Cross-platform user identity alignment processing method and device, equipment and medium

Abstract

The invention discloses a cross-platform user identity alignment processing method and apparatus, a device and a medium. The method comprises the steps of obtaining user behavior data of a plurality of users of different platforms; determining identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by adopting sample user data and an initial training model of different platforms to obtain an embedded vector of heterogeneous information, encoding the embedded vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result; the method comprises the following steps: acquiring identity information of a plurality of users, determining alignment information of the plurality of users according to the identity information, constructing a heterogeneous graph neural network model, comprehensively considering various heterogeneous information of the users, and coding and fusing the information by using a deep learning technology so as to realize more accurate and efficient cross-platform user identity alignment processing of user identity alignment.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and in particular, relates to a method, device, equipment and medium for processing cross-platform user identity alignment. Background Art

[0002] With the rapid development of social networking technology and the diversification of online social platforms, the identity information of users on different social networking platforms has shown explosive growth. This information contains rich user behavior and social interaction data, which is of great value for user behavior analysis, personalized content recommendation, network security management, etc. However, due to factors such as data isolation between different platforms and user privacy protection, the identity data of users in different networks is often difficult to obtain, which greatly limits the comprehensive understanding and effective use of users' cross-platform behavior patterns.

[0003] Most existing user identity alignment methods are limited to single-dimensional attribute matching analysis or local network structure mining, and fail to fully utilize the multidimensional characteristics of users in social networks and the complex connections between user attributes, content, and social relationships. At the same time, traditional machine learning methods often find it difficult to effectively capture deep user characteristics and global social context when processing such high-dimensional, highly heterogeneous data containing rich social information. How to comprehensively process deep user characteristics and global social context to identify users' cross-platform identities is an urgent problem to be solved. Summary of the invention

[0004] The present application aims to provide a method, device, equipment and medium for processing cross-platform user identity alignment to solve the deficiencies in the prior art. The technical problem to be solved by the present application is achieved through the following technical solutions.

[0005] In a first aspect, an embodiment of the present application provides a method for processing cross-platform user identity alignment, the method comprising:

[0006] Obtain user behavior data of multiple users on different platforms;

[0007] Determine the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result;

[0008] Alignment information of the multiple users is determined according to the identity information.

[0009] Optionally, the identity alignment model is obtained in the following manner:

[0010] Obtain sample user data from different platforms;

[0011] Determining a user heterogeneous information data set based on the sample user data;

[0012] Encoding the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, the content embedding vector, and the time information embedding vector;

[0013] Constructing a heterogeneous graph according to the embedding vector of the heterogeneous information;

[0014] Performing adaptive weight aggregation on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector;

[0015] Inputting the model input vectors of users of different platforms into the multilayer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users of different platforms;

[0016] According to the output results corresponding to the model input vectors of the users of different platforms, it is determined whether the sample user data of different platforms are from the same user.

[0017] Optionally, encoding the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information includes:

[0018] The BERT model is used to encode the basic attribute information in the sample user data to obtain an embedding vector of the text attribute; wherein the basic attribute information includes at least a user name and a nickname;

[0019] The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector at least includes a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector;

[0020] The embedding vector of the text attribute, the content embedding vector and the time information embedding vector are fused to obtain an embedding vector of multi-dimensional heterogeneous information.

[0021] Optionally, constructing a heterogeneous graph according to the embedding vector of the heterogeneous information includes:

[0022] The heterogeneous graph is determined by taking the embedding vector of the text attribute, the content embedding vector and the time information embedding vector as nodes, and taking the relationship between the social relationship and the content relationship as an edge.

[0023] Optionally, the step of performing adaptive weight aggregation on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector includes:

[0024] Based on the graph attention model, corresponding weight values ​​are set for the nodes of the embedding vector of each text attribute, the content embedding vector, and the time information embedding vector respectively;

[0025] The model input vector is determined according to the embedding vector and the corresponding weight value.

[0026] In a second aspect, an embodiment of the present application provides a processing device for cross-platform user identity alignment, the device comprising:

[0027] The acquisition module is used to obtain user behavior data of multiple users on different platforms;

[0028] A determination module is used to determine the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result;

[0029] An alignment module is used to determine alignment information of the multiple users according to the identity information.

