Method and device for estimating social network forwarding behavior based on large model enhancement

Abstract

The embodiment of the present application discloses a method and device for estimating social network forwarding behavior based on large model enhancement, wherein the method includes: using a social media content body encoder to extract a vector representation of text information of candidate social media content to obtain a first feature vector of the candidate social media content; using a user encoder to extract a vector representation of historical forwarding information to obtain a second feature vector of the user; generating prompt information based on model control instructions, historical forwarding information, social network platform account ID and user ID, and inputting the prompt information into a large language model for processing to obtain an intermediate result; splicing the first feature vector, the second feature vector and the intermediate result and inputting them into a multi-layer perceptron, and the multi-layer perceptron estimates the user's forwarding behavior for the candidate social media content to obtain an estimated result. This solution effectively improves the accuracy of estimating user forwarding behavior.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data analysis technology, and specifically to a method and device for estimating social network forwarding behavior based on large model enhancement. Background Art

[0002] With the widespread popularity of social networking platforms, user-generated content is growing rapidly, and other accompanying social network data (such as forwarding behavior data, etc.) is also increasing. It is very necessary to study and estimate the forwarding behavior of users in social networks, which can not only enhance the overall user experience of social networks, but also help study the propagation mechanism in social networks, for example, helping to track the propagation path of hot social media content, thereby revealing the propagation mode of hot social media content.

[0003] In the prior art, social network forwarding behavior prediction models are mainly based on graph convolutional neural networks or attention networks, which make predictions by capturing text content and user features. However, the semantic understanding capabilities of these network structures are limited, resulting in low accuracy in estimating social network forwarding behavior. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a method and device for estimating social network forwarding behavior based on large model enhancement, which overcomes the above problems or at least partially solves the above problems.

[0005] According to one aspect of an embodiment of the present application, a method for estimating social network forwarding behavior based on large model enhancement is provided, comprising:

[0006] Obtain candidate social media content, user's historical forwarding information, social network platform account ID, and user ID;

[0007] Extracting a vector representation of text information of the candidate social media content using a social media content body encoder to obtain a first feature vector of the candidate social media content;

[0008] The user encoder is used to extract the vector representation of the historical forwarding information to obtain the second feature vector of the user;

[0009] Generate prompt information based on model control instructions, historical forwarding information, social network platform account ID and user ID, and input the prompt information into the large language model for processing to obtain intermediate results;

[0010] The first eigenvector, the second eigenvector and the intermediate result are concatenated and input into a multi-layer perceptron, which estimates the user's forwarding behavior for the candidate social media content to obtain an estimated result.

[0011] According to another aspect of an embodiment of the present application, a device for estimating social network forwarding behavior based on large model enhancement is provided, comprising:

[0012] An acquisition module, adapted to acquire candidate social media content, a user's historical forwarding information, a social network platform account ID, and a user ID;

[0013] The first processing module is adapted to extract a vector representation of text information of candidate social media content using a social media content body encoder to obtain a first feature vector of the candidate social media content; and to extract a vector representation of historical forwarding information using a user encoder to obtain a second feature vector of the user;

[0014] The second processing module is adapted to generate prompt information according to the model control instruction, historical forwarding information, social network platform account ID and user ID, and input the prompt information into the large language model for processing to obtain an intermediate result;

[0015] The estimation module is suitable for splicing the first feature vector, the second feature vector and the intermediate result and inputting them into a multi-layer perceptron, and the multi-layer perceptron estimates the user's forwarding behavior for the candidate social media content to obtain an estimation result.

[0016] According to another aspect of the embodiment of the present application, there is provided a computing device, including: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0017] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned social network forwarding behavior prediction method based on large model enhancement.

[0018] According to another aspect of the embodiments of the present application, a computer storage medium is provided, in which at least one executable instruction is stored, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned method for estimating social network forwarding behavior based on large model enhancement.

[0019] According to another aspect of the embodiments of the present application, a computer program product is provided, including at least one executable instruction, which enables a processor to perform operations corresponding to the above-mentioned method for estimating social network forwarding behavior based on large model enhancement.

