A social platform guide topic key node tracing method

Abstract

The application belongs to the technical field of internet application, and particularly relates to a key node tracing method for guiding a topic on a social platform, comprising the following steps: obtaining user data; pre-processing the user data; extracting attribute features of the pre-processed user data; processing the user data by using a hybrid random walk strategy to obtain user hidden relationships; performing sentiment analysis on topic information in the user data to obtain user sentiment features; performing vectorization representation on the user attribute features, the user hidden relationships and the user sentiment features by using a URE2vec algorithm, and performing dimension reduction on the vectorization representation result to obtain hidden information; inputting the hidden information into a key node tracing model to obtain a topic propagation key node and a topic guiding intention; and the application adopts a hybrid random walk strategy, finds out implicit friends with similar preferences for the user by using the hybrid random walk strategy and utilizing explicit relationships and implicit relationships, and relieves data sparseness.

Description

Technical Field

[0001] This invention belongs to the field of Internet application technology, specifically relating to a method for tracing the key nodes of topics guided by social platforms. Background Technology

[0002] In today's digital age, social media has become a crucial platform for people to access information, share opinions, and build connections. With the help of the internet, the speed and reach of information dissemination are faster and wider than ever before, generating a greater impact. Compared to traditional media such as television, newspapers, and radio, social media platforms like Weibo have significant advantages. Users can update and share information in real time, easily comment, forward, and interact with content through likes and comments. This increased engagement facilitates faster information dissemination and makes it easier for users to participate in various discussions. However, information overload and the issue of information credibility have become critical challenges that urgently need to be addressed in the current social network environment. In the social media environment, the guiding influence of specific topics has enormous power, shaping user attention and behavior and profoundly impacting public opinion, the market, and society.

[0003] In recent years, research on the discovery of key node influence in various topics has mainly focused on three aspects. First, research based on network structure uses graph theory to analyze network structure and the connections between nodes, employing metrics such as node degree centrality to assess node influence. Second, methods based on community discovery identify nodes with high influence by recognizing community structures. Third, research based on user behavior analyzes user interactions such as likes, comments, and shares to measure the strength of influence among users.

[0004] However, existing technologies still face several challenges in tracing the key nodes of topic dissemination: 1. Hidden relationships among users within a topic. The relationships between users in a topic network are complex and diverse. Implicit relationships can lead to missing node information, making accurate node representation difficult and affecting the analysis and understanding of the topic network. Discovering and representing implicit user relationships from the topic network structure is a challenge. 2. The concealment of the guide's intentions within the topic space. Due to the hidden intentions in guiding topics, guides often deliberately guide or conceal certain information to lead the public. How to uncover the guide's intentions within a topic is also a major challenge. 3. How to quantify user influence and uncover the guides behind it. Guiding topics are usually proposed by one or more users to achieve a certain purpose, thereby stimulating various behaviors among ordinary users. Quantifying the influence of key users and uncovering the guides is also a significant challenge. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method for tracing the key nodes of topics guided by social media platforms. This method includes:

[0006] Acquire user data, including basic user information, historical user behavior, and topic information; preprocess the user data;

[0007] Extract attribute features from preprocessed user data; use a hybrid random walk strategy to process user data and obtain hidden user relationships; perform sentiment analysis on topic information in user data to obtain user sentiment features;

[0008] The URE2vec algorithm is used to vectorize user attribute features, hidden user relationships, and user sentiment features. The dimensionality of the vectorized representation is then reduced to obtain the hidden information.

[0009] By inputting the hidden information into the key node tracing model, we can obtain the key nodes of topic propagation and the topic guidance intention.

[0010] The beneficial effects of this invention are:

[0011] This invention constructs a hybrid random walk strategy, which uses both explicit and implicit relationships to find implicit friends with similar preferences for users, thus alleviating data sparsity. This invention employs the URE2vec algorithm to construct hidden information between user relationships and emotions, and uses the RE-GCN user relationship-emotion feature classification model to analyze the influence of user behavior and emotions on the topic propagation process, thereby achieving accurate identification of key nodes. Attached Figure Description

[0012] Figure 1 This is an overall flowchart of the present invention;

[0013] Figure 2 This is a schematic diagram of the hybrid random walk of the present invention;

