Social media hot topic prediction method, system and device and storage medium

Abstract

The invention discloses a social media hot topic prediction method, system and device, and a storage medium. The method comprises the following steps: obtaining a query post, a target user, a historical post set related to the target user, and a recent post set published by a followers of the target user; generating interest representation of the target user based on the historical post set; on the basis of the recent post set and in combination with the query post, a social exposure representation of the target user is generated by applying a self-attention shielding mechanism, and the social exposure representation reflects hot topic information related to the query post in the recent post set; fusing the user interest representation and the social contact exposure representation into a comprehensive representation; and predicting a possibility index of forwarding the query post by the target user based on the comprehensive representation. According to the invention, more accurate prediction of the user forwarding behavior is realized, and the sensitivity and prediction efficiency of hot topic identification are significantly improved.

Description

Technical Field

[0001] The present invention belongs to the field of public opinion monitoring and relates to a method, system, device and storage medium for predicting hot topics on social media. Background Art

[0002] Social media users post millions of microblogs daily on a variety of topics. Retweets play a crucial role in spreading topics on social media. Retweet prediction has garnered significant attention in recent years. Existing retweet prediction methods mostly focus on modeling user preferences using a variety of information, such as user profiles, user posting histories, and user following relationships. However, given that users may engage with hot topics discussed by their followers rather than being limited to their prior interests, the user's social exposure to their followers' real-time microblog posts significantly contributes to retweet prediction.

[0003] Current social media retweet prediction technologies primarily rely on static user profiles (such as registration information and historical posts) and explicit social relationships (such as follow lists) to build prediction models. Representative approaches include collaborative filtering-based user interest modeling and graph neural network-integrated social relationship reasoning. However, these approaches have significant limitations: First, traditional models struggle to effectively capture the real-time dynamics of a user's social circle and cannot reflect the impact of sudden, instantaneous hot topics generated by discussions among a user's supporters on their retweet decisions. Second, existing hierarchical attention mechanisms lack the ability to differentiate between heterogeneous information sources when processing historical user behavior and social exposure, resulting in inadequate fusion of features from real-time topics and long-term interests. Finally, most systems employ a fixed-time window sequence modeling approach, which is unable to adapt to the dynamic nature of non-stationary information flows in social media. These technical shortcomings result in reduced prediction accuracy and response delays in existing prediction systems when faced with the spread of sudden hot topics, making them difficult to meet the demands of real-time applications. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a social media hot topic prediction method, system, device and storage medium, which can achieve more accurate prediction of user forwarding behavior and significantly improve the sensitivity and prediction efficiency of hot topic identification.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for predicting hot topics on social media includes the following steps: Obtain the query post, the target user, a collection of historical posts related to the target user, and a collection of recent posts published by the target user's followers; Generate interest representation of target users based on historical post collection; Based on the recent post collection and the query post, a masked self-attention mechanism is applied to generate a social exposure representation of the target user. The social exposure representation reflects the hot topic information related to the query post in the recent post collection. Merge user interest representation and social exposure representation into a comprehensive representation; Based on the comprehensive representation, the likelihood indicator of the target user forwarding the query post is predicted.

[0006] Preferably, the process of generating the interest representation of the target user is: obtaining a set of historical posts related to the author of the query post; applying a hierarchical attention mechanism, combining the query post, the target user's historical post set and the author's historical post set, calculating the historical interest similarity between the user and the author, and using the similarity as or including it in the user interest representation.

[0007] Preferably, before generating the interest representation, the method further includes: using a bidirectional recurrent neural network to encode the query post, the posts in the historical post collection, and the posts in the recent post collection to obtain an initial feature representation of the post.

[0008] Preferably, the process of applying the masked self-attention mechanism to generate a social exposure representation is: calculating the relevance score of each post in the recent post set with the query post; generating a mask based on the relevance score; and when applying self-attention calculation to the recent post set, combining the mask to focus on the content of posts related to the query post, thereby extracting the hot topic information.

