Social media soft advertisement user responsivity prediction method
By constructing a multimodal content diffusion model, the problem of difficulty in capturing the interactive influence between soft advertising posts on social media is solved, user responsiveness prediction with high accuracy and strong generalization ability is achieved, and the importance of influencing posts is revealed.
Patent Information
- Application Number
- CN202510696227.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2025-09-16
AI Technical Summary
Existing social media user responsiveness prediction methods find it difficult to effectively capture the interactive influence between diverse posts, which limits the accuracy and generalization ability of the prediction model.
A user responsiveness prediction method based on multimodal content diffusion is adopted. By collecting soft advertisement post data from social media platforms, feature extraction and clustering are performed, and a user responsiveness prediction model based on multimodal content diffusion is constructed. The model includes a soft advertisement post activation module, a diffusion module, and a user responsiveness prediction module. Pre-trained text and image embedding models are used for feature extraction and fusion to predict user responsiveness.
The accuracy and generalization ability of predicting user responsiveness to social media soft advertisements are improved, and it reveals which posts contribute more to the overall responsiveness, which has a certain degree of interpretability.
Smart Images
Figure CN120655348A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence marketing, and in particular relates to a method for predicting user responsiveness of social media soft advertisements. Background Art
[0002] Social media soft advertising, where businesses collaborate with bloggers to create and publish social media posts on social media platforms to promote their brands or products, has become a widely adopted and popular advertising and marketing method on social media platforms. Predicting the likely user response to these posts (including likes, favorites, forwarding, and comments) helps businesses predict the success of advertising campaigns and, in turn, refine their soft advertising posting strategies.
[0003] Existing methods for predicting user responsiveness on social media are mostly based on deep learning models. These methods map multimodal social media posts into a vector space and then predict the number of user responses. However, existing methods for predicting user responsiveness mostly focus on static post content features and fail to adequately model the collaborative diffusion of posts. Social media soft advertising typically includes a variety of posts with different styles, promoting a brand or product from different perspectives. These different posts often have complex semantic connections. Existing methods struggle to effectively capture the interactive influences between diverse posts, such as the driving effect of popular content on related content, limiting the accuracy and generalization of prediction models. Summary of the Invention
[0004] In order to solve the problems existing in the background technology, the present invention provides a method for predicting user responsiveness of social media soft advertisements, which solves the technical problem that the existing technology is difficult to effectively capture the interactive influence between diversified posts.
[0005] The technical solution adopted by the user responsiveness prediction method of the present invention includes the following steps:
[0006] S1. Collect the text and image content of all soft advertising posts published in several soft advertising activities on social media platforms, and obtain the user response degree of each soft advertising post. Use the text and image content of each soft advertising post as input and the corresponding user response degree as label to construct a soft advertising dataset.
[0007] S2. Perform feature extraction and clustering processing on the soft advertisement dataset in sequence to obtain the soft advertisement dataset after clustering processing.
[0008] S3. Construct a user responsiveness prediction model based on multimodal content diffusion in a computer, input the clustered soft advertisement data set into the user responsiveness prediction model based on multimodal content diffusion for training, and obtain a trained user responsiveness prediction model based on multimodal content diffusion.
[0009] S4. Input the graphic and text contents of all posts in the soft advertising campaign to be tested into the trained user responsiveness model based on multimodal content diffusion to predict the user responsiveness of the soft advertising campaign to be tested, that is, to obtain the popularity of the soft advertising campaign to be tested.
[0010] The obtained popularity of the soft advertisement activity to be tested is used to evaluate the effectiveness of the implementation of the soft advertisement to be tested.
[0011] The user response degree of each soft advertisement post in step S1 is the sum of the number of likes, comments, favorites and reposts.
[0012] The step S2 is specifically as follows:
[0013] S21. Use the word frequency-inverse document frequency method to extract features of each soft advertisement post in the soft advertisement dataset to obtain the topic tag features of the corresponding soft advertisement post.
[0014] S22. Clustering the soft advertisement dataset using a hierarchical clustering method according to the topic tag features of all soft advertisement posts to obtain a clustered soft advertisement dataset.