[0014] Figure 3 This is a schematic diagram of the RE-GCN of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] A method for tracing the key nodes that guide topics on social media platforms, such as Figure 1As shown, the method includes: acquiring user data, which includes basic user information, historical user behavior, and topic information; preprocessing the user data; extracting attribute features from the preprocessed user data; processing the user data using a hybrid random walk strategy to obtain hidden user relationships; performing sentiment analysis on the topic information in the user data to obtain user sentiment features; using the URE2vec algorithm to vectorize the user attribute features, hidden user relationships, and user sentiment features, and reducing the dimensionality of the vectorized representation to obtain hidden information; and inputting the hidden information into a key node tracing model to obtain key nodes for topic propagation and topic guidance intentions.

[0017] like Figure 1 The diagram illustrates the overall process of this invention, showing that the inputs are basic user information, historical user behavior data, and basic topic information from a social media platform. The output after processing by the source tracing model identifies the key propagation nodes. Specific steps include:

[0018] S1: Online Data Acquisition. Data acquisition will be done by accessing Twitter and Weibo datasets. The required data includes basic information about users participating in guiding topics, such as interests, followings, and followers; historical user behavior data within the topics (e.g., browsing, saving, liking, commenting, and retweeting); and the content of the topics themselves.

[0019] S2: Extract relevant attributes. A dual-graph hybrid random walk algorithm is designed to mine potential relationships and hidden information among users in a topic, thereby obtaining a more detailed network propagation topology. Then, graph attention network mining is introduced to guide user sentiment in the topic, resulting in a representation of user sentiment features.

[0020] S3: Model Building. A URE2vec algorithm is designed to represent user relationships and sentiment information as vectors. Then, a graph convolutional neural network is used to construct a guided activity-user relationship sentiment classification model (RE-GCN). This model analyzes the influence of user nodes on topic dissemination during the guided activity propagation process and classifies key propagation nodes accordingly.

[0021] In this embodiment, data acquisition includes: acquiring raw data. Raw data can be obtained through public APIs of social networks or by directly downloading existing data sources. Simple data cleaning. Raw data is usually unstructured and cannot be directly used for data analysis. Simple data cleaning can structure most unstructured data. For example, deleting duplicate data and cleaning up invalid information.

[0022] In this embodiment, the attribute features extracted from the preprocessed user data include: In a guiding topic network, the guide usually has their own purpose and emotional inclination, and uses the concealment of user relationships and guiding intentions in the network to exert influence. Based on this, this paper extracts factors influencing the topic from two aspects: potential user relationships and emotional characteristics, as follows:

[0023] S21: Extract potential user relationships;

[0024] The process is mainly divided into three stages, such as Figure 2 As shown in the diagram. The first stage involves constructing a user-user relationship graph based on basic user attributes and the relationship network structure. The second stage involves constructing a user-topic-user relationship hypergraph based on user historical behavior and topic content information. The third stage involves performing random walks on the graph and hypergraph, combined with user similarity, to obtain node sequences representing complex user relationships.

[0025] In topic networks, the degree of similarity between users can effectively measure the relationships between them. Similar behaviors between users include following, sharing, replying, and mentioning. Generally, the more frequent the interaction or the more similar the behaviors between users, the higher their similarity. To calculate user similarity, different weight coefficients are assigned to different user behaviors; therefore, user similarity is defined as follows:

[0026]

[0027] Where, im(u i ,u j ) represents user u i and user u j The degree of similarity between them, Num[Link k (u i ,u j )] represents user u i and user u j The total number of interactions under k different behaviors, Num[Link] k u i ] and Num[Link k u j ] represents user u i and u j The total number of interactions with all users in the network under k different behaviors. k This represents the interaction weight of the k-th interaction method. In interaction behavior, mentioning a relationship is often considered to indicate a closer relationship than forwarding a message. W k The value is greater.

[0028] Based on user basic attribute characteristics Fea(u iConstruct a user-user relationship graph Gg, with the following basic user attribute features:

[0029] Fea(u i ) = [fans(u i ),follows(u i ),field(u i )]

[0030] Among them, fans(u i ) represents user u i fans, followers(u i ) represents user u i attention, field(u i ) represents user u i In the field of authentication, this value may be empty. These basic characteristics have a certain influence on users during the dissemination of topics.