[0030] Optionally, the device further comprises a training module, wherein the training module is used to:

[0031] Obtain sample user data from different platforms;

[0032] Determining a user heterogeneous information data set based on the sample user data;

[0033] Encoding the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, the content embedding vector, and the time information embedding vector;

[0034] Constructing a heterogeneous graph according to the embedding vector of the heterogeneous information;

[0035] Performing adaptive weight aggregation on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector;

[0036] Inputting the model input vectors of users of different platforms into the multilayer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users of different platforms;

[0037] According to the output results corresponding to the model input vectors of the users of different platforms, it is determined whether the sample user data of different platforms are from the same user.

[0038] Optionally, the training module is used to:

[0039] The BERT model is used to encode the basic attribute information in the sample user data to obtain an embedding vector of the text attribute; wherein the basic attribute information includes at least a user name and a nickname;

[0040] The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector at least includes a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector;

[0041] The embedding vector of the text attribute, the content embedding vector and the time information embedding vector are fused to obtain an embedding vector of multi-dimensional heterogeneous information.

[0042] Optionally, the training module is used to:

[0043] The heterogeneous graph is determined by taking the embedding vector of the text attribute, the content embedding vector and the time information embedding vector as nodes, and taking the relationship between the social relationship and the content relationship as an edge.

[0044] Optionally, the training module is used to:

[0045] Based on the graph attention model, corresponding weight values ​​are set for the nodes of the embedding vector of each text attribute, the content embedding vector, and the time information embedding vector respectively;

[0046] The model input vector is determined according to the embedding vector and the corresponding weight value.

[0047] In a third aspect, an embodiment of the present application provides a terminal device, including: at least one processor and a memory;

[0048] The memory stores a computer program; the at least one processor executes the computer program stored in the memory to implement the processing method for cross-platform user identity alignment provided in the first aspect.

[0049] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed, the method for processing cross-platform user identity alignment provided in the first aspect is implemented.

[0050] The embodiments of the present application include the following advantages:

[0051] The processing method, device, equipment and medium for cross-platform user identity alignment provided by the embodiment of the present application obtains user behavior data of multiple users on different platforms; determines the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model uses sample user data and an initial training model of different platforms to obtain an embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain an encoding result, and performs adaptive weight aggregation on the encoding result; determines the alignment information of the multiple users according to the identity information, and comprehensively considers the various heterogeneous information of the users by constructing a heterogeneous graph neural network model, and uses deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight allocation mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the existing technical solutions, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0053] Figure 1 A flowchart of a method for processing cross-platform user identity alignment provided by an embodiment of the present application;

[0054] Figure 2 A schematic diagram of a process of a cross-platform user identity alignment method based on a heterogeneous graph neural network provided in one embodiment of the present application;

[0055] Figure 3 A schematic diagram of a process for embedding and extracting cross-platform user attributes based on a heterogeneous graph neural network provided by an embodiment of the present invention;

[0056] Figure 4A schematic diagram of a process for cross-platform user attribute enhancement and unified embedding representation based on a heterogeneous graph neural network provided by an embodiment of the present invention;

[0057] Figure 5 It is a structural block diagram of an embodiment of a processing device for cross-platform user identity alignment of the present application;

[0058] Figure 6 It is a structural diagram of a terminal device of the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. 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 the field without creative work are within the scope of protection of the present application.

[0060] An embodiment of the present application provides a method for aligning user identities across platforms, for aligning user identities. The execution subject of this embodiment is a device for aligning user identities across platforms, which is arranged on a terminal device, for example, the terminal device at least includes a computer terminal.

[0061] Reference Figure 1 , shows a flowchart of a method for processing cross-platform user identity alignment according to an embodiment of the present application, which method may specifically include the following steps:

[0062] S101, obtaining user behavior data of multiple users on different platforms;

[0063] Specifically, the terminal device obtains user behavior data of different platforms through various interfaces.

[0064] S102, determining identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result;

[0065] Specifically, the terminal device obtains sample user data from different platforms, and inputs the sample user data into the initial training model to obtain an embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain an encoding result, and adaptively weights the encoding result. By continuously training the initial training model, an identity alignment model is obtained, and the identity alignment model is used to determine whether user data from different platforms come from the same user.

[0066] S103: Determine alignment information of multiple users according to the identity information.