[0015] The user responsiveness prediction model based on multimodal content diffusion in step S3 includes a soft advertising post activation module, a soft advertising post diffusion module and a user responsiveness prediction module; the graphic content of each soft advertising post in the soft advertising data set after clustering processing is input into the soft advertising post activation module for processing, and the multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts and the activation vector of each post in each category of soft advertising posts are respectively obtained; the multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts and the activation vector of each post in each category of soft advertising posts are input into the soft advertising post diffusion module for processing, and the weight of each post in each category of soft advertising posts and the updated features of the intra-class representation features of each post in each category of soft advertising posts are respectively obtained; the weight of each post in each category of soft advertising posts and the updated features of the intra-class representation features of each post in each category of soft advertising posts are input into the user responsiveness prediction module for processing to obtain user responsiveness.
[0016] The weight of each post in each category of soft advertisement posts output by the soft advertisement post diffusion module reveals which posts contribute more to the overall responsiveness.
[0017] The soft advertisement post activation module includes a picture-text multimodal representation unit and a picture-text attention fusion unit, and the picture-text multimodal representation unit includes a text representation unit and an image representation unit; the text content of the picture-text content of each soft advertisement post in the soft advertisement data set after clustering processing is input into the text representation unit for processing to obtain text vector features; the image content of the picture-text content of each soft advertisement post in the soft advertisement data set after clustering processing is input into the image representation unit for processing to obtain image vector features; the text vector features and the image vector features are input together into the picture-text attention fusion unit for processing to obtain the multimodal activation value of the first post in each category of soft advertisement posts, the collective influence matrix representation of all posts in each category of soft advertisement posts and the activation vector of each post in each category of soft advertisement posts, and the multimodal activation value of the first post in each category of soft advertisement posts, the collective influence matrix representation of all posts in each category of soft advertisement posts and the activation vector of each post in each category of soft advertisement posts are input together into the soft advertisement post diffusion module for processing.
[0018] The text representation unit adopts the pre-trained text embedding model XLM-Roberta model, and the image representation unit adopts the pre-trained image embedding model Vision Transformer model.
[0019] The image-text attention fusion unit is set according to the following formula:
[0020]
[0021] Among them, k and s both represent indexes; α k,1 represents the multimodal activation value of the first post in the k-th soft advertisement post; f() represents the Sigmoid activation function; Represents the collective influence matrix representation of all posts in the k-th category of soft advertising posts The first vector in ; represents the collective influence matrix representation of all posts in the k-th category of soft advertisement posts; softmax() represents the softmax() function; A matrix representation of the activation vectors of the first post representing all categories; Representation matrix representation Dimensions; represents the activation vector of the first post in the first category of soft advertisement posts; represents the activation vector of the first post in the second category of soft advertisement posts; represents the activation vector of the first post in the k-th category of soft advertisement posts; represents the activation vector of the first post in the Nth type of soft advertisement posts; represents the activation vector of the sth post in the kth category of soft advertisement posts; Represents the text vector features of the sth post in the kth category of soft advertisement posts; Represents the image-text fusion features of the sth post in the kth category of soft advertising posts; Represents the image vector features of the sth post in the kth category of soft advertisement posts; represents transposition; [·||·] represents feature concatenation; Represents image vector features Dimension; W α 、 W Te 、W Im1 and W Im2 Both represent weight matrices; b α and Both indicate bias.
[0022] The soft advertisement post diffusion module is set according to the following formula:
[0023]
[0024] Among them, β k，s represents the weight of the sth post in the kth category of soft advertisement posts; α k，1 represents the multimodal activation value of the first post in the k-th soft advertisement post; i, j, t and l all represent indexes; It represents one of the outputs of the soft advertisement post diffusion module, and also represents the updated feature of the intra-class representation feature of the s-th post in the k-th category of soft advertisement posts; The feature obtained by concatenating the updated features of the in-class representation features of all soft advertisement posts; represents transposition; M represents the number of soft advertising posts in each category; Represents the collective influence matrix representation of all posts in the k-th category of soft advertising posts; represents the intra-class representation feature of the sth post in the kth category of soft advertisement posts; f( ) represents the Sigmoid activation function; softmax( ) represents the softmax( ) function; and They represent the activation vector of the lth post and the activation vector of the sth post in the kth category of soft advertisement posts respectively; δ represents the weight coefficient; [·||·] represents feature concatenation; f( ) represents the Sigmoid activation function; softmax( ) represents the softmax( ) function; e, and Both represent weight matrices; represents the transpose of the weight matrix e.