[0031] Based on the user-topic rating matrix R N×M Constructing a hypergraph G H User u is taken as the superpoint and topic t as the superedge. Based on the scenario of social recommendation, a weight calculation method for the superpoint and superedge pairs is designed. The scoring formula ImplicitScore(u,i) is defined as follows:

[0032]

[0033] Wherein, matrix R N×M element R in u,t The rating represents user u's rating of topic t. The rating value can reflect the user's level of interest in the topic. Two users who give high ratings to the same topic have similar preferences. Such a pair of users are potential implicit friends and will co-occur more frequently in the random walk sequence.

[0034] In a normal graph, starting from the current node u i Starting with probability P ij Jump to another user node u j , where P ij With the edge weight w ij Proportional, defined as follows:

[0035]

[0036] Among them, δ(u i ) indicates that with user u i The set of directly connected user nodes.

[0037] In the hypergraph, when a random walk reaches a user node u i At that time, you can choose to jump to the page with u iAny related topic node t j Then jump from the topic node to t j Other related user nodes, redirection probability The definition is as follows:

[0038]

[0039] Where, δ t (u i This indicates that the topic node is used to connect with user u. i The set of connected user nodes This represents the weight of the superedge.

[0040] To utilize both explicit and implicit user relationships, the MixRandom strategy combines a hypergraph and a regular graph for random walks. Before transitioning from the current node to the next node, a probability parameter α (0 < α < 1) is introduced to control whether the transition occurs on the hypergraph or the regular graph, and the value of α is determined by the current user u. i and the next candidate user u j The similarity between them is used for adjustment. The specific steps of the hybrid random walk are as follows:

[0041] Set a similarity threshold θ, if the user similarity Sim(u) i ,u j If the similarity is greater than θ, it indicates high user similarity. The next step is to perform a random walk on the ordinary graph to explore the social relationships between similar users. Conversely, if the similarity is less than θ, it indicates low user similarity. In this case, more user behavior and other features should be considered to uncover implicit relationships. The next step is to perform a random walk on the hypergraph. If no new nodes are found during the walk, the termination condition is met.

[0042] By using a hybrid random walk, the user relationship node representation F can be obtained. u =[u1,u2,u3,...,u k ], where u k Let be the sequence of nodes for a random walk by user k, representing users in the network that are related to k.

[0043] S22: Extract user emotional characteristics;

[0044] The process mainly consists of three stages. The first stage involves analyzing topic content to calculate the grammatical and relative distances between aspects and context words, thereby achieving effective utilization of positional information. The second stage involves constructing a syntactic dependency tree and a graph attention network to fuse syntactic dependency information. The third stage involves establishing an interactive attention model between aspects and context words to achieve coordinated optimization between them.

[0045] Using the dependency parser in the Spacy text processing library to generate a syntactic dependency tree of a sentence, the grammatical distance can be obtained from the syntactic dependency tree, defined as d. i For aspect words w a and context words w i The syntactic distance between them is calculated using the following formula:

[0046]

[0047] Based on the above expression, we can obtain the grammatical distance feature vector S = (s1, s2, ..., s...) between aspects and context words in a sentence. n ), where d max s represents the maximum path length. i This represents the grammatical distance feature between aspects and context words, where 'a' represents the position index of the aspect word in the sentence. The sentence contains n words, with m aspects. The relative distance can be obtained from the path lengths of the aspects and context words in the sentence. The formula for calculating the relative distance is:

[0048]

[0049] The relative distance feature vector R = (r1, r2, ..., r) is obtained from this expression. n ), where r i The feature representing the relative distance between the aspect and the context words.

[0050] Combining syntactic distance features and relative distance features to generate a new distance feature representation is as follows:

[0051]

[0052] Where, p i This represents the distance features between aspects in the sentence and their context words. Finally, by introducing distance features to update the hidden state of the sentence, the newly generated sentence feature representation contains both semantic and positional information of the words. The update formula is:

[0053] h i =p i h i

[0054] The GAT layer updates node representations through a multi-head attention mechanism, capturing the syntactic dependencies between words.

[0055]

[0056] Where || denotes the concatenation operation, and N(i) represents the adjacent nodes of node i. W represents the weight coefficients calculated by the k-th attention head of the l-th layer GAT. lk and These are weight parameters. Let represent the hidden state vector of adjacent nodes, and LeakyReLU represent the activation function. The output of the k-layer GATs is then obtained.