[0067] In the embodiment of the present application, BERTopic and BiLSTM are used to capture the deep feature information between nodes and nodes, which can not only capture the complex relationships between users on different social platforms, but also learn the intrinsic connections between user attributes, content and social behaviors, thereby significantly improving the accuracy and efficiency of user identity alignment. By utilizing the advantages of GNN and BERT, the problem of user identity alignment is effectively solved, the ability to capture user cross-platform behavioral characteristics is enhanced, and accurate alignment of user identities is successfully achieved, which not only improves the accuracy of user identity recognition, but also can better understand and mine the complex behavior patterns of users in social networks, thereby significantly improving the personalization level and security of services in practical applications such as personalized recommendations, advertising targeting, network security, etc.

[0068] The processing method for cross-platform user identity alignment provided by the embodiment of the present application is obtained by obtaining user behavior data of multiple users on different platforms; according to the pre-trained identity alignment model, the identity information corresponding to the user behavior data is determined; wherein the pre-trained identity alignment model adopts sample user data of different platforms and the initial training model to obtain the embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain the encoding result, and the encoding result is obtained by adaptive weight aggregation; according to the identity information, the alignment information of multiple users is determined, and by constructing a heterogeneous graph neural network model, a variety of heterogeneous information of users is comprehensively considered, and deep learning technology is used to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight distribution mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0069] Another embodiment of the present application further supplements the processing method for cross-platform user identity alignment provided by the above embodiment.

[0070] Figure 2 A schematic diagram of a process of a cross-platform user identity alignment method based on a heterogeneous graph neural network provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, including:

[0071] S11: Data collection and preprocessing: Collect user data from different platforms through interfaces, clean the data, remove noise and standardize the format, build a user heterogeneous information data set and divide it into training set and test set;

[0072] S12: User attribute embedding extraction: Encode various heterogeneous information of users based on BERT and BiLSTM, and extract the embedding representation of user attributes in a unified way;

[0073] S13: User attribute enhancement and unified embedding representation: BERTopic model is used to extract user’s text style and interests, these deep features are integrated into the heterogeneous graph, and GAT (graph attention network) is used to perform adaptive weight aggregation on user’s heterogeneous information nodes to obtain a unified user embedding representation;

[0074] S14: User identity alignment: Based on the MLP model, input the user embedding representations in two different platforms and predict whether they belong to the same user;

[0075] S15: Model performance evaluation: Evaluate the performance of the model on the user alignment task, and use a variety of evaluation indicators to verify the alignment accuracy of the model.

[0076] Optionally, the identity alignment model is obtained as follows:

[0077] Obtain sample user data from different platforms;

[0078] Specifically, user data from different platforms are collected through interfaces, the data is cleaned, noise is removed and the format is standardized, a user heterogeneous information dataset is constructed and divided into training and test sets;

[0079] Determine a user heterogeneous information data set based on sample user data;

[0080] Encode the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, a content embedding vector, and a time information embedding vector;

[0081] Construct a heterogeneous graph based on the embedding vectors of heterogeneous information;

[0082] Adaptively weight the information of each user's heterogeneous information node in the heterogeneous graph to obtain the model input vector;

[0083] Inputting the model input vectors of users on different platforms into the multi-layer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users on different platforms;

[0084] According to the output results corresponding to the model input vectors of users on different platforms, it is determined whether the sample user data of different platforms come from the same user, the user heterogeneous information data is received from the data acquisition module, and the user identities are aligned using the heterogeneous graph neural network architecture.

[0085] The pre-defined evaluation indicators are used to evaluate and improve the results of user identity alignment, and the effect of user alignment is objectively evaluated to ensure the accuracy and reliability of the user identity alignment results.

[0086] Optionally, the heterogeneous information in the user heterogeneous information dataset is encoded based on a deep learning network model to obtain an embedding vector of the heterogeneous information, including:

[0087] The BERT model is used to encode the basic attribute information in the sample user data to obtain the embedding vector of the text attribute; wherein the basic attribute information includes at least the user name and nickname;

[0088] The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector includes at least a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector;

[0089] The embedding vectors of text attributes, content embedding vectors and time information embedding vectors are fused to obtain multi-dimensional embedding vectors of heterogeneous information.

[0090] Specifically, the user attribute embedding and extraction step in S12 includes:

[0091] Build a text attribute embedding module: Encode the user's basic attribute information (such as username, nickname, personal profile, etc.) using the BERT model to obtain a deep semantic representation;

[0092] Build a content embedding module: process user-generated content (such as text and images), where text content is embedded through the BERT model, and image content is further encoded using BERT after features are extracted through a pre-trained CNN model;

[0093] Among them: BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained deep learning model that can achieve good results in a variety of NLP tasks. BERTopic is a text clustering tool based on BERT, mainly used for text clustering and topic classification. It clusters similar texts together by encoding the text, extracting features from the text, and then clustering the features. The main advantage of BERTopic is that it uses the BERT model, which has strong capabilities in processing natural language. The BERT model can capture contextual information in the text and extract important features from it. In addition, the BERT model can be fine-tuned through multi-task learning to make it more suitable for specific text clustering tasks. In BERTopic, the text is first preprocessed into word-level vector representations. Then, these vectors are encoded using the BERT model to obtain the contextual embedding representation of each word. These representations are further processed into sentence-level vector representations, and then they are clustered by a clustering algorithm. Finally, by analyzing each cluster, the topic classification results of the text can be obtained.

[0094] Build a time information embedding module: standardize the user's timestamp information (such as registration time, activity recording time) and convert it into a format suitable for model input;

[0095] Build a serialized information embedding module: Use the BiLSTM model to process serialized data, such as user activity logs or time series data, to capture temporal dependencies;

[0096] Heterogeneous information fusion module: The embedding representations of the above different types of attributes are fused to form a comprehensive user attribute embedding vector to fully reflect the multi-dimensional characteristics of the user.

[0097] Figure 3 A schematic diagram of a process for embedding and extracting cross-platform user attributes based on a heterogeneous graph neural network provided by an embodiment of the present invention specifically includes:

[0098] S21. Build a text attribute embedding module: Encode the user's basic attribute information (such as user name, nickname, personal profile, etc.) using the BERT model to obtain a deep semantic representation;

[0099] Considering the existence of short text attributes such as usernames and nicknames and long text attributes such as tweet text, the model extracts text features using the MaxPooling method. In this way, for short text attributes, one segment is generally sufficient, so it is not affected by the splitting and Pooling methods. For long texts such as tweets, they can be input into the BERT model by splitting, and then the features of multiple segments can be aggregated by MaxPooling. As shown in Formula 1, the model first splits the filtered text according to the input length specified by BERT to obtain texts, and then sends the segmented text set to the BERT pre-trained model for embedding extraction, and then performs MaxPooling on the text segment embedding representation set to obtain the final text embedding representation e text .

[0100] e text =MaxPooling(BERT(texts)) (Formula 1)

[0101] S22. Build a content embedding module: process user-generated content (such as text and images), where text content is embedded through the BERT model, and image content is further encoded using BERT after features are extracted through a pre-trained CNN model;

[0102] S23. Build a time information embedding module: standardize the user's timestamp information (such as registration time, activity record time) and convert it into a format suitable for model input;

[0103] Time is key information in social networking platforms, such as user registration time, user login time, and tweet publishing time. This information is a very critical factor in user features. It can be used to weight other features or as a feature embedding for unified processing. However, different social networking platforms have different ways of representing time. Some social networking platforms use Unix timestamp format (such as 1675524522), ISO 8601 (International Organization for Standardization) format (such as 2023-02-04T15:28:42.178Z), RFC 1123 (Request For Comments) format (such as Sat, 04Feb 202315:28:42GMT), etc. These times in different formats are difficult to process, and time attributes in different formats will also lead to mismatched embedding between different platforms. In addition, the BERT pre-trained model can only accept time in text format, so the model converts time into a unified text form. First, the model converts the time in different formats into Unix timestamps to obtain a numeric time. Then, through the time formatting operation, the numeric time is converted into a text time in the format of yyyy-MM-dd HH:mm:ss. Then, the unified time text is processed according to the text embedding extraction task flow to obtain the final time embedding representation. time .

[0104] S24. Build a serialized information embedding module: Use the BiLSTM model to process serialized data, such as user activity logs or time series data, to capture temporal dependencies.

[0105] S25, heterogeneous information fusion module: fuse the embedding representations of the above different types of attributes to form a comprehensive user attribute embedding vector to fully reflect the multi-dimensional characteristics of the user. Optionally, construct a heterogeneous graph based on the embedding vector of the heterogeneous information, including:

[0106] The embedding vectors of text attributes, content embedding vectors, and time information embedding vectors are taken as nodes, and the relationships between social relations and content relations are taken as edges to determine a heterogeneous graph.