[0020] According to the method and device for estimating forwarding behavior in a social network based on large model enhancement provided by the embodiment of the present application, a large language model is applied to the forwarding behavior estimation task, and user-related ID sequence information is injected into the large language model, social media content and user-related ID sequence information, etc. are integrated, and the powerful semantic understanding ability and extensive pre-training knowledge of the large language model are utilized to align it with the estimation task through fine-tuning, thereby helping to obtain better estimation results; the estimation ability of user forwarding behavior in a social network is enhanced by the large language model, and the estimation accuracy of user forwarding behavior is effectively improved; and the large language model has good interpretability for the estimation results, and the obtained estimation results are returned to the large language model, and explanation information can also be generated, which effectively enhances the interpretability of forwarding behavior estimation in the social network; the solution can be deployed in a computer or server, etc., and can analyze user forwarding behavior data in a social network platform, improve user experience, and help study the propagation mechanism in the social network.

[0021] The above description is only an overview of the technical solution of the embodiment of the present application. In order to more clearly understand the technical means of the embodiment of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiment of the present application more obvious and easy to understand, the specific implementation method of the embodiment of the present application is listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only used for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the embodiments of the present application. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0023] Figure 1 A schematic diagram of a process of estimating a social network forwarding behavior based on large model enhancement according to an embodiment of the present application is shown;

[0024] Figure 2 A schematic diagram of the model architecture corresponding to the social network forwarding behavior prediction method based on large model enhancement is shown;

[0025] Figure 3 A structural block diagram of a social network forwarding behavior prediction device based on large model enhancement according to an embodiment of the present application is shown;

[0026] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0028] Figure 1 A schematic diagram of a process of estimating a social network forwarding behavior based on a large model enhancement according to an embodiment of the present application is shown. Figure 1 As shown, the method comprises the following steps:

[0029] Step S101, obtaining candidate social media content, user's historical forwarding information, social network platform account ID and user ID.

[0030] Considering that the Large Language Model (LLM) has rich knowledge, deep semantic understanding capabilities, and has been proven to be effective in various natural language processing tasks, this application proposes a novel social network forwarding behavior prediction method, which introduces LLM into the estimation processing of social network forwarding behavior. By processing social media content (such as social media content historically forwarded by users, etc.) and user-related ID sequence information through LLM, the estimation potential is effectively enhanced by using LLM, and the estimation accuracy of user forwarding behavior is effectively improved. In addition, the model has good interpretability of the estimation results.

[0031] In order to accurately predict the forwarding behavior of the user for the candidate social media content, it is necessary to obtain the candidate social media content, the user's historical forwarding information, the social network platform account ID and the user ID. Among them, the candidate social media content may include candidate microblogs, etc., and may specifically include their text information and topic information, etc. The number of candidate social media content may be one or more, which is not limited here; the user's historical forwarding information may include multiple social media content forwarded by the user in history, etc.; the social network platform account ID is the identification information used by the social network platform to uniquely identify the user's social network platform account, which is used to distinguish different social network platform accounts, such as microblog ID, etc. The social network platform account ID is generally the identification information (such as a string) automatically assigned by the social network platform to the user. Some social network platforms also allow users to customize their social network platform account ID; the user ID is the identification information used to identify the user, such as the nickname, user name, mobile phone number, etc. set by the user. The social network platform account ID and the user ID belong to the user-related ID sequence information, which contains a lot of collaborative information and helps to enhance the accuracy of the forwarding behavior prediction.

[0032] Step S102: using a social media content body encoder to extract a vector representation of text information of the candidate social media content, and obtaining a first feature vector of the candidate social media content.

[0033] Figure 2 The schematic diagram of the model architecture corresponding to the social network forwarding behavior prediction method based on large model enhancement is shown as follows: Figure 2 As shown, the forwarding behavior prediction model architecture of this embodiment includes multiple parts, specifically including a first prediction part, a second prediction part and a model fusion part, etc. Among them, the first prediction part includes a social media content body encoder and a user encoder.

[0034] The social media content body encoder is used to learn the representation of the content body from the text information of social media content with different semantic scales. The social media content body encoder may include: a text encoder, an element encoder, and an attention-based pooling layer. In step S102, the text encoder is used to learn the body representation from the body text information of the candidate social media content, and the element encoder is used to learn the body representation from the topic information of the candidate social media content. The semantic information of the body with different semantic qualities is captured through the attention-based pooling layer to obtain the first feature vector of the candidate social media content, that is, the candidate social media content vector representation.

[0035] Specifically, the text encoder is used to learn the text representation from the text information of social media content. For example, the text information is: "When can we really ride in a driverless car anytime and anywhere? How can every "traveler" travel fast and not miss the beautiful scenery along the way?" The text encoder may include three layers. The first layer is the word embedding layer, which uses Glove (Global Vectors for Word Representation) to obtain each word and the corresponding vector representation, and converts the text into a low-dimensional word vector; the second layer is the convolutional neural network layer, which focuses on the local context of the word; the third layer is an additive attention network, which can focus on more comprehensive important information from a global perspective.