[0009] Preferably, a masked self-attention mechanism is applied to process the recent post set using at least one universal encoder, and the relevance score is calculated by combining the output of the universal encoder and the query post.

[0010] Preferably, the process of fusing the user interest representation and the social exposure representation into a comprehensive representation is: splicing the user interest representation and the social exposure representation; inputting the spliced ​​representation into at least one multi-layer perceptron for processing to obtain the comprehensive representation.

[0011] Preferably, the process of predicting the likelihood index of the target user forwarding the query post is: inputting the comprehensive representation into a classification layer; and outputting a value representing the probability of forwarding or not forwarding through the classification layer as the likelihood index.

[0012] A social media hot topic prediction system, comprising: A data acquisition module is used to obtain query posts, target users, a collection of historical posts related to the target user, and a collection of recent posts published by followers of the target user; The interest representation generation module is used to generate the interest representation of the target user based on the historical post collection; A social exposure representation generation module is used to generate a social exposure representation of the target user based on a recent post collection and in combination with a query post, applying a masked self-attention mechanism. The social exposure representation reflects hot topic information related to the query post in the recent post collection; Comprehensive representation fusion module, used to fuse user interest representation and social exposure representation into a comprehensive representation; The prediction module is used to predict the likelihood indicator of the target user forwarding the query post based on the comprehensive representation.

[0013] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the social media hot topic prediction method are implemented.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the social media hot topic prediction method.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention achieves significant technological advancements in the task of predicting the amount of forwarding by employing a shielded self-attention model (SSAM) and a masked self-attention mechanism. Combining user interest perception with hot topic perception, a hierarchical attention mechanism is utilized to comprehensively model user interest preferences and social exposure. At the same time, hot topic features are extracted from the most recent microblog posts by users' followers, rapidly capturing current hot topics and their social influence, thereby significantly improving the accuracy of forwarding volume prediction. Compared to traditional methods that rely solely on historical user behavior or topics of interest, this invention takes into account both individual interests and group behavior, more accurately reflecting dynamic changes in user interests and sensitivity to sudden hot events. It also possesses excellent versatility and scalability, making it applicable to a variety of social media scenarios. Furthermore, by shielding the self-attention mechanism and optimizing the feature extraction process, computational complexity is effectively reduced and operational efficiency is improved, providing reliable support for practical applications such as content recommendation, advertising placement, and public opinion monitoring. It also facilitates social hot topic research and public decision-making, possessing significant technical value and social significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the social media hot topic prediction method of the present invention; Figure 2 Schematic diagram of the stacked self-attention mechanism model; Figure 3 Schematic diagram of the masked attention mechanism used in this invention. DETAILED DESCRIPTION

[0017] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0018] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like, indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention pertains. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, they can be fixedly connected, detachably connected, or integrally connected; they can be mechanically connected, electrically connected, or able to communicate with each other; they can be directly connected, or indirectly connected through an intermediate medium, or they can be internally connected between two elements or an interactive relationship between two elements. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0020] In the present invention, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may include the first and second features being in direct contact, or may include the first and second features being in contact not directly but through another feature between them. Furthermore, a first feature being "above," "above," and "above" a second feature may include the first feature being directly above or obliquely above the second feature, or may simply mean that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may include the first feature being directly below or obliquely below the second feature, or may simply mean that the first feature is lower in level than the second feature.

[0021] The disclosure below provides many different embodiments or examples for realizing different structures of the present invention. In order to simplify the disclosure of the present invention, the components and settings of specific examples are described below. Of course, they are merely examples and are not intended to limit the present invention. In addition, the present invention may repeat reference numbers and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or settings discussed. In addition, the present invention provides examples of various specific processes and materials, but those skilled in the art will recognize the application of other processes and / or the use of other materials.