[0107] Figure 3 A schematic diagram of a process for cross-platform user attribute enhancement and unified embedding representation based on a heterogeneous graph neural network provided by an embodiment of the present invention;

[0108] S31. Build a text style and interest extraction module: Use the BERTopic model to perform topic modeling on user-generated text content, extract topic information from the text and convert it into an embedding vector;

[0109] After the corpus and parameters are fed into the BERTopic framework, a topic model TopicModel is obtained, which can output a topicText, that is, a text type topic, for each input document doc.

[0110] After training the topic model TopicModel, the model will first extract the topic of each tweet. c The text in is sent to the topic model topicText to obtain the text type topic TopicModel of all tweets, as shown in Formula 2.

[0111] topicText=topicModel(getText(V c )) (Formula 2)

[0112] In order to maintain a uniform embedding vector length of 768, the model converts each text topic topicText into a real number topicIndex through the topic indexer topicIndexer. Its implementation is just a mapping function that maps text to a real number. As shown in Formula 3, each text topic input to the topic indexer will be converted into a real number between [0,767].

[0113] topicIndex=TopicIndexer(topicText),topicIndex∈[0,767](Formula 3)

[0114] S32, heterogeneous graph construction module: various attributes of users are embedded, style is embedded, and interest is embedded as nodes, and edges between nodes are constructed through social relationships and content relationships to form a comprehensive heterogeneous graph structure;

[0115] S33, build GAT embedding aggregation module: use GAT model to perform adaptive weight aggregation on nodes in heterogeneous graphs to generate user embedding representation that integrates multiple information;

[0116] Through the above processing, each user's attributes have been converted into feature vectors, including the user's portrait embedding e p , a user's tweet list is embedded c , user style embedding style and user interest embedding interestDifferent embedding features have different weights, so when the model aggregates these features, it needs to assign different weights to different types of features. Most of the existing models give each feature an average weight, or give different weights to different types of features based on personal subjective feelings, or assign weights based on some statistics or experience. These practices are subjective to a certain extent, and for the scenario of heterogeneous social network graphs with many types of features, this manual weight assignment method is very cumbersome and not conducive to the addition of new features.

[0117] In order to assign an objective weight to different features and avoid manually assigning weights to features, the model uses the GAT model to aggregate these heterogeneous types of embedded features. The GAT model can assign weights to embedded features based on the importance of the embedded features when aggregating feature embeddings. Therefore, the user embedding feature we finally obtain is an accurate embedding with adaptive weights that can accurately describe the user. In order to use the GAT model to aggregate the user's embedded features, the model first embeds the user's portrait into the e p , a user's tweet list is embedded c , user style embedding style and user interest embedding interest Model it into a heterogeneous graph and establish edges between embedded nodes. Here, the model embeds node e with user profiles. p As the central node, connecting the other three nodes c 、e style and e interest , thus completing the transformation of embedded feature nodes to heterogeneous graphs. As shown in Formula 4, this paper uses g to represent a heterogeneous social network graph with users as the central nodes, which contains the user's embedded feature nodes e p 、e c and e style and e interest , and the edges e between embedded feature nodes p →e c 、e p →e style and e p →e interest .

[0118] g=({e p ,e c ,e style e interest},{e p →e c ,e p →e style ,e p →e interest})(Formula 4)

[0119] Then, the model feeds the heterogeneous graph g based on user embedding features into the GAT network model to obtain a user embedding list that aggregates other embedding features. The embedding here includes the embedding of each embedding feature node, and the model only selects the central node e p The embedding of is used as the final embedding representation of the user. As shown in Formula 5, e U represents the final user embedding representation, and Get represents obtaining the final e obtained by GAT p Embedded representation of a node.

[0120] e U =Get(GAT(g),e p ) (Formula 5)

[0121] S34, feature fusion and optimization module: The features extracted by BERT, BiLSTM and BERTopic are integrated to finally generate a unified embedding representation of the user that integrates various heterogeneous information of the user. This representation can comprehensively reflect the user's behavior and characteristics in the social network.

[0122] Optionally, adaptive weight aggregation is performed on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector, including:

[0123] Based on the graph attention model, corresponding weight values ​​are set for the nodes of the embedding vectors of each text attribute, the content embedding vector, and the time information embedding vector;

[0124] According to the embedding vector and the corresponding weight value, the model input vector is determined.