[0036] The element encoder is used to learn the text representation from the topic information of social media content, such as the topic information "driverless landing". The element encoder can include two layers: the topic ID embedding layer and the dense layer. Each neuron in the dense layer receives the output of all neurons in the previous layer as input, and performs a weighted sum of these inputs, adds a bias term, and then converts it into a nonlinear output through an activation function (such as ReLU, sigmoid, or tanh, etc.).

[0037] Finally, an attention-based pooling layer is used to capture the body semantic information of different semantic qualities to learn accurate body representation of social media content.

[0038] Step S103: using a user encoder to extract a vector representation of the historical forwarding information to obtain a second feature vector of the user.

[0039] The user encoder includes a multi-head self-attention network and an additive attention network. The user encoder applies the multi-head self-attention network and the additive attention network in sequence. The final user vector representation (i.e., the second eigenvector) is the weighted sum of the representations of the social media content forwarded by the user in history, which can be used to reflect which social media content the user likes.

[0040] Step S104, generating prompt information according to the model control instruction, historical forwarding information, social network platform account ID and user ID, and inputting the prompt information into the large language model for processing to obtain an intermediate result.

[0041] like Figure 2 As shown, the main body of the second estimation part uses a large language model, specifically a large language model after parameter fine-tuning using the Low-Rank Adaptation (LoRA) method. Specifically, the ChatGLM3-6B model after parameter fine-tuning using the LoRA method is named LLMForward. Among them, ChatGLM3 is a dialogue pre-training model jointly released by Zhipu AI and Tsinghua University KEG Laboratory, which can improve performance while using less computing resources and data. LoRA allows some dense layers in the neural network to be indirectly trained by optimizing the rank decomposition matrix of the dense layer changes during the adaptation process, while keeping the pre-trained weights frozen. During training, the calculation method of the weight update process is as follows:

[0042] W updated =W+ΔW;

[0043] Among them, W updated represents the updated weight matrix; W represents the pre-trained weight matrix; ΔW represents the cumulative gradient update during the adaptation process. In the scenario of using a large language model, ΔW contains a huge number of parameters. Using LoRA saves memory and computing resources by decomposing the weight change ΔW into a low-rank representation.

[0044] In this embodiment, the prompt information (ie, Prompt) can be generated using the template officially provided by ChatGLM3, as shown below:

[0045] <|system|>

[0046] You are an intelligent AI assistant, your name is ChatGLM3. Please follow the user's instructions carefully.

[0047] <|user|>

[0048] Given a user's ID sequence information and some historically forwarded social media content and a candidate social media content. Historically forwarded social media content is separated by "\", and candidate social media content starts with "#". Please predict whether the user will forward this candidate social media content. If yes, output 1; if not, output 0. Please note that the output is only 0 or 1.

[0049] <|assistant|>

[0050] User-related ID sequence information may include social network platform account ID and user ID. The <|user|> part may include model control instructions in natural language form, historical forwarding information of the user (such as historical forwarded social media content), and ID sequence information such as social network platform account ID and user ID. Since the large language model cannot directly understand the ID sequence information containing collaborative information, the embodiment of the present application first maps the low-dimensional ID embedding to the same dimension as the word embedding of the large language model through a linear layer. Specifically, the social network platform account ID and user ID are mapped to a preset dimension (such as 4096 dimensions) that conforms to the word embedding of the large language model, and the third feature vector of the social network platform account ID and the fourth feature vector of the user ID are obtained. After obtaining the third feature vector of the social network platform account ID and the fourth feature vector of the user ID, prompt information can be generated based on the model control instructions, historical forwarding information, the third feature vector and the fourth feature vector.

[0051] like Figure 2 As shown, the social network platform account ID is mapped to the preset dimension of the word embedding that conforms to the large language model through a linear layer to obtain the third feature vector of the social network platform account ID, that is, the social network platform account ID vector representation; the user ID is mapped to the preset dimension of the word embedding that conforms to the large language model through a linear layer to obtain the fourth feature vector of the user ID, that is, the user ID vector representation; then the social network platform account ID vector representation, the user ID vector representation and the other text content in the above <|user|> are concatenated by the vector representation obtained by encoding with the large language model to generate prompt information, which is input into LLMForward.