[0022] The proposed social media hot topic prediction method utilizes an innovative masked self-attention model (SSAM) framework to achieve dynamic forwarding prediction by integrating user interest perception with the hot topic perception of the user's supporter group. The core modules include a microblog content encoder, a hierarchical attention network, a masked self-attention module, and a fusion prediction layer. During the initialization phase, each microblog is encoded using a bidirectional long short-term memory (Bi-LSTM) network to generate word-level and sentence-level feature representations. Subsequently, a hierarchical attention mechanism is used to calculate the historical interest similarity vector between the author and the user. , and design a masked self-attention mechanism to construct a social exposure vector for hot topic perception The masked self-attention module filters social exposure content and captures hot topics through a universal stacked encoder, a mask generator, and a mask encoder stack, combined with query microblogs. and After concatenation, the data is fed into a multi-layer perceptron, and a softmax classifier is used to predict the user's forwarding probability. The model is optimized using a stochastic gradient descent algorithm with the Adam strategy, with the objective function minimizing the negative log-likelihood loss. This dual-channel parallel processing mechanism preserves users' long-term interests while capturing emerging hot topics in social networks in real time, effectively overcoming the technical bottlenecks of traditional methods in processing dynamic information flows and integrating heterogeneous features.

[0023] like Figure 1As shown in FIG, the social media hot topic prediction method of the present invention includes the following steps: S1, obtain the query post, the target user, the collection of historical posts related to the target user, and the collection of recent posts published by the followers of the target user.

[0024] S,2,generates the interest representation of the target user based on the historical post collection.

[0025] Obtain a collection of historical posts related to the author of the query post; apply a hierarchical attention mechanism, combine the query post, the target user's historical post collection, and the author's historical post collection, calculate the historical interest similarity between the user and the author, and use the similarity as or include it in the user interest representation.

[0026] Before generating the interest representation, a bidirectional recurrent neural network is used to encode the query post, the posts in the historical post collection, and the posts in the recent post collection to obtain an initial feature representation of the post.

[0027] S3, based on the recent post collection and combined with the query post, applies the masked self-attention mechanism to generate the social exposure representation of the target user, which reflects the hot topic information related to the query post in the recent post collection.

[0028] Calculating a relevance score between each post in the recent post set and the query post; generating a mask based on the relevance score; and combining the mask with the self-attention calculation to focus on the content of posts related to the query post when applying the self-attention calculation to the recent post set, thereby extracting the hot topic information.

[0029] A masked self-attention mechanism is applied to process the recent post set using at least one universal encoder, and the relevance score is calculated by combining the output of the universal encoder and the query post.

[0030] S4, integrates user interest representation and social exposure representation into a comprehensive representation.

[0031] The user interest representation and the social exposure representation are spliced ​​together; the spliced ​​representation is input into at least one multi-layer perceptron for processing to obtain the comprehensive representation.

[0032] S5, based on the comprehensive representation, predicts the likelihood indicator of the target user forwarding the query post.

[0033] The process of predicting the likelihood index of the target user forwarding the query post is as follows: inputting the comprehensive representation into a classification layer; and outputting a value representing the probability of forwarding or not forwarding through the classification layer as the likelihood index.

[0034] The present invention combines user interest perception with hot topic perception of user supporter groups through a hierarchical attention mechanism, constructs feature expressions from multiple dimensions such as user posting history, real-time discussion content of user supporters, and social exposure rate, and uses word-level and Weibo-level attention feature vectors to capture the dynamic changes and social influence of hot topics, thus breaking through the limitation of traditional methods that only rely on static user interests, achieving more accurate prediction of user forwarding behavior, and significantly improving the sensitivity and prediction efficiency of hot topic identification.

[0035] The specific process of the above method is: A novel masked self-attention model is designed to learn from users' forwarding behavior, leveraging a hierarchical attention mechanism to model user interests. This model simultaneously considers the content of the microblog, the author's history, the forwarding user's historical interests, and the forwarding user's social exposure. It also accounts for user interests and the hot topics followed by their followers, enabling better prediction of microblog forwarding volume.