[0125] Specifically, the user attribute enhancement and unified embedding representation step in S13 includes:

[0126] Build a text style and interest extraction module: Use the BERTopic model to perform topic modeling on user-generated text content, extract topic information from the text and convert it into an embedding vector;

[0127] Heterogeneous graph construction module: various user attribute embeddings, style embeddings, and interest embeddings are used as nodes, and edges between nodes are constructed through social relationships and content relationships to form a comprehensive heterogeneous graph structure;

[0128] Constructing a GAT embedding aggregation module: Using the GAT model to adaptively weight nodes in heterogeneous graphs, generating a user embedding representation that integrates multiple information;

[0129] Feature fusion and optimization module: The features extracted by BERT, BiLSTM and BERTopic are fused to finally generate a unified embedding representation of the user that integrates multiple heterogeneous information of the user. This representation can fully reflect the user's behavior and characteristics in the social network. The user identity alignment step in S15 includes:

[0130] After completing the unified embedding representation of users, use the trained MLP model to input the unified embedding representation of users in the two platforms as features;

[0131] The MLP model performs identity alignment prediction based on the learned user features to determine whether two users have the same identity.

[0132] The embodiment of the present application also includes the step of optimizing the model performance:

[0133] According to the evaluation results, the model parameters are adjusted, including but not limited to the number of GNN layers, node embedding dimension, time embedding vector dimension, the number of BERT model layers and hidden layer size, as well as the optimizer selection and dynamic adjustment strategy of the learning rate, to improve the overall performance of the model on the user identity alignment task.

[0134] Evaluate and improve the aligned user identity results to ensure the accuracy and reliability of the alignment results. This module receives the aligned user identity data from the user identity alignment module and uses pre-defined evaluation indicators (including accuracy, F1 score, precision, recall, etc.) to objectively evaluate the alignment effect. These indicators can help evaluate the difference and similarity between the alignment results and the real data.

[0135] Based on the evaluation results, the effect evaluation module can also propose improvement methods and suggestions, such as adjusting model parameters, retraining the model, or adopting other completion strategies, to further improve the quality and stability of the completion results.

[0136] The embodiment of the present application constructs a heterogeneous graph neural network model, comprehensively considers various heterogeneous information of users, and uses deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment. The embodiment of the present application not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight allocation mechanism, the embodiment of the present application also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management. It can comprehensively process heterogeneous social information and effectively identify users' cross-platform identities, so as to more accurately simulate and predict user behavior, and improve the personalization level of social network services and the accuracy of data mining.

[0137] The processing method for cross-platform user identity alignment provided by the embodiment of the present application is obtained by obtaining user behavior data of multiple users on different platforms; according to the pre-trained identity alignment model, the identity information corresponding to the user behavior data is determined; wherein the pre-trained identity alignment model adopts sample user data of different platforms and the initial training model to obtain the embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain the encoding result, and the encoding result is obtained by adaptive weight aggregation; according to the identity information, the alignment information of multiple users is determined, and by constructing a heterogeneous graph neural network model, a variety of heterogeneous information of users is comprehensively considered, and deep learning technology is used to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight distribution mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0138] Another embodiment of the present application provides a cross-platform user identity alignment processing device, which is used to execute the cross-platform user identity alignment processing method provided by the above embodiment.

[0139] Reference Figure 5 , shows a structural block diagram of an embodiment of a processing device for cross-platform user identity alignment of the present application, the device may specifically include the following modules: an acquisition module 501, a determination module 502 and an alignment module 503, wherein:

[0140] The acquisition module 501 is used to acquire user behavior data of multiple users on different platforms;

[0141] The determination module 502 is used to determine the identity information corresponding to the user behavior data according to the pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result;

[0142] The alignment module 503 is used to determine alignment information of multiple users according to the identity information.

[0143] The processing device for cross-platform user identity alignment provided by the embodiment of the present application obtains user behavior data of multiple users on different platforms; determines the identity information corresponding to the user behavior data according to the pre-trained identity alignment model; wherein the pre-trained identity alignment model adopts sample user data of different platforms and the initial training model to obtain the embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain the encoding result, and performs adaptive weight aggregation on the encoding result; determines the alignment information of multiple users according to the identity information, and comprehensively considers the various heterogeneous information of users by constructing a heterogeneous graph neural network model, and uses deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight allocation mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0144] Another embodiment of the present application further supplements the processing device for cross-platform user identity alignment provided by the above embodiment.