[0057] Calculation of interaction attention weights β and Φ between aspect word a and context word c:

[0058]

[0059]

[0060] Among them, a i and c i It is a trainable parameter vector.

[0061] Weights are normalized using the softmax function:

[0062]

[0063] Weighted representation of context and aspect:

[0064]

[0065] The final result is a concatenation of aspects and contextual terms:

[0066] F e = [c; a]

[0067] In this embodiment, inputting hidden information into the key node tracing model includes two main stages: hidden information representation, generalized behavior discovery, and model algorithm design. In the first stage, starting from the complex relationship between user behavior and topic content on social platforms, the URE2vec algorithm is proposed. Knowledge representation is used to vectorize entities and embed them into a low-dimensional dense vector space to obtain a hidden information network. In the second stage, a graph convolutional network classification model, RE-GCN, based on user relationships and sentiment information, is established to analyze the influence of topic activities on users and, by integrating the above information, discover the key nodes affecting topic dissemination.

[0068] Specifically, it includes:

[0069] S31: Hidden Information Representation; The URE2vec algorithm first uses a knowledge representation learning algorithm to vectorize the user-relationship-sentiment feature triples, obtaining a vectorized user set. Then, it performs dimensionality reduction on the user set, embedding the users into a low-dimensional dense vector space to obtain the hidden information network.

[0070] S32: A key node discovery model for guiding topics based on RE-GCN;

[0071] Existing research on node influence rarely considers the complex social relationships and emotional intentions among users during topic dissemination. This paper combines the previously mentioned hidden relationships and emotional features, and utilizes an attention mechanism to fuse the node vector representations of both aspects of the network, proposing an influence discovery model based on hidden relationships and emotional intentions, such as... Figure 3 As shown in the figure. This paper defines the final discovery of user influence as a binary classification problem, namely, whether the user has influence.

[0072] The outputs of S21 and S22 yield vector representations of user relationships and sentiment features in the topic network. The URE2vec method is then used to extract information from the topic network and embed it into a low-dimensional, dense vector space. The output of the URW2vec algorithm is represented as follows:

[0073] F = [f1, f2, f3, ..., f k ]

[0074] Where F represents the user relationships and the user sentiment feature vector matrix in the topic network, f k This represents the vector representation of the k-th user node.

[0075] Because social networks are typically non-Euclidean structured data, traditional discrete convolutions are inadequate for processing them. Graph Convolutional Neural Networks (GCNs), based on graph theory, implement convolution operations on graphs. Therefore, this invention chooses GCNs to process guiding topic data. Considering the complex relationships and emotional guidance characteristics of guiding topics in their dissemination, a key node discovery model for relational sentiment information dissemination based on RE-GCNs is proposed. The objective of this application is to discover key user nodes in a topic. By judging the user's influence in the network, it determines whether the user is a key user node, thus transforming it into a binary classification task. The model input is as follows: Feature matrix X = N × F, where N represents the number of user nodes in the topic dissemination network, F is the input feature dimension of each node, and the features are information processed by URE2vec. The adjacency matrix under topic dissemination is also included.

[0076] In this application, a two-layer graph convolutional neural network with an intermediate Dropout layer is used as the key node classification model. First, the weights and biases are randomly initialized. Then, X is multiplied by W and the bias is added, and then... Multiplication is performed. Then, the ReLU function is used as the activation function for this layer, and Dropout is performed during model training. Finally, the softmax activation function is used to represent the convolution output as the probability values ​​of different node categories. The specific formula is expressed as follows:

[0077]

[0078] Among them, W i Let be the weight matrix corresponding to the i-th layer in the graph convolutional network.

[0079]

[0080] Since this paper discusses a binary classification problem, let the model output Z = C(a,b|u) i The specific definition is as follows:

[0081]

[0082] Where, if the corresponding Y=1, then determine user u i This refers to key nodes with influence in the topic; if Y=0, then determine user u. i These are key nodes in the topic that do not have influence.