[0036] 1), Problem definition.

[0037] Given a query Weibo and a Weibo user The task of this invention is to predict the user's Will you forward this Weibo? The three sets are described as: Post query microblog author Home page's historical Weibo posts , which can express the author's interests.

[0038] Weibo users Historical Weibo posts , can represent the user interest.

[0039] Real-time microblogs posted by the user's followers , which may contain the query of Weibo author and Weibo users Interests not related to hot topics. Indicates the number of microblogs in each collection.

[0040] The social exposure of each user can be regarded as the information flow within a small community. Users in different communities may have common topics due to common interests. The social exposure of a user is composed of microblogs posted by different user supporters in real time. The resulting information flow does not have time series consistency. Therefore, a self-attention mechanism with a maximum length of 1 can be used to construct contextual expressions. In order to model the historical interest similarity between users and authors, the query microblog is introduced. , for the author's history and user history Establish a hierarchical mechanism. Based on querying microblogs , Author Release History , user publishing history and user social exposure , the output layer can represent the probability of user forwarding behavior.

[0041] 2), algorithm model.

[0042] This paper will introduce the basic framework of the method of the present invention in detail. The overall structure of the model is as follows Figure 1 As shown in the figure, the masked self-attention mechanism is the gray module (Masked Self-Attentive) in the computational structure on the right side of the figure. First, the present invention uses a bidirectional long short-term memory network (Bi-LSTM) to encode each Weibo post. In order to capture the semantic representation of different levels, the hidden vector corresponding to each Weibo post in the author's posting history and each Weibo post in the user's posting history is retained, but the last hidden vector is used to query the Weibo post. and user social exposure Every microblog in the author's post history and user publishing history After multi-hop hierarchical attention, the historical interest similarity vector can be obtained At the same time, the proposed masked self-attention mechanism constructs the social exposure vector of hot topic perception Then the splicing layer is applied to combine the historical interest similarity and perceived social exposure of hot topics Finally, a fully connected softmax layer is used to predict the forwarding behavior.

[0043] For each input Weibo post, the word vectors for each word in the Weibo post are obtained by querying the word vector list. This list is trained on massive Weibo data using the GloVe unsupervised method. In this way, a word-level Weibo feature representation can be obtained: (1) in is the maximum length of a microblog. Next, we use bidirectional long short-term memory units and word vector concatenation to obtain the sentence-level feature expression of the microblog: (2) (3) in, and Respectively represent the forward and backward LSTM at time steps At each time step, the two hidden state vectors are concatenated to construct a set of text feature vectors : (4) In the formula For the The context feature vector of each word in the entire sentence. Considering that the following two attention mechanisms focus on semantic representations at different levels, the present invention specifically stores all the implicit vectors of Bi-LSTM and the implicit vector of the last time step. The last implicit vector is used to construct the query microblog and all Weibo posts exposed by users on social media Feature expression (Weibo-level vector).

[0044] In order to construct a high-level expression between the similarity of author and user interests, the user history and author history memory are superimposed Hierarchical attention mechanism. The feature representation of the query microblog As Initial query vector for the layer-stacked hierarchical attention mechanism Then, use To construct the attention probability distribution of the author and user's historical Weibo, and construct the next query vector . The output of the hierarchical attention mechanism constitutes the final interest similarity vector .

[0045] like Figure 3 As shown in the figure, the masked attention mechanism is a special design structure of the model. The following will focus on the internal calculation process of this structure. The key challenge in the learning process of this calculation structure is how to model the hot topics contained in the recent Weibo posts of user supporters. Considering that the content recently posted by user supporters has discontinuous characteristics, it is not appropriate to use a sequence model like a recurrent neural network to construct context information. This project uses a self-attention mechanism with a maximum length of 1 to learn the user supporters. Characteristic expressions of recent Weibo.