[0025] The user responsiveness prediction module obtains the user responsiveness by processing according to the following formula:
[0026]
[0027] in, represents user responsiveness; k and s both represent indexes; N represents the number of categories obtained by clustering; M represents the number of soft advertising posts contained in each category; β k，s represents the weight of the sth post in the kth category of soft advertisement posts; W represents the updated features of the intra-class representation features of the s-th post in the k-th soft advertisement post; pre represents the weight matrix; b pre Indicates bias.
[0028] The beneficial effects of the present invention are:
[0029] 1. This paper effectively predicts the overall user responsiveness of social media soft advertisements by modeling the activation and diffusion process between multimodal social media soft advertisement posts. It also reveals which posts contribute more to the overall responsiveness, making the prediction process interpretable.
[0030] 2. The method of the present invention has high accuracy and strong generalization ability in predicting user responsiveness to social media soft advertisements. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 Flowchart of the method of the present invention. DETAILED DESCRIPTION
[0032] The present invention is described in more detail below with reference to the accompanying drawings and examples. However, the present invention is not limited thereto. A person skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are considered to be within the scope of protection of the present invention. Any matters not described in detail in this specification constitute prior art known to those skilled in the art.
[0033] like Figure 1 As shown, the embodiment of the present invention is implemented according to the following steps:
[0034] S1. Collect the text and image content of all soft advertising posts published in several historical soft advertising campaigns on social media platforms, and obtain the user response degree of each soft advertising post. Use the text and image content of each soft advertising post as input and the corresponding user response degree as label to construct a soft advertising dataset.
[0035] In specific implementations, the image and text content of each soft advertising post is used as input, and the corresponding user response level is used as a label to obtain the data of the corresponding soft advertising post. The data of all soft advertising posts is merged to form a soft advertising dataset. In specific implementations, a soft advertising campaign contains several soft advertising posts. Social media platforms are online software platforms that provide social interaction services over the internet; soft advertising campaigns are advertising activities conducted through content marketing. Soft advertising usually appears in the form of articles, stories, news, comments, etc.
[0036] The user response of each soft advertising post is the sum of the number of likes, comments, favorites and reposts.
[0037] S2. Perform feature extraction and clustering processing on the soft advertisement dataset in sequence to obtain the soft advertisement dataset after clustering processing.
[0038] S21. Use the word frequency-inverse document frequency method to extract features of each soft advertisement post in the soft advertisement dataset to obtain the topic tag features of the corresponding soft advertisement post.
[0039] In a specific implementation, the feature extraction unit extracts features from topic tags in soft advertisement posts; topic tags are keywords starting with “#” in soft advertisement posts; and the first soft advertisement post in each category is taken as the category center post.
[0040] S22. Clustering the soft advertisement dataset using a hierarchical clustering method according to the topic tag features of all soft advertisement posts to obtain a clustered soft advertisement dataset.
[0041] In the specific implementation, the soft advertising dataset after clustering processing is specifically: the number of categories is N and the number of soft advertising posts contained in each category is M. Based on the soft advertising dataset after clustering processing, the image and text content of each soft advertising post in each category can be obtained, including the image and text content of the first soft advertising post in each category, and the first soft advertising post is the class center post of the category.
[0042] In a specific implementation, determining the number of categories N finally obtained by the hierarchical clustering algorithm includes calculating the silhouette coefficients under different K values and selecting the N value that maximizes the silhouette coefficient as the optimal number of clusters.
[0043] In the specific implementation, during the clustering process, a maximum number M is preset for each category. When the number of soft advertisement posts in a certain category exceeds M, the posts published later are truncated in order of their publishing time.
[0044] S3. Construct a user responsiveness prediction model based on multimodal content diffusion in a computer, input the clustered soft advertisement data set into the user responsiveness prediction model based on multimodal content diffusion for training, and obtain a trained user responsiveness prediction model based on multimodal content diffusion.
[0045] The user responsiveness prediction model based on multimodal content diffusion in step S3 includes a soft advertising post activation module, a soft advertising post diffusion module and a user responsiveness prediction module; the text content and image content of the graphic content of each soft advertising post in the soft advertising data set after clustering processing are input into the soft advertising post activation module for processing, and the multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts and the activation vector of each post in each category of soft advertising posts are obtained respectively; the multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts and the activation vector of each post in each category of soft advertising posts are input into the soft advertising post diffusion module for processing, and the weight of each post in each category of soft advertising posts and the updated features of the intra-class representation features of each post in each category of soft advertising posts are obtained respectively; the weight of each post in each category of soft advertising posts and the updated features of the intra-class representation features of each post in each category of soft advertising posts are input into the user responsiveness prediction module for processing to obtain user responsiveness.
[0046] The weight of each post in each category of soft advertisement posts output by the soft advertisement post diffusion module reveals which posts contribute more to the overall responsiveness.
[0047] Furthermore, in the subsequent soft advertising campaign post delivery process, content similar to the text and image content of the posts with greater weights is delivered, which has guiding significance for the delivery of posts in the soft advertising campaign.
[0048] The soft advertisement post activation module includes a picture-text multimodal representation unit and a picture-text attention fusion unit. The picture-text multimodal representation unit includes a text representation unit and an image representation unit. The text content of the picture-text content of each soft advertisement post in the soft advertisement dataset after clustering processing is input into the text representation unit for processing to obtain text vector features. The image content of the picture-text content of each soft advertisement post in the soft advertisement dataset after clustering processing is input into the image representation unit for processing to obtain image vector features. The text vector features and the image vector features are input into the picture-text attention fusion unit together for processing to obtain the multimodal activation value of the first post in each category of soft advertisement posts, the collective influence matrix representation of all posts in each category of soft advertisement posts, and the activation vector of each post in each category of soft advertisement posts. The multimodal activation value of the first post in each category of soft advertisement posts, the collective influence matrix representation of all posts in each category of soft advertisement posts, and the activation vector of each post in each category of soft advertisement posts are input into the soft advertisement post diffusion module together for processing.
[0049] The text representation unit uses the pre-trained text embedding model XLM-Roberta model, and the image representation unit uses the pre-trained image embedding model Vision Transformer model.
[0050] The image-text attention fusion unit is set according to the following formula:
[0051]
[0052] Among them, k and s both represent indexes; α k,1 represents one of the outputs of the image-text attention fusion unit, and also represents the multimodal activation value of the first post in the k-th category of soft advertisement posts; f( ) represents the Sigmoid activation function; Represents the collective influence matrix representation of all posts in the k-th category of soft advertising posts The first vector in ; represents one of the outputs of the image-text attention fusion unit, and also represents the collective influence matrix representation of all posts in the k-th category of soft advertisement posts; softmax() represents the softmax() function; A matrix representation of the activation vectors of the first post representing all categories; Representation matrix representation Dimensions; represents the activation vector of the first post in the first category of soft advertisement posts; represents the activation vector of the first post in the second category of soft advertisement posts; represents the activation vector of the first post in the k-th category of soft advertisement posts; Represents the activation vector of the first post in the Nth category (the last category) of soft advertisement posts; represents one of the outputs of the image-text attention fusion unit, and also represents the activation vector of the sth post in the kth category of soft advertisement posts; It represents one of the inputs of the image-text attention fusion unit and also represents the text vector feature of the sth post in the kth category of soft advertisement posts; Represents the image-text fusion features of the sth post in the kth category of soft advertising posts; It represents another input of the image-text attention fusion unit, and also represents the image vector feature of the sth post in the kth category of soft advertisement posts; represents transposition; [·||·] represents feature concatenation; Represents image vector features Dimension; W α 、 W Te 、W Im1 and W Im2 Both represent weight matrices; b α and Both indicate bias.
[0053] The soft advertisement post diffusion module is set according to the following formula:
[0054]
[0055] Among them, β k，s represents one of the outputs of the soft advertisement post diffusion module, and also represents the weight of the sth post in the kth category of soft advertisement posts; α k，1 represents one of the inputs of the soft advertisement post diffusion module, and also represents the multimodal activation value of the first post in the k-th category of soft advertisement posts; i, j, t, and l all represent indexes; It represents one of the outputs of the soft advertisement post diffusion module, and also represents the updated feature of the intra-class representation feature of the s-th post in the k-th category of soft advertisement posts; The updated features representing the intra-class representation of all soft advertising posts (N categories, M in each category) are concatenated according to the number of posts. represents transposition; M represents the number of soft advertising posts in each category; It represents one of the inputs of the soft advertisement post diffusion module and also represents the collective influence matrix representation of all posts in the k-th category of soft advertisement posts; represents the intra-class representation feature of the sth post in the kth category of soft advertisement posts; f( ) represents the Sigmoid activation function; softmax( ) represents the softmax( ) function; softmax l ( ) represents the lth softmax() function; and represents one of the inputs of the soft advertisement post diffusion module, and also represents the activation vector of the lth post and the sth post in the kth category of soft advertisement posts; δ represents the weight coefficient; [·||·] represents feature concatenation; f( ) represents the Sigmoid activation function; softmax( ) represents the softmax( ) function; e, and Both represent weight matrices; represents the transpose of the weight matrix e.