[0145] Optionally, the device further comprises a training module, the training module being used for:

[0146] Obtain sample user data from different platforms;

[0147] Determine a user heterogeneous information data set based on sample user data;

[0148] Encode the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, a content embedding vector, and a time information embedding vector;

[0149] Construct a heterogeneous graph based on the embedding vectors of heterogeneous information;

[0150] Adaptively weight the information of each user's heterogeneous information node in the heterogeneous graph to obtain the model input vector;

[0151] Inputting the model input vectors of users on different platforms into the multi-layer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users on different platforms;

[0152] According to the output results corresponding to the model input vectors of the users of different platforms, it is determined whether the sample user data of different platforms are from the same user.

[0153] Optionally, the training module is used to:

[0154] The BERT model is used to encode the basic attribute information in the sample user data to obtain the embedding vector of the text attribute; wherein the basic attribute information includes at least the user name and nickname;

[0155] The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector includes at least a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector;

[0156] The embedding vectors of text attributes, content embedding vectors and time information embedding vectors are fused to obtain multi-dimensional embedding vectors of heterogeneous information.

[0157] Optionally, the training module is used to:

[0158] The embedding vectors of text attributes, content embedding vectors, and time information embedding vectors are taken as nodes, and the relationships between social relations and content relations are taken as edges to determine a heterogeneous graph.

[0159] Optionally, the training module is used to:

[0160] Based on the graph attention model, corresponding weight values ​​are set for the nodes of the embedding vectors of each text attribute, the content embedding vector, and the time information embedding vector;

[0161] According to the embedding vector and the corresponding weight value, the model input vector is determined.

[0162] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0163] The processing device for cross-platform user identity alignment provided by the embodiment of the present application obtains user behavior data of multiple users on different platforms; determines the identity information corresponding to the user behavior data according to the pre-trained identity alignment model; wherein the pre-trained identity alignment model adopts sample user data of different platforms and the initial training model to obtain the embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain the encoding result, and performs adaptive weight aggregation on the encoding result; determines the alignment information of multiple users according to the identity information, and comprehensively considers the various heterogeneous information of users by constructing a heterogeneous graph neural network model, and uses deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight allocation mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0164] Yet another embodiment of the present application provides a terminal device for executing the processing method for cross-platform user identity alignment provided in the above embodiment.

[0165] Figure 6 is a schematic diagram of the structure of a terminal device of the present application, such as Figure 6 As shown, the terminal device includes: at least one processor 601 and a memory 602;

[0166] The memory stores a computer program; and at least one processor executes the computer program stored in the memory to implement the processing method for cross-platform user identity alignment provided in the above embodiment.

[0167] The terminal device provided in this embodiment obtains user behavior data of multiple users from different platforms; determines the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model uses sample user data from different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain an encoding result, and performs adaptive weight aggregation on the encoding result; determines the alignment information of multiple users according to the identity information, and comprehensively considers the various heterogeneous information of the users by constructing a heterogeneous graph neural network model, and uses deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the adaptability and generalization ability of the model to different types of social platform data. Through the automated feature extraction and weight distribution mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0168] Yet another embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed, the processing method for cross-platform user identity alignment provided by any of the above embodiments is implemented.

[0169] According to the computer-readable storage medium of this embodiment, by obtaining user behavior data of multiple users on different platforms; according to the pre-trained identity alignment model, determine the identity information corresponding to the user behavior data; wherein the pre-trained identity alignment model uses sample user data of different platforms and the initial training model to obtain the embedding vector of heterogeneous information, and encodes the embedding vector of the heterogeneous information to obtain the encoding result, and adaptively weights the encoding result to obtain; according to the identity information, determine the alignment information of multiple users, by constructing a heterogeneous graph neural network model, comprehensively consider the user's various heterogeneous information, and use deep learning technology to encode and fuse this information to achieve more accurate and efficient user identity alignment, which not only improves the accuracy of user identity alignment, but also enhances the model's adaptability and generalization ability to different types of social platform data. Through the automated feature extraction and weight distribution mechanism, the present invention also significantly improves the efficiency of processing large-scale social network data, and provides an effective technical means for social network analysis and user identity management.

[0170] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0171] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0172] 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 terms 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.

[0173] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0174] For ease of description, spatially relative terms, such as "above", "above", "on the upper surface of", "above", etc., may be used herein to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" may include both "above" and "below". The device may also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatially relative descriptions used herein are interpreted accordingly.

[0175] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0176] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for processing cross-platform user identity alignment, characterized in that: The method comprises: Obtain user behavior data of multiple users on different platforms; Determine the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result; Alignment information of the multiple users is determined according to the identity information.