[0046] First, this paper introduces the computational approach of the Transformer, as it is a key component of the masked self-attention mechanism. This paper focuses on the scaled dot product attention, which is the basic computational structure of the Transformer. Given a query , a set of keys Sum ,in , scaled dot product attention output weighted sum value , where the attention probability is given by the query and key The dot product of is obtained. The attention calculation formula is as follows: (5) The attention of the bulls is The multi-head attention model is composed of parallel scaled dot product attention layers, called "heads". Each "head" is an independent dot product attention. Multi-head attention enables the model to jointly focus on information from different representation subspaces at different locations. The output of the multi-head attention is as follows: (6) (7) in, is the independent head mapping matrix, ,and .

[0047] like Figure 2 As shown in the figure, the present invention uses two parallel stacked encoders to construct topic-aware representations in the user's social exposure. The masked self-attention module consists of three components: a general stacked encoder, a mask generator, and a mask encoder stack. The main task of these two parallel stacked encoders is to focus on learning contextual representations. The mask encoder multiplies the input of each encoder layer by a mask matrix to obtain hot topics. The mask matrix is ​​represented by the output of the general stacked encoder and the query microblog. The general encoder is composed of The stack consists of 100 identical layers. Each layer consists of a multi-head self-attention layer and a fully connected feedforward network structure. Residual connections and layer normalization are introduced next to every two sub-layers. The fully connected feedforward network consists of two linear transformations with a ReLU activation function in the middle: (8) in ,and The fully connected feedforward network can be viewed as two convolutions with a kernel size of 1. In other words, FFN can add nonlinear features while keeping the feature map scale unchanged. The layer formula is as follows: (9) (10) in, represents the i-th layer output of the same encoder, The social exposure rate is represented by the matrix Initialization. The difference is that the additional mask matrix helps capture the popular topics recently posted by the user’s advocates. The mask generator module is based on the feature representation of social exposure microblogs and combined with the query microblog To calculate the relevant probability of each social exposure Weibo : (11) in The sigmoid activation function limits the value of each channel to between 0 and 1. is the last layer output of the universal encoder stack. After the above calculation, the present invention uses the query microblog Query the output of the mask encoder and generate hot topic representations based on the attention probability distribution : (12) (13) (14) The present invention expresses the historical interest similarity obtained in the above process and hot topic expressions The final feature expression is obtained by concatenating the features of Finally, a multi-layer perceptron and a single-layer softmax classifier are used to determine whether the user will forward the query microblog: (15) Among them, MLP is a multi-layer perceptron. is the nonlinear activation function sigmoid. The final prediction formula is as follows: (16) In the formula It is The class weight vector, .

[0048] 3) Model optimization.

[0049] The model objective function is as follows: (17) in, For the training set. It is a quadruple When the label Represents a user Retweet query Weibo ; Represents a user Do not forward. is the entire parameter set of the model.

[0050] The optimization algorithm uses the stochastic gradient descent algorithm based on the Adam strategy.

[0051] The following describes the technical implementation process of the present invention in detail through specific implementation cases: On the Weibo platform, when user "Digital Expert Xiao Zhang" posts an original Weibo post about "New Technology for Folding Screen Phones," the goal is to predict whether his follower, "Tech Enthusiast Li Si," will retweet the post. First, the query Weibo post is preprocessed: pre-trained GloVe word vectors (dimension 300) are used to convert the text into word-level feature representations. These are then fed into a bidirectional LSTM network (hidden layer dimension 256) to generate context-aware sentence-level feature vectors.

[0052] To model user Li Si's historical behavior, we extracted his past 100 Weibo posts and processed them using a three-layer hierarchical attention mechanism. Using the query Weibo's feature vector as the initial query anchor, we calculated semantic similarity with historical Weibo posts layer by layer. The first layer focused on historical content related to "smartphone technology," the second layer captured relevance to the topic of "electronic product reviews," and the third layer extracted deeper interest features such as "technology exhibition participation," ultimately generating a user interest similarity vector.

[0053] During the social exposure processing phase, real-time Weibo data from the last 24 hours of posts by Li Si's 500 followers was collected. A universal encoder stack (consisting of two Transformer encoding layers, 8 heads, and a feed-forward network dimension of 512) was used to extract global feature representations. A mask generator dynamically calculated the relevance probability of each follower's Weibo post, combining the query Weibo content. This process assigns a high weight (with a probability of 0.8-1.0) to posts containing keywords such as "folding screen" and "flexible screen technology," while suppressing irrelevant entertainment content (with a probability of 0.1-0.3). After processing with the mask encoder, hot topic features such as "flexible screen material breakthrough" and "foldable screen phone durability test" were identified.

[0054] During the feature fusion stage, the user interest similarity vector (dimension 128) and the hot topic perception vector (dimension 64) were concatenated and subjected to a nonlinear transformation using a three-layer MLP (with 192, 64, and 16 neurons in each layer, respectively). The softmax classifier ultimately outputted a forwarding probability of 0.78, indicating that Li Si had a high likelihood of forwarding the post. Actual verification showed that the user did indeed forward the post within two hours, confirming the accuracy of the model's predictions.

[0055] During model training, the Adam optimizer (learning rate 0.001, β1=0.9, β2=0.999) was used on a dataset containing 100,000 annotated samples. Dynamically adjusting the attention weight parameters through backpropagation achieved an accuracy of 82.3% and an F1 score of 0.81 on the validation set, a 12.6 percentage point improvement over the traditional LSTM model. This implementation demonstrates the effectiveness of this invention in dynamic social scenarios, particularly in capturing the coupling effect between emerging hot topics (such as the release of new mobile phone technology) and long-term user interests (such as technology product reviews).

[0056] The following are device embodiments of the present invention, which can be used to perform the method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0057] In another embodiment of the present invention, a social media hot topic prediction system is provided. The social media hot topic prediction system can be used to implement the above-mentioned social media hot topic prediction method. Specifically, the social media hot topic prediction system includes a data acquisition module, an interest representation generation module, a social exposure representation generation module, a comprehensive representation fusion module and a prediction module.

[0058] The data acquisition module is used to obtain query posts, target users, a collection of historical posts related to the target user, and a collection of recent posts published by followers of the target user.

[0059] The interest representation generation module is used to generate interest representations of target users based on a collection of historical posts.

[0060] The social exposure representation generation module is used to generate the social exposure representation of the target user based on the recent post collection and in combination with the query post, applying the masked self-attention mechanism. The social exposure representation reflects the hot topic information related to the query post in the recent post collection.

[0061] The comprehensive representation fusion module is used to fuse user interest representation and social exposure representation into a comprehensive representation; The prediction module is used to predict the likelihood index of the target user forwarding the query post based on the comprehensive representation.

[0062] In yet another embodiment of the present invention, a terminal device is provided. The terminal device includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal and is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to implement corresponding method processes or corresponding functions. The processor described in an embodiment of the present invention can be used for the operation of a social media hot topic prediction method, including: obtaining a query post, a target user, a collection of historical posts related to the target user, and a collection of recent posts published by followers of the target user; generating an interest representation of the target user based on the collection of historical posts; generating a social exposure representation of the target user based on the collection of recent posts and in combination with the query post, applying a masked self-attention mechanism, wherein the social exposure representation reflects hot topic information related to the query post in the collection of recent posts; fusing the user interest representation and the social exposure representation into a comprehensive representation; and predicting a likelihood indicator of the target user forwarding the query post based on the comprehensive representation.

[0063] In another embodiment, the present invention further provides a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium herein may include both built-in storage media in the terminal device and, of course, extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device.