[0056] The user responsiveness prediction module is set according to the following formula:
[0057]
[0058] in, represents the output of the user responsiveness prediction module and also represents the user responsiveness; k and s both represent indexes; N represents the number of categories obtained by clustering; M represents the number of soft advertising posts contained in each category; β k，s It represents one of the inputs of the user responsiveness prediction module and also represents the weight of the sth post in the kth category of soft advertisement posts; represents one of the inputs of the user responsiveness prediction module, and also represents the updated features of the intra-class representation features of the s-th post in the k-th soft advertisement post; W pre represents the weight matrix; b pre Indicates bias.
[0059] S4. Input the graphic and text contents of all posts in the soft advertising campaign to be tested into the trained user responsiveness model based on multimodal content diffusion to predict the user responsiveness of the soft advertising campaign to be tested, that is, to obtain the popularity of the soft advertising campaign to be tested.
[0060] The obtained popularity of the soft advertisement activity to be tested is used to evaluate the effectiveness of the implementation of the soft advertisement to be tested.
[0061] This paper effectively predicts the overall user responsiveness of social media soft advertisements by modeling the activation and diffusion process between multimodal social media soft advertisement posts. It also reveals which posts contribute more to the overall responsiveness, making the prediction process interpretable to a certain extent.
[0062] The above embodiments are merely preferred embodiments for the purpose of fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are within the scope of protection of the present invention. The scope of protection of the present invention shall be subject to the claims.
Claims
1. A method for predicting user responsiveness of social media soft advertisements, characterized in that: The steps include: S1. Collect the text and image content of all soft advertising posts published in several soft advertising campaigns on social media platforms, and obtain the user response degree of each soft advertising post. Use the text and image content of each soft advertising post as input and the corresponding user response degree as label to construct a soft advertising dataset. S2. performing feature extraction and clustering processing on the soft advertisement dataset in sequence to obtain a soft advertisement dataset after clustering processing; S3. Construct a user responsiveness prediction model based on multimodal content diffusion, input the clustered soft advertisement dataset into the user responsiveness prediction model based on multimodal content diffusion for training, and obtain a trained user responsiveness prediction model based on multimodal content diffusion; S4. Input the graphic and text contents of all posts in the soft advertising campaign to be tested into the trained user responsiveness model based on multimodal content diffusion to predict the user responsiveness of the soft advertising campaign to be tested, that is, to obtain the popularity of the soft advertising campaign to be tested.
2. The method for predicting user responsiveness of social media soft advertisements according to claim 1, characterized in that: The user response degree of each soft advertisement post in step S1 is the sum of the number of likes, comments, favorites and reposts.
3. The method for predicting user responsiveness of social media soft advertisements according to claim 1, characterized in that: The step S2 is specifically as follows: S21, using the word frequency-inverse document frequency method to extract features of each soft advertisement post in the soft advertisement dataset to obtain the topic tag features of the corresponding soft advertisement post; S22. Clustering the soft advertisement dataset using a hierarchical clustering method according to the topic tag features of all soft advertisement posts to obtain a clustered soft advertisement dataset.
4. The method for predicting user responsiveness of social media soft advertisements according to claim 1, characterized in that: The user responsiveness prediction model based on multimodal content diffusion in step S3 includes a soft advertisement post activation module, a soft advertisement post diffusion module and a user responsiveness prediction module; The text and image content of each soft advertisement post in the clustered soft advertisement dataset is input into the soft advertisement post activation module for processing. The multimodal activation value of the first post in each category of soft advertisement posts, the collective influence matrix representation of all posts in each category of soft advertisement posts, and the activation vector of each post in each category of soft advertisement posts are obtained respectively. The multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts, and the activation vector of each post in each category of soft advertising posts are input into the soft advertising post diffusion module for processing, and the weight of each post in each category of soft advertising posts and the updated features of the intra-class representation features of each post in each category of soft advertising posts are obtained respectively; The weight of each post in each category of soft advertisement posts and the updated features of the intra-category representation features of each post in each category of soft advertisement posts are input into the user responsiveness prediction module for processing to obtain the user responsiveness.
