A data delivery optimization method and system based on multi-modal data

Abstract

The application discloses a data putting optimization method based on multi-modal data, comprising the following steps: obtaining user identity information and associated multi-modal data, and inputting the multi-modal data to a portrait establishment model; the portrait establishment model adjusts the analysis weight of the multi-modal data based on platform traffic type information, establishes user portrait information by analyzing the multi-modal information; a basic putting strategy is generated based on the user portrait information; a propagation relationship chain including a sharer user and an accepter user is determined based on the user identity information; historical sharing information and social network structure information in the propagation relationship chain are obtained, and a social influence model is constructed; and the basic putting strategy is optimized based on the bidirectional association features output by the social influence model to generate an optimized putting strategy. The application adjusts the analysis weight of the multi-modal data and optimizes the basic putting strategy based on bidirectional association features, thereby solving the problem of low precision and conversion rate in the traditional data putting method.

Description

Technical Field

[0001] This invention relates to the field of data delivery technology, specifically to a data delivery optimization method and system based on multimodal data. Background Technology

[0002] Data-driven advertising refers to the process of showcasing advertising content to potential consumers through various channels and platforms to achieve business goals such as brand promotion, product marketing, and sales growth. In current data-driven advertising practices, to improve targeting accuracy, user profiles (i.e., multi-dimensional information models) are typically created, and the most compelling data is selected based on these profiles for targeted advertising, thereby enhancing advertising effectiveness and conversion rates.

[0003] However, existing data delivery methods still have the following shortcomings:

[0004] 1. Traditional data delivery methods rely on collecting multimodal data (such as text and video) including user behavior and social attribute data. This data is then analyzed and processed with equal weights, and a multidimensional information model of the user is constructed based on the analysis results. However, the core content formats of different platforms may vary significantly. For example, 70% of a short video platform user's interest characteristics may be contained within the video frames, but existing data analysis methods typically use an average distribution approach, leading to an unreasonable allocation of analytical resources and thus affecting the accuracy of user profile information.

[0005] 2. Current data delivery methods generally employ a one-way propagation model (i.e., from sharer to receiver) when processing social network data among users. For example, they adjust the receiver's profile solely based on the information shared by the sharer, ignoring the two-way transmission effect between users in social networks.

[0006] Therefore, there is an urgent need for a data delivery optimization method and system that can adaptively adjust the weight of multimodal data processing according to the mainstream data types of the platform and activate the value of social networks among users. Summary of the Invention

[0007] To optimize the uneven distribution of data analysis capacity in traditional data delivery and activate the two-way transmission effect between users, thereby improving the accuracy of data delivery, this invention provides a data delivery optimization method and system based on multimodal data.

[0008] The above-mentioned objective of this application is achieved through the following technical solution:

[0009] A data delivery optimization method based on multimodal data includes the following steps:

[0010] Obtain user identity information and related multimodal data as input to the pre-trained profile building model. The multimodal data includes historical user behavior information, corresponding content feature information, and platform traffic type information.

[0011] The user profile model adjusts the analysis weights of multimodal data based on platform traffic type information, and establishes user profile information by analyzing the historical user behavior information and the content feature information.

[0012] When user profile information is received from the user profile building model, a basic targeting strategy is generated based on the user profile information.

[0013] The propagation relationship chain is determined based on user identity information, and the propagation relationship chain includes sharing users and receiving users;

[0014] Obtain historical sharing information and social network structure information between sharer and recipient users in the propagation relationship chain, and construct a social influence model;

[0015] The basic targeting strategy is optimized based on the bidirectional correlation features output by the social influence model, resulting in an optimized targeting strategy.

[0016] By adopting the above technical solution, user identity information and its associated multimodal data are obtained and input into a user profile building model. The user profile building model adjusts the analysis weights of the multimodal data based on platform browsing type information in the multimodal data, and builds user profile information by analyzing historical user behavior information and content feature information in the multimodal data. Based on the user profile information, a basic targeting strategy is generated. Based on user identity information, a propagation relationship chain including sharing users and receiving users is determined. Historical sharing information and social network structure information between sharing users and receiving users in the propagation relationship chain are obtained to build a social influence model. Finally, the basic targeting strategy is optimized based on the bidirectional correlation features output by the social influence model to generate an optimized targeting strategy. This application optimizes the allocation of analysis capacity for multimodal data by adjusting the analysis weights of multimodal data based on platform traffic type information, and optimizes the basic targeting strategy by building a social influence model through the establishment of a propagation relationship chain, thereby activating the value of social networks among users and improving the accuracy and conversion rate of data targeting among users.

[0017] In a preferred embodiment, this application can be further configured as follows: the user profile building model includes a data classification layer, a weight adjustment layer, and a feature fusion layer. The user profile building model adjusts the analysis weights of multimodal data based on platform traffic type information. The steps for building user profile information by analyzing the historical user behavior information and the content feature information include the following steps:

[0018] The data classification layer performs data modality classification on historical user behavior information and its corresponding content feature information to obtain feature tensors of different modalities;

[0019] The weight adjustment layer analyzes the weight information of feature tensors of different modalities based on platform traffic type information and then adjusts the weights accordingly.

[0020] The feature fusion layer performs feature extraction and feature fusion on historical user behavior information and content feature information based on the weight adjustment results, and outputs user profile information.

[0021] By adopting the above technical solution, data modality classification is performed on historical user behavior information and corresponding content feature information. Based on platform traffic type information, the feature tensor is weighted and adjusted. Based on the weight adjustment results, historical user behavior information and corresponding content feature information are processed, thereby obtaining user profiles with higher accuracy more efficiently.

[0022] In a preferred embodiment, this application can be further configured as follows: the step of obtaining historical sharing information and social network structure information between sharer users and recipient users in the propagation relationship chain, and constructing a social influence model, includes the following steps:

[0023] Obtain historical sharing information between sharer users and recipient users, and generate low-dimensional vectors of user nodes with each user as a node, thereby generating social network structure information;

[0024] A social influence model is constructed based on historical sharing information and social network structure information.

[0025] By adopting the above technical solution, historical sharing information of sharing users and receiving users is obtained, corresponding low-dimensional vectors are generated with users as nodes, and social network structure information is generated based on this. Thus, a social influence model is constructed based on historical sharing information and social network structure information, and the bidirectional transmission effect of users in the social network is realized by adopting a bidirectional propagation model between users.

[0026] In a preferred example, this application may be further configured to include the following steps before the step of optimizing the basic delivery strategy based on the bidirectional correlation features output by the social influence model to generate an optimized delivery strategy:

[0027] The system obtains users' historical consumption information as input to a pre-trained consumption prediction model, and matches it with the generated predicted consumption information to obtain the corresponding predicted consumption level information.

[0028] By adopting the above technical solution, users' historical consumption information is obtained and input into the consumption prediction model. The generated predicted consumption information is then matched to obtain the corresponding predicted consumption level information, which facilitates the subsequent optimization of content delivery based on the predicted consumption level information.

[0029] In a preferred embodiment, this application can be further configured as follows: the step of obtaining the user's historical consumption information as input to a pre-trained consumption prediction model, and matching it with the generated predicted consumption information to obtain the corresponding predicted consumption level information, includes the following steