations based on the user's multimodal preference input data is achieved, thereby solving the technical problem of unsatisfactory accuracy of user virtual scene recommendation results caused by complex and changeable user input and dynamically changing scene requirements.

[0073] Based on the above embodiments and optional embodiments, the present application proposes an optional implementation method for recommending virtual scenes to users based on the user's multimodal preference data. With the rapid development of virtual reality (VR) and augmented reality (AR) technologies, providing users with personalized and immersive virtual scene experiences has become an industry trend. However, how to accurately understand the diverse needs of users, such as conversation content, clothing choices, music preferences, etc., and map them to appropriate virtual scenes is a major challenge currently faced. Related technologies often rely on artificial rules or simple statistical models, which are difficult to handle complex and changeable user inputs and dynamically changing scene requirements. Therefore, the present application proposes an intelligent recommendation system based on machine learning technology, which aims to automatically learn the mapping relationship between user input and virtual scenes, and provide users with more accurate and personalized scene recommendations.

[0074] According to the virtual scene recommendation method of the present application, the problems of multi-source heterogeneous data fusion, complex mapping relationships, and low model fusion and decision-making accuracy faced by virtual scene recommendation can be effectively solved. For multi-source heterogeneous data fusion, it refers to the feature fusion of user input data such as conversation content, clothing selection, and music preferences. Since the above data are multi-source heterogeneous data and have different data structures and representations, how to effectively fuse these heterogeneous data to fully reflect user interests is a key issue facing virtual scene recommendation. For complex mapping relationship learning, since the mapping relationship between user input and virtual scenes is complex and changeable, it is necessary to build a machine learning model that can capture this complex relationship. For model fusion and decision-making, designing an effective fusion model to integrate the outputs of multiple independent models and making the final scene recommendation decision based on this is an important way to improve the accuracy of virtual scene recommendation results.

[0075] Step S1, data preprocessing and feature extraction, obtaining the user's corresponding conversation information, clothing information and music preference information, and preprocessing the above information obtained.

[0076] For conversation content data, NLP technology is used for text preprocessing, including word segmentation, stop word removal, word embedding, etc. For clothing selection data, the preprocessing method is to encode clothing features (such as category, color, style) into one-hot encoding (One-Hot Encoding) or embedding vectors (If the number of categories is small, one-hot encoding is used; if the number of categories is large, an embedding layer is used). For music preference data, similar to clothing selection, the preprocessing method is to encode music features (such as type, composer, genre) into one-hot encoding or embedding vectors.

[0077] Step S2, model construction training, constructing a preference area recommendation model including a conversation content model, a clothing selection model, a music preference model, and a fusion neural network.

[0078] For the conversation content data, the BERT model is selected as the feature extraction model to convert the conversation content into a fixed-length vector representation. The vector output by BERT is used as input, and the output of BERT is mapped to the classification space of the virtual scene through the fully connected layer, and the relevance or classification label of the conversation content and the virtual scene is output, that is, the unimodal feature corresponding to the conversation content modal data.

[0079] Select and load a pre-trained BERT model as a feature extractor. Input the pre-processed conversation content into the BERT model and extract the embedded representation of the CLS tag (Class Token, a special tag used to summarize the information of the entire sequence in sequence classification tasks for the final classification decision) as a fixed-length vector of the conversation. Add one or more fully connected layers after the BERT output to map the BERT output to the classification space of the virtual scene and output the relevance or classification label of the conversation content to the virtual scene. In the training phase, use the annotated data of the conversation content to train the conversation content model.

[0080] For clothing selection data, a multi-layer perceptron (MLP) is used as the feature extraction model. The embedding vector of the clothing is input, and the feature transformation is performed through the hidden layer. The relevance between the clothing and the virtual scene is output, that is, the unimodal feature corresponding to the clothing selection modal data.

[0081] The clothing selection model includes an MLP layer, a hidden layer, and an output layer. A multi-layer perceptron (MLP) is designed, whose input layer receives the embedding vector of clothing. One or more hidden layers are added after the input layer for feature transformation. The last layer outputs the relevance between clothing and virtual scenes. In the training phase, the clothing selection model is trained using labeled data of clothing and virtual scenes.

[0082] For music preference data, the MLP model is also used as the feature extraction model, which inputs the embedding vector of music preference and outputs the correlation between music and virtual scenes, that is, the unimodal features corresponding to the music preference modal data.

[0083] The music preference model includes an MLP layer, a hidden layer and an output layer. The MLP layer is used to process the embedding vector of the music preference. The hidden layer and the output layer are used to output the relevance between the music and the virtual scene. In the training phase, the music preference model is trained using the annotated data of the music preference and the virtual scene.

[0084] After obtaining the unimodal features corresponding to the conversation content data, clothing selection data, and music preference data, the fusion model is used to obtain the fusion features of the target account. A fusion neural network is designed, whose input layer receives the outputs of the conversation content model, clothing selection model, and music preference model, that is, the unimodal features corresponding to the conversation content data, clothing selection data, and music preference data. The fusion neural network adopts a multi-layer fully connected network structure, combines these inputs through nonlinear transformations and activation functions, and finally outputs the virtual scene area that the user may be interested in.

[0085] The fusion neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the output of the dialogue content model, the clothing selection model, and the music preference model as input. The hidden layer is used to add multiple fully connected layers and use nonlinear activation functions such as ReLU (Rectified Linear Unit) to perform feature fusion and transformation. The output layer is used to output the scene score of the virtual scene area that the user may be interested in. In the training phase, the fusion neural network is trained using a comprehensive data set containing all model outputs to learn the interactive relationship between different features.

[0086] After obtaining the scene scores of each virtual scene using the fusion neural network, the virtual scene with the highest score is selected as the recommendation result, that is, the target virtual scene. At the same time, a threshold mechanism can be introduced to make recommendations only when the highest score exceeds a certain threshold (that is, a predetermined score threshold) to improve the confidence of the recommendation and user satisfaction.

[0087] Step S3, the recommendation system is implemented. Based on the new user input (conversation content, clothing selection, music preference), the trained preference area recommendation model is used to obtain the preference virtual area and lead the user to enter the preference virtual area.

[0088] The virtual scene recommendation system includes multimodal data input processing, fusion output, recommendation decision, threshold mechanism and output recommendation result. Multimodal data input processing is used to process new user input data (conversation content, clothing selection, music preference) through their own trained models to obtain corresponding outputs. The fusion output is used to use the outputs of the three models as the input of the fusion neural network to obtain the scene score of the fused virtual scene area. The recommendation decision is used to select the highest score, that is, to select the scene with the highest score from all virtual scene areas as the recommendation result. The threshold mechanism is used to set a predetermined score threshold, and the recommendation is only made when the highest score exceeds the predetermined score threshold. This helps to improve the confidence of the recommendation and user satisfaction. The output recommendation result is used to present the recommended virtual scene area to the user.

[0089] Step S4: system evaluation and optimization, determining whether it is necessary to optimize the preferred area recommendation model based on the evaluation results.

[0090] System evaluation and optimization include performance evaluation, model optimization, and user feedback. Performance evaluation refers to the use of test sets to evaluate the performance of the recommendation system. Evaluation indicators include accuracy, recall, F1 score (an indicator for evaluating model performance, commonly used in classification task evaluation in fields such as information retrieval, natural language processing, and medical diagnosis). Optimizing the model refers to adjusting the model structure, hyperparameters, or training process based on the evaluation results to improve system performance. User feedback refers to collecting user feedback to understand whether the recommendation results meet user expectations and further optimize the system.

[0091] The above virtual scene recommendation method achieves the following effects: by automatically learning the mapping relationship between user input and virtual scenes, the accuracy and personalization of recommendations are improved; customized virtual scene recommendations are provided according to the diverse needs of users, enhancing user immersion and satisfaction; the constructed recommendation system can flexibly handle new user input types and virtual scenes, supporting the continuous upgrading and expansion of the system.

[0092] The virtual scene recommendation method provided in the embodiment of the present application obtains input data of N modes of the target account, wherein the N modes are the component modes required for the virtual scene, and N is a positive integer greater than or equal to 2; based on the input data of the N modes, the fusion features of the N modes are determined; based on the fusion features, the target virtual scene recommended to the target account is determined among the M candidate virtual scenes, wherein M is a positive integer greater than N. The problem of unsatisfactory accuracy of the user virtual scene recommendation results in the related art is solved. Thus, the effect of providing users with more accurate and personalized virtual scene recommendations based on the user's multimodal preference input data is achieved.

[0093] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0094] Example 2

[0095] The embodiment of the present application also provides a virtual scene recommendation device. It should be noted that the virtual scene recommendation device of the embodiment of the present application can be used to execute the virtual scene recommendation method provided by the embodiment of the present application. The virtual scene recommendation device provided by the embodiment of the present application is introduced below.

[0096] According to an embodiment of the present application, a device for implementing the above virtual scene recommendation method is also provided. Figure 3is a schematic diagram of a virtual scene recommendation device provided according to an embodiment of the present application, such as Figure 3 As shown, the device includes: an input data acquisition module 301, used to acquire input data of N modes of the target account, wherein the N modes are respectively the component modes required for the virtual scene, and N is a positive integer greater than or equal to 2; a fusion feature determination module 302, connected to the input data acquisition module 301, used to determine the fusion features of the N modalities based on the input data of the N modalities; a target virtual scene determination module 303, connected to the fusion feature determination module 302, used to determine the target virtual scene recommended to the target account from among the M candidate virtual scenes based on the fusion features, wherein M is a positive integer greater than N.

[0097] The virtual scene recommendation device provided in the embodiment of the present application is used to obtain the input data of N modes of the target account through the input data acquisition module 301, wherein the N modes are the component modes required for the virtual scene, and N is a positive integer greater than or equal to 2; the fusion feature determination module 302 is connected to the input data acquisition module 301, and is used to determine the fusion features of the N modes based on the input data of the N modes; the target virtual scene determination module 303 is connected to the fusion feature determination module 302, and is used to determine the target virtual scene recommended to the target account from the M candidate virtual scenes based on the fusion features, wherein M is a positive integer greater than N. The problem of unsatisfactory accuracy of the user virtual scene recommendation results in the related art is solved. Thereby achieving the effect of providing users with more accurate and personalized virtual scene recommendations based on the user's multimodal preference input data.

[0098] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the target virtual scene determination module includes: a scene score determination module, used to determine the scene scores corresponding to M candidate virtual scenes respectively according to the fusion features; a first scene determination module, used to select the candidate virtual scene corresponding to the highest score as the target virtual scene when the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold.

[0099] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the device also includes: a login time determination module, used to determine the login time of the target account; a scene set determination module, used to determine the scene sets corresponding to P time periods, wherein the scenes included in the scene sets corresponding to the P time periods are different, and P is a positive integer; a target time period determination module, used to determine the target time period corresponding to the login time in the P time periods; and a candidate scene determination module, used to obtain M candidate virtual scenes based on the scene sets corresponding to the target time periods.

[0100] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the device also includes: an interaction information determination module, used to obtain the interaction information of the target account in the target virtual scene when the target account responds to the target virtual scene; a correction information determination module, used to determine the recommended correction information based on the interaction information, wherein the recommended correction information is used to correct the virtual scene recommended to the target account.

[0101] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the target virtual scene determination module includes: a resource requirement determination module, which is used to obtain the scene resource requirements corresponding to the M candidate virtual scenes respectively when there is historical usage data of the target account in the M candidate virtual scenes; a loading time determination module, which is used to determine the historical loading times corresponding to the M candidate virtual scenes respectively based on the historical usage data corresponding to the M candidate virtual scenes respectively; a delay tolerance acquisition module, which is used to obtain the delay tolerance of the target account; and a second scene determination module, which is used to determine the target virtual scene based on the fusion features, delay tolerance, and the scene resource requirements and historical loading times corresponding to the M candidate virtual scenes respectively.