2. The method for processing cross-platform user identity alignment according to claim 1, characterized in that: The identity alignment model is obtained in the following way: Obtain sample user data from different platforms; Determining a user heterogeneous information data set based on the sample user data; Encoding the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, the content embedding vector, and the time information embedding vector; Constructing a heterogeneous graph according to the embedding vector of the heterogeneous information; Performing adaptive weight aggregation on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector; Inputting the model input vectors of users of different platforms into the multilayer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users of different platforms; According to the output results corresponding to the model input vectors of the users of different platforms, it is determined whether the sample user data of different platforms are from the same user.

3. The method for processing cross-platform user identity alignment according to claim 2, characterized in that: The encoding of the heterogeneous information in the user heterogeneous information data set based on the deep learning network model to obtain an embedding vector of the heterogeneous information includes: The BERT model is used to encode the basic attribute information in the sample user data to obtain an embedding vector of the text attribute; wherein the basic attribute information includes at least a user name and a nickname; The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector at least includes a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector; The embedding vector of the text attribute, the content embedding vector and the time information embedding vector are fused to obtain an embedding vector of multi-dimensional heterogeneous information.

4. The method for processing cross-platform user identity alignment according to claim 2, characterized in that: The constructing a heterogeneous graph according to the embedding vector of the heterogeneous information includes: The heterogeneous graph is determined by taking the embedding vector of the text attribute, the content embedding vector and the time information embedding vector as nodes, and taking the relationship between the social relationship and the content relationship as an edge.

5. The method for processing cross-platform user identity alignment according to claim 2, characterized in that: The adaptive weight aggregation of the heterogeneous type information node information of each user in the heterogeneous graph to obtain the model input vector includes: Based on the graph attention model, corresponding weight values ​​are set for the nodes of the embedding vector of each text attribute, the content embedding vector, and the time information embedding vector respectively; The model input vector is determined according to the embedding vector and the corresponding weight value.

6. A processing device for aligning user identities across platforms, characterized in that: The device comprises: The acquisition module is used to obtain user behavior data of multiple users on different platforms; A determination module is used to determine the identity information corresponding to the user behavior data according to a pre-trained identity alignment model; wherein the pre-trained identity alignment model is obtained by using sample user data of different platforms and an initial training model to obtain an embedding vector of heterogeneous information, and encoding the embedding vector of the heterogeneous information to obtain an encoding result, and performing adaptive weight aggregation on the encoding result; An alignment module is used to determine alignment information of the multiple users according to the identity information.

7. The cross-platform user identity alignment processing device according to claim 6, characterized in that: The device also includes a training module, which is used to: Obtain sample user data from different platforms; Determining a user heterogeneous information data set based on the sample user data; Encoding the heterogeneous information in the user heterogeneous information dataset based on a deep learning network model to obtain an embedding vector of the heterogeneous information, wherein the embedding vector of the heterogeneous information at least includes an embedding vector of a text attribute, the content embedding vector, and the time information embedding vector; Constructing a heterogeneous graph according to the embedding vector of the heterogeneous information; Performing adaptive weight aggregation on the heterogeneous type information node information of each user in the heterogeneous graph to obtain a model input vector; Inputting the model input vectors of users of different platforms into the multilayer perceptron model respectively, and obtaining output results corresponding to the model input vectors of users of different platforms; According to the output results corresponding to the model input vectors of the users of different platforms, it is determined whether the sample user data of different platforms are from the same user.

8. The cross-platform user identity alignment processing device according to claim 6, characterized in that: The training module is used to: The BERT model is used to encode the basic attribute information in the sample user data to obtain an embedding vector of the text attribute; wherein the basic attribute information includes at least a user name and a nickname; The BERTopic model is used to encode the user-generated content in the sample user data to obtain a content embedding vector, wherein the content embedding vector at least includes a topic information embedding vector and an interest information embedding vector; the timestamp information in the sample user data is encoded to obtain a time information embedding vector; The embedding vector of the text attribute, the content embedding vector and the time information embedding vector are fused to obtain an embedding vector of multi-dimensional heterogeneous information.

9. A terminal device, characterized in that: include: at least one processor and memory; The memory stores a computer program; The at least one processor executes the computer program stored in the memory to implement the cross-platform user identity alignment processing method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, implements the cross-platform user identity alignment processing method according to any one of claims 1 to 5.

Images