[0064] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the social media hot topic prediction method in the above-mentioned embodiment; one or more instructions in the computer-readable storage medium are loaded by the processor and execute the following steps: obtaining a query post, a target user, a collection of historical posts related to the target user, and a collection of recent posts published by the target user's followers; generating an interest representation of the target user based on the collection of historical posts; based on the collection of recent posts and in combination with the query post, applying a masked self-attention mechanism to generate a social exposure representation of the target user, the social exposure representation reflecting the hot topic information related to the query post in the collection of recent posts; fusing the user interest representation and the social exposure representation into a comprehensive representation; and based on the comprehensive representation, predicting a possibility index of the target user forwarding the query post.

[0065] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0069] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0070] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0071] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0072] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0073] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

[0074] It should be understood that the above description is for illustration and not for limitation. Many embodiments and many applications beyond the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this patent should not be determined with reference to the above description, but rather with reference to the preceding claims and the full scope of equivalents to which such claims are entitled. For the purpose of completeness, all articles and references, including disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein from the preceding claims is not a disclaimer of such subject matter, nor should it be considered that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

Claims

1. A method for predicting hot topics on social media, characterized in that: The following processes are included: Obtain the query post, the target user, a collection of historical posts related to the target user, and a collection of recent posts published by the target user's followers; Generate interest representation of target users based on historical post collection; Based on the recent post collection and the query post, a masked self-attention mechanism is applied to generate a social exposure representation of the target user. The social exposure representation reflects the hot topic information related to the query post in the recent post collection. Merge user interest representation and social exposure representation into a comprehensive representation; Based on the comprehensive representation, the likelihood indicator of the target user forwarding the query post is predicted.

2. The method for predicting hot topics in social media according to claim 1, characterized in that: The process of generating the interest representation of the target user is as follows: obtaining a set of historical posts related to the author of the query post; applying a hierarchical attention mechanism, combining the query post, the historical post set of the target user, and the historical post set of the author, calculating the historical interest similarity between the user and the author, and using the similarity as or including it in the user interest representation.

3. The method for predicting hot topics in social media according to claim 2, characterized in that: Before generating the interest representation, the method further includes: using a bidirectional recurrent neural network to encode the query post, the posts in the historical post collection, and the posts in the recent post collection to obtain an initial feature representation of the post.

4. The method for predicting hot topics in social media according to claim 1, wherein: The process of applying the masked self-attention mechanism to generate social exposure representation is as follows: calculating the relevance score of each post in the recent post set and the query post; A mask is generated based on the relevance score; and when applying self-attention calculation to the recent post set, the mask is combined to focus on the post content related to the query post, thereby extracting the hot topic information.

5. The method for predicting hot topics in social media according to claim 4, characterized in that: A masked self-attention mechanism is applied to process the recent post set using at least one universal encoder, and the relevance score is calculated by combining the output of the universal encoder and the query post.

6. The method for predicting hot topics in social media according to claim 1, wherein: The process of fusing the user interest representation and the social exposure representation into a comprehensive representation is as follows: concatenating the user interest representation and the social exposure representation; The concatenated representations are input into at least one multi-layer perceptron for processing to obtain the comprehensive representation.

7. The method for predicting hot topics in social media according to claim 1, wherein: The process of predicting the likelihood index of the target user forwarding the query post is as follows: inputting the comprehensive representation into a classification layer; and outputting a value representing the probability of forwarding or not forwarding through the classification layer as the likelihood index.

8. A social media hot topic prediction system, characterized by: include: A data acquisition module is used to obtain query posts, target users, a collection of historical posts related to the target user, and a collection of recent posts published by followers of the target user; The interest representation generation module is used to generate the interest representation of the target user based on the historical post collection; A social exposure representation generation module is used to generate a social exposure representation of the target user based on a recent post collection and in combination with a query post, applying a masked self-attention mechanism. The social exposure representation reflects hot topic information related to the query post in the recent post collection; Comprehensive representation fusion module, used to fuse user interest representation and social exposure representation into a comprehensive representation; The prediction module is used to predict the likelihood indicator of the target user forwarding the query post based on the comprehensive representation.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the social media hot topic prediction method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the social media hot topic prediction method according to any one of claims 1 to 7 are implemented.

Images