[0083] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for tracing the key nodes of a topic on a social media platform, characterized in that, include: Acquire user data, which includes basic user information, user history, and topic information; Preprocess user data; Extract the attribute features of the preprocessed user data; use a hybrid random walk strategy to process the user data to obtain hidden user relationships; Sentiment analysis is performed on topic information in user data to obtain user sentiment characteristics; The process of processing user data using a hybrid random walk strategy includes: constructing a user-user relationship graph based on user attribute features and relationship network structure; constructing a user-topic-user relationship hypergraph based on the user-user relationship graph, user historical behavior, and topic information; calculating the similarity between users; and using the hybrid random walk strategy to perform random walks on the user-user relationship graph and the user-topic-user relationship hypergraph based on the similarity to obtain a node sequence of complex user relationships. Constructing a user-user relationship graph includes: obtaining basic user attribute features Fea(u i ), and based on the user's basic attribute features Fea(u i Construct a user-user relationship graph, with the following basic user attributes: Fea(u i )=[fans(u i ),follows(u i ),field(u i )] Among them, fans(u i ) represents user u i fans, followers(u i ) represents user u i attention, field(u i ) represents user u i The certification field; Based on the user-topic rating matrix R N×M Constructing a hypergraph G H User u is taken as the superpoint and topic t as the superedge. Based on the scenario of social recommendation, a weight calculation method for the superpoint and superedge pairs is designed. The scoring formula ImplicitScore(u,i) is defined as follows: Among them, R u,t This represents user u's rating of topic t; Sentiment analysis of topic information in user data includes: calculating the grammatical and relative distances between aspects and context words in the topic information; constructing a syntactic dependency tree and a graph attention network, and fusing the grammatical distances of aspect words and context words through the syntactic dependency tree and graph attention network to obtain grammatical distance features; fusing the relative distances of aspect words and context words through the syntactic dependency tree and graph attention network to obtain relative distance features; fusing the grammatical distance features and relative distance features to obtain new distance features; constructing an interaction attention model between aspects and context words; inputting the new distance features into the interaction attention model between aspects and context words to obtain user sentiment features; updating the hidden state of the sentence by introducing distance features, the newly generated sentence feature representation contains semantic and positional information of words; the update formula is: h g =p g h g The interactive attention model processes the input data in the following ways: the GAT layer updates node representations through a multi-head attention mechanism, capturing syntactic dependencies between words. Where || denotes the concatenation operation, and N(x) represents the adjacent nodes of node x. W represents the weight coefficients calculated by the k-th attention head of the l-th layer GAT. lk and These are weight parameters. Let represent the hidden state vector of adjacent nodes, and LeakyReLU represent the activation function; Calculation of interaction attention weights β and Φ between aspect word a and context word c: Among them, a e and c e It is a trainable parameter vector; Weights are normalized using the softmax function: Weighted representation of context and aspect: By combining aspects and contextual terms, user sentiment characteristics can be obtained. The URE2vec algorithm is used to vectorize user attribute features, hidden user relationships, and user sentiment features. The dimensionality of the vectorized representation is then reduced to obtain the hidden information. By inputting the hidden information into the key node tracing model, we can obtain the key nodes of topic propagation and the topic guidance intention; The key node tracing model processes hidden information by: inputting user information into the key node tracing model and extracting node features, relationship features, and sentiment features through multi-layer graph convolution operations; fusing the extracted node features, relationship features, and sentiment features and embedding them into a low-dimensional dense vector space; and classifying nodes into key nodes and ordinary nodes based on their feature representations.

2. The method for tracing the key nodes of a topic on a social platform according to claim 1, characterized in that, Preprocessing user data includes: obtaining raw data, deleting duplicate data and invalid information from the raw data, and obtaining structured data.

3. The method for tracing the key nodes of a topic on a social platform according to claim 1, characterized in that, User data attributes include: user behavior characteristics, social relationship characteristics, and topic content characteristics.

4. The method for tracing the key nodes of a topic on a social platform according to claim 1, characterized in that, The URE2vec algorithm is used to vectorize user attribute features, hidden user relationships, and user sentiment features. This includes: converting user attribute features, hidden user relationships, and user sentiment features into user-relationship-sentiment feature triples; using a knowledge learning algorithm to vectorize the user-relationship-sentiment feature triples; and embedding the vectorized representation into a low-dimensional dense vector space to obtain the hidden information.

5. The method for tracing the key nodes of a topic on a social platform according to claim 1, characterized in that, The key node tracing model adopts the RE-GCN network model.

Images