[0102] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the fusion feature determination module includes: a feature extraction model determination module, used to determine the feature extraction models corresponding to N modalities, wherein the feature extraction models corresponding to the N modalities have different feature extraction capabilities in different modalities; a single modal feature determination module, used to use the feature extraction models corresponding to the N modalities, perform feature extraction based on the input data of the N modalities, and obtain single modal features corresponding to the N modalities; a first feature determination module, used to perform feature fusion based on the single modal features corresponding to the N modalities, to obtain fused features.

[0103] Optionally, in the virtual scene recommendation device provided in the embodiment of the present application, the first feature determination module includes: an active frequency determination module, which is used to determine the active frequencies corresponding to N modalities based on the historical behavior data of the target account; a fusion weight determination module, which is used to determine the fusion weights corresponding to the N modalities according to the active frequencies corresponding to the N modalities; and a second feature determination module, which is used to perform feature fusion based on the fusion weights corresponding to the N modalities and the single modal features to obtain a fused feature.

[0104] It should be noted here that the above-mentioned input data acquisition module 301, fusion feature determination module 302 and target virtual scene determination module 303 correspond to steps S201 to S203 in Example 1, and the two modules and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above-mentioned modules can also be run in the computer terminal 10 provided in Example 1 as part of the device.

[0105] Example 3

[0106] An embodiment of the present application may provide an electronic device, Figure 4 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 4 As shown, the electronic device may include: one or more ( Figure 4 (only one is shown) processor 402, memory 404, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0107] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0108] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain input data of N modes of the target account, where the N modes are the component modes required for the virtual scene, and N is a positive integer greater than or equal to 2; based on the input data of the N modes, determine the fusion features of the N modalities; based on the fusion features, determine the target virtual scene recommended to the target account from among the M candidate virtual scenes, where M is a positive integer greater than N.

[0109] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the scene scores corresponding to the M candidate virtual scenes according to the fusion features; when the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold, use the candidate virtual scene corresponding to the highest score as the target virtual scene.

[0110] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the login time of the target account; determine the scene sets corresponding to P time periods, wherein the scenes included in the scene sets corresponding to the P time periods are different, and P is a positive integer; determine the target time period corresponding to the login time among the P time periods; and obtain M candidate virtual scenes based on the scene sets corresponding to the target time period.

[0111] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: when the target account responds to the target virtual scene, obtain the interaction information of the target account in the target virtual scene; based on the interaction information, determine the recommended correction information, wherein the recommended correction information is used to correct the virtual scene recommended to the target account.

[0112] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: when there is historical usage data of the target account in the M candidate virtual scenes, obtain the scene resource requirements corresponding to the M candidate virtual scenes respectively; based on the historical usage data corresponding to the M candidate virtual scenes respectively, determine the historical loading time corresponding to the M candidate virtual scenes respectively; obtain the delay tolerance of the target account; determine the target virtual scene based on the fusion features, delay tolerance, and the scene resource requirements and historical loading time corresponding to the M candidate virtual scenes respectively.

[0113] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the feature extraction models corresponding to the N modalities, wherein the feature extraction models corresponding to the N modalities have different feature extraction capabilities in different modalities; use the feature extraction models corresponding to the N modalities to perform feature extraction based on the input data of the N modalities to obtain single modal features corresponding to the N modalities; perform feature fusion based on the single modal features corresponding to the N modalities to obtain fused features.

[0114] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the active frequencies corresponding to the N modalities based on the historical behavior data of the target account; determine the fusion weights corresponding to the N modalities according to the active frequencies corresponding to the N modalities; perform feature fusion based on the fusion weights corresponding to the N modalities and single modal features to obtain fused features.

[0115] By adopting the embodiment of the present application, a solution of a virtual scene recommendation method is provided. By extracting and fusing the user's multimodal preference data to capture the mapping relationship between the user input and the virtual scene, the purpose of providing the user with more accurate and personalized virtual scene recommendations based on the user's multimodal preference input data is achieved, thereby solving the technical problem of unsatisfactory accuracy of the user's virtual scene recommendation results caused by complex and changeable user input and dynamically changing scene requirements.