5. The method for predicting user responsiveness of social media soft advertisements according to claim 4, characterized in that: The soft advertisement post activation module includes a picture-text multimodal representation unit and a picture-text attention fusion unit, wherein the picture-text multimodal representation unit includes a text representation unit and an image representation unit; the text content of the picture-text content of each soft advertisement post in the soft advertisement dataset after clustering processing is input into the text representation unit for processing to obtain text vector features; The image content of the text and image content of each soft advertisement post in the soft advertisement dataset after clustering is input into the image representation unit for processing to obtain image vector features; The text vector features and image vector features are input together into the image-text attention fusion unit for processing to obtain the multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts, and the activation vector of each post in each category of soft advertising posts. The multimodal activation value of the first post in each category of soft advertising posts, the collective influence matrix representation of all posts in each category of soft advertising posts, and the activation vector of each post in each category of soft advertising posts are input together into the soft advertising post diffusion module for processing.
6. The method for predicting user responsiveness of social media soft advertisements according to claim 5, characterized in that: The text representation unit adopts the pre-trained text embedding model XLM-Roberta model, and the image representation unit adopts the pre-trained image embedding model Vision Transformer model.
7. The method for predicting user responsiveness of social media soft advertisements according to claim 5, characterized in that: The image-text attention fusion unit is set according to the following formula: Among them, k and s both represent indexes; α k,1 represents the multimodal activation value of the first post in the k-th soft advertisement post; f() represents the Sigmoid activation function; Represents the collective influence matrix representation of all posts in the k-th category of soft advertising posts The first vector in ; represents the collective influence matrix representation of all posts in the k-th category of soft advertisement posts; softmax() represents the softmax() function; A matrix representation of the activation vectors of the first post representing all categories; Representation matrix representation Dimensions; represents the activation vector of the first post in the first category of soft advertisement posts; represents the activation vector of the first post in the second category of soft advertisement posts; represents the activation vector of the first post in the k-th category of soft advertisement posts; represents the activation vector of the first post in the Nth type of soft advertisement posts; represents the activation vector of the sth post in the kth category of soft advertisement posts; Represents the text vector features of the sth post in the kth category of soft advertisement posts; Represents the image-text fusion features of the sth post in the kth category of soft advertising posts; Represents the image vector features of the sth post in the kth category of soft advertising posts; () T represents transposition; [·||·] represents feature concatenation; Represents image vector features Dimension; W α 、 w Te 、W Im and W Im Both represent weight matrices; b α and Both indicate bias.
8. The method for predicting user responsiveness of social media soft advertisements according to claim 5, characterized in that: The soft advertisement post diffusion module is set according to the following formula: Among them, β k,s represents the weight of the sth post in the kth category of soft advertisement posts; α k,1 represents the multimodal activation value of the first post in the k-th soft advertisement post; i, j, t and l all represent indexes; Represents the updated features of the intra-class representation features of the s-th post in the k-th category of soft advertising posts; The feature obtained by concatenating the updated features of the in-class representation features of all soft advertisement posts; () T represents transposition; M represents the number of soft advertising posts in each category; Represents the collective influence matrix representation of all posts in the k-th category of soft advertising posts; represents the intra-class representation feature of the sth post in the kth category of soft advertisement posts; f() represents the Sigmoid activation function; softmax() represents the softmax() function; and They represent the activation vector of the lth post and the activation vector of the sth post in the kth category of soft advertisement posts respectively; δ represents the weight coefficient; [·||·] represents feature splicing; f() represents the Sigmoid activation function; softmax() represents the softmax() function; e, and Both represent weight matrices; e T represents the transpose of the weight matrix e.
9. The method for predicting user responsiveness of social media soft advertisements according to claim 5, characterized in that: The user responsiveness prediction module obtains the user responsiveness by processing according to the following formula: in, represents user responsiveness; k and s both represent indexes; N represents the number of categories obtained by clustering; M represents the number of soft advertising posts contained in each category; β k,s represents the weight of the sth post in the kth category of soft advertisement posts; W represents the updated features of the intra-class representation features of the s-th post in the k-th soft advertisement post; pre represents the weight matrix; b pre Indicates bias.