[0052] The prompt information is input into the large language model for processing, and the output result of the output part mapped from the preset dimension to the low dimension through the linear layer is selected as the intermediate result. Taking the preset dimension as 4096 as an example, the output result of the part of the output part mapped from 4096 dimensions to the low dimension through the linear layer is selected as the intermediate result.

[0053] like Figure 2 As shown, in the process of generating the prompt information, "Answer:" should be added to prompt the large language model to start answering from here.

[0054] Existing methods for estimating forwarding behavior in social networks usually mine user interests through text mining of social media content, ignoring the fact that user-related ID sequence information contains a large amount of collaborative information. The embodiments of the present application overcome the defect that large language models cannot model ID sequence information, and can use ID sequence information as input, thereby effectively enhancing the accuracy of forwarding behavior estimation.

[0055] Step S105, the first feature vector, the second feature vector and the intermediate result are concatenated and input into a multi-layer perceptron, and the multi-layer perceptron estimates the user's forwarding behavior for the candidate social media content to obtain an estimation result.

[0056] like Figure 2 As shown, the model fusion part concatenates the output results of the first estimation part (i.e., the first feature vector of the candidate social media content and the second feature vector of the user) and the output results of the second estimation part and sends them to a multilayer perceptron (MLP). The multilayer perceptron estimates the user's forwarding behavior for the candidate social media content and whether the user will forward the candidate social media content, thereby obtaining an estimation result.

[0057] Multilayer Perceptron is a deep learning model based on Feedforward Neural Network, which consists of multiple neuron layers, each of which is fully connected to the previous layer. Multilayer Perceptron can be used to solve various machine learning problems such as classification, regression and clustering. Each neuron layer of Multilayer Perceptron is composed of many neurons, where the input layer receives input features, the output layer gives the final estimated results, and the hidden layer in the middle is used to extract features and perform nonlinear transformations. Each neuron receives the output of the previous layer, performs weighting, activation function and other operations, and obtains the output of the current layer. Through continuous iterative training, Multilayer Perceptron can automatically learn the complex relationship between input features and predict new data.

[0058] Step S106: Generate explanation information corresponding to the estimation result using the large language model.

[0059] The large language model used in the embodiment of the present application not only learns the knowledge of forwarding behavior prediction, but also has a wealth of other knowledge. The large language model can also be used to generate corresponding explanatory information for the prediction results to explain the prediction results. For example, the generated explanatory information may be: "The user's historical forwarding behavior shows that he prefers science and technology topics. In the provided list of historical forwarded blog posts, there is no evidence of consumption of sports-related blog posts. The candidate blog post is about the ranking of football teams, and this topic does not appear in the user's historical forwarded blog posts. Therefore, based on the user's historical forwarding behavior and the content analysis of the candidate blog post, the user is unlikely to forward it.". The present application introduces a large language model into the estimation process of forwarding behavior on social networks, which not only effectively improves the accuracy of the estimation of user forwarding behavior, but also improves the interpretability of forwarding behavior estimation in social networks.

[0060] According to the social network forwarding behavior prediction method based on large model enhancement provided by the embodiment of the present application, the large language model is applied to the forwarding behavior prediction task, and the user-related ID sequence information is injected into the large language model, and the social media content and the user-related ID sequence information are integrated. The powerful semantic understanding ability and extensive pre-training knowledge of the large language model are utilized, and it is aligned with the prediction task through fine-tuning to help obtain better prediction results; the prediction ability of user forwarding behavior in the social network is enhanced by the large language model, and the estimation accuracy of user forwarding behavior is effectively improved; and the large language model has good interpretability for the prediction results, and the obtained prediction results are returned to the large language model, and explanation information can also be generated, which effectively enhances the interpretability of forwarding behavior prediction in the social network; the solution can be deployed in a computer or server, etc., and can analyze the user forwarding behavior data in the social network platform, improve the user experience, and help study the propagation mechanism in the social network.

[0061] Figure 3 FIG. 1 shows a structural block diagram of a device for estimating social network forwarding behavior based on large model enhancement according to an embodiment of the present application. Figure 3 As shown, the device includes: an acquisition module 310 , a first processing module 320 , a second processing module 330 and an estimation module 340 .

[0062] The acquisition module 310 is adapted to: acquire candidate social media content, historical forwarding information of a user, a social network platform account ID, and a user ID.

[0063] The first processing module 320 is adapted to: extract the vector representation of the text information of the candidate social media content using the social media content body encoder to obtain the first feature vector of the candidate social media content; and extract the vector representation of the historical forwarding information using the user encoder to obtain the second feature vector of the user.

[0064] The second processing module 330 is adapted to generate prompt information according to the model control instruction, historical forwarding information, social network platform account ID and user ID, and input the prompt information into the large language model for processing to obtain an intermediate result.

[0065] The estimation module 340 is adapted to: concatenate the first feature vector, the second feature vector and the intermediate result and input them into a multi-layer perceptron, and the multi-layer perceptron estimates the user's forwarding behavior for the candidate social media content to obtain an estimation result.

[0066] Optionally, the social media content body encoder includes: a text encoder, an element encoder, and an attention-based pooling layer. The first processing module 320 is further adapted to: using the text encoder to learn the body representation from the body text information of the candidate social media content, using the element encoder to learn the body representation from the topic information of the candidate social media content, and capturing the body semantic information of different semantic qualities through the attention-based pooling layer to obtain the first feature vector of the candidate social media content.

[0067] Optionally, the second processing module 330 is further adapted to: map the social network platform account ID and the user ID to a preset dimension of word embedding that conforms to the large language model, to obtain a third feature vector of the social network platform account ID and a fourth feature vector of the user ID; and generate prompt information based on model control instructions, historical forwarding information, the third feature vector and the fourth feature vector.

[0068] Optionally, the second processing module 330 is further adapted to: input the prompt information into the large language model for processing, and select the output result of the output part mapped from the preset dimension to the low dimension through the linear layer as the intermediate result.

[0069] Optionally, the estimation module 340 is further adapted to: generate explanation information corresponding to the estimation result using a large language model.

[0070] Optionally, the large language model is a large language model after parameter fine-tuning using a low-rank adaptive method.

[0071] The description of each module above refers to the corresponding description in the method embodiment and will not be repeated here.

[0072] According to the device for estimating forwarding behavior of a social network based on large model enhancement provided by an embodiment of the present application, a large language model is applied to the forwarding behavior estimation task, and user-related ID sequence information is injected into the large language model, social media content and user-related ID sequence information, etc. are integrated, and the powerful semantic understanding ability and extensive pre-training knowledge of the large language model are utilized to align it with the estimation task through fine-tuning, thereby helping to obtain better estimation results; the estimation ability of user forwarding behavior in a social network is enhanced by the large language model, and the estimation accuracy of user forwarding behavior is effectively improved; and the large language model has good interpretability of the estimation results, and the obtained estimation results are returned to the large language model, and explanation information can also be generated, which effectively enhances the interpretability of forwarding behavior estimation in the social network; the solution can be deployed in a computer or server, etc., and can analyze user forwarding behavior data in a social network platform, thereby improving user experience and facilitating the study of propagation mechanisms in social networks.

[0073] An embodiment of the present application provides a non-volatile computer storage medium, which stores at least one executable instruction or computer program, which enables a processor to perform operations corresponding to the method for estimating social network forwarding behavior based on large model enhancement in any of the above method embodiments.

[0074] An embodiment of the present application provides a computer program product, which includes at least one executable instruction or computer program, which can enable a processor to perform operations corresponding to the social network forwarding behavior prediction method based on large model enhancement in any of the above method embodiments.

[0075] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. The specific embodiment of the present application does not limit the specific implementation of the computing device.

[0076] like Figure 4 As shown, the computing device may include: a processor (processor) 402 , a communications interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0077] Wherein: the processor 402, the communication interface 404, and the memory 406 communicate with each other through the communication bus 408. The communication interface 404 is used to communicate with other devices such as a client or other server network elements. The processor 402 is used to execute the program 410, which can specifically execute the relevant steps in the embodiment of the social network forwarding behavior prediction method based on large model enhancement for the computing device.

[0078] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0079] The processor 402 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0080] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0081] Program 410 can be specifically used to enable processor 402 to execute the social network forwarding behavior prediction method based on large model enhancement in any of the above-mentioned method embodiments. The specific implementation of each step in program 410 can refer to the corresponding descriptions in the corresponding steps and units in the above-mentioned social network forwarding behavior prediction embodiment based on large model enhancement, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-mentioned devices and modules can refer to the corresponding process description in the above-mentioned method embodiment, which will not be repeated here.

[0082] The algorithm and display provided herein are not inherently related to any specific computer, virtual system or other equipment. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application embodiment described herein, and the above description of specific languages ​​is to disclose the best mode of implementation of the present application embodiment.

[0083] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0084] Similarly, it should be understood that in order to streamline the present disclosure and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following intention: the claimed embodiments of the present application require more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the embodiment of the present application.

[0085] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0086] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the embodiments of the present application and to form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0087] The various component embodiments of the embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some or all components according to the embodiments of the present application. The embodiments of the present application can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the embodiments of the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0088] It should be noted that the above embodiments illustrate rather than limit the embodiments of the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The embodiments of the present application may be implemented by means of hardware including several different elements and by means of appropriately programmed computers. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

Claims

1. A method for estimating social network forwarding behavior based on large model enhancement, characterized in that: include: Obtain candidate social media content, user's historical forwarding information, social network platform account ID, and user ID; Extracting a vector representation of text information of the candidate social media content using a social media content body encoder to obtain a first feature vector of the candidate social media content; Extracting the vector representation of the historical forwarding information using a user encoder to obtain a second feature vector of the user; Generate prompt information according to the model control instruction, the historical forwarding information, the social network platform account ID and the user ID, and input the prompt information into the large language model for processing to obtain an intermediate result; The first feature vector, the second feature vector and the intermediate result are concatenated and input into a multilayer perceptron, and the multilayer perceptron estimates the user's forwarding behavior for the candidate social media content to obtain an estimation result; wherein the multilayer perceptron is composed of multiple neuron layers, and each neuron layer is fully connected to the previous layer; The step of generating prompt information based on the model control instruction, the historical forwarding information, the social network platform account ID, and the user ID, and inputting the prompt information into the large language model for processing to obtain the intermediate result further includes: Mapping the social network platform account ID and the user ID to a preset dimension of word embedding that conforms to the large language model to obtain a third feature vector of the social network platform account ID and a fourth feature vector of the user ID; generating the prompt information according to the model control instruction, the historical forwarding information, the third feature vector and the fourth feature vector; The prompt information is input into the large language model for processing, and an output result in the output part that is mapped from a preset dimension to a low dimension through a linear layer is selected as the intermediate result.

2. The method according to claim 1, characterized in that The social media content body encoder includes: a text encoder, an element encoder, and an attention-based pooling layer; The step of extracting the vector representation of the text information of the candidate social media content by using the social media content body encoder to obtain the first feature vector of the candidate social media content further comprises: The text encoder is used to learn the body representation from the body text information of the candidate social media content, and the element encoder is used to learn the body representation from the topic information of the candidate social media content. The body semantic information of different semantic qualities is captured through an attention-based pooling layer to obtain the first feature vector of the candidate social media content.

3. The method according to claim 1, characterized in that After obtaining the estimation result, the method further includes: The large language model is used to generate explanation information corresponding to the estimation result.

4. The method according to any one of claims 1 to 3, characterized in that: The large language model is a large language model after parameter fine-tuning using a low-rank adaptive method.

5. A social network forwarding behavior prediction device based on large model enhancement, characterized in that: include: An acquisition module, adapted to acquire candidate social media content, a user's historical forwarding information, a social network platform account ID, and a user ID; The first processing module is adapted to extract the vector representation of the text information of the candidate social media content using the social media content body encoder to obtain the first feature vector of the candidate social media content; and extract the vector representation of the historical forwarding information using the user encoder to obtain the second feature vector of the user; A second processing module is adapted to generate prompt information according to the model control instruction, the historical forwarding information, the social network platform account ID and the user ID, and input the prompt information into the large language model for processing to obtain an intermediate result; an estimation module, adapted to concatenate the first feature vector, the second feature vector and the intermediate result and input the concatenated result into a multilayer perceptron, wherein the multilayer perceptron estimates the forwarding behavior of the user for the candidate social media content to obtain an estimation result; wherein the multilayer perceptron is composed of a plurality of neuron layers, and each neuron layer is fully connected to the previous layer; Wherein, the second processing module is further adapted to: Mapping the social network platform account ID and the user ID to a preset dimension of word embedding that conforms to the large language model to obtain a third feature vector of the social network platform account ID and a fourth feature vector of the user ID; generating the prompt information according to the model control instruction, the historical forwarding information, the third feature vector and the fourth feature vector; The prompt information is input into the large language model for processing, and an output result in the output part that is mapped from a preset dimension to a low dimension through a linear layer is selected as the intermediate result.

6. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the social network forwarding behavior prediction method based on large model enhancement as described in any one of claims 1-4.

7. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the method for estimating social network forwarding behavior based on large model enhancement as described in any one of claims 1-4.

8. A computer program product, comprising at least one executable instruction, wherein the executable instruction enables a processor to perform operations corresponding to the method for estimating social network forwarding behavior based on large model enhancement as described in any one of claims 1-4.

Images