[0116] It can be understood by those skilled in the art that Figure 4 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, or other terminal devices. Figure 4 The structure of the electronic device is not limited. Figure 4 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 4 Different configurations are shown.

[0117] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0118] Example 4

[0119] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the virtual scene recommendation method provided in the first embodiment.

[0120] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0121] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the virtual scene recommendation method.

[0122] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0123] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the above-mentioned units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0125] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0126] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0127] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the above-mentioned methods of each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0128] The above are only preferred implementations of the present application. It should be pointed out that ordinary technicians in this technical field can make several improvements and modifications without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A virtual scene recommendation method, characterized in that: include: Obtain input data of N modes of the target account, wherein the N modes are component modes required by the virtual scene, and N is a positive integer greater than or equal to 2; Determining fusion features of the N modalities based on the input data of the N modalities; Based on the fusion feature, a target virtual scene recommended to the target account is determined from M candidate virtual scenes, where M is a positive integer greater than N.

2. The method according to claim 1, characterized in that The determining, based on the fusion feature, a target virtual scene recommended to the target account from among the M candidate virtual scenes comprises: Determining scene scores corresponding to the M candidate virtual scenes respectively according to the fusion features; When the highest score among the scene scores corresponding to the M candidate virtual scenes is greater than a predetermined score threshold, the candidate virtual scene corresponding to the highest score is used as the target virtual scene.

3. The method according to claim 1, characterized in that The method further comprises: Determining a login time of the target account; Determine scene sets corresponding to P time periods respectively, wherein the scenes included in the scene sets corresponding to the P time periods respectively are different, and P is a positive integer; Determining a target time period corresponding to the login time in the P time periods; Based on the scene set corresponding to the target time period, the M candidate virtual scenes are obtained.

4. The method according to claim 1, characterized in that The method further comprises: When the target account responds to the target virtual scene, acquiring interaction information of the target account in the target virtual scene; Based on the interaction information, recommendation correction information is determined, wherein the recommendation correction information is used to correct the virtual scene recommended to the target account.

5. The method according to claim 1, characterized in that The determining, based on the fusion feature, a target virtual scene recommended to the target account from among the M candidate virtual scenes comprises: In a case where historical usage data of the target account exists in the M candidate virtual scenes, obtaining scene resource requirements corresponding to the M candidate virtual scenes respectively; Determine the historical loading durations respectively corresponding to the M candidate virtual scenes based on the historical usage data respectively corresponding to the M candidate virtual scenes; Obtaining the latency tolerance of the target account; The target virtual scene is determined based on the fusion feature, the delay tolerance, and the scene resource requirements and historical loading durations respectively corresponding to the M candidate virtual scenes.

6. The method according to any one of claims 1 to 5, characterized in that The determining, based on the input data of the N modalities, fusion features of the N modalities includes: Determining feature extraction models corresponding to the N modes respectively, wherein the feature extraction models corresponding to the N modes respectively have different feature extraction capabilities in different modes; Using feature extraction models corresponding to the N modes respectively, extracting features based on input data of the N modes, and obtaining single-modal features corresponding to the N modes respectively; Feature fusion is performed based on the single modal features corresponding to the N modalities respectively to obtain the fused feature.

7. The method according to claim 6, characterized in that The step of performing feature fusion based on the single-modal features corresponding to the N modalities to obtain the fused features includes: Based on the historical behavior data of the target account, determine the activity frequencies corresponding to the N modes respectively; Determining fusion weights corresponding to the N modes respectively according to the active frequencies corresponding to the N modes respectively; Feature fusion is performed based on the fusion weights and single-modal features respectively corresponding to the N modes to obtain the fused features.

8. A virtual scene recommendation device, characterized in that: include: An input data acquisition module, used to acquire input data of N modes of the target account, wherein the N modes are component modes required by the virtual scene, and N is a positive integer greater than or equal to 2; A fusion feature determination module, used to determine the fusion features of the N modalities based on the input data of the N modalities; The target virtual scene determination module is used to determine the target virtual scene recommended to the target account from M candidate virtual scenes based on the fusion feature, where M is a positive integer greater than N.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the virtual scene recommendation method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 7 when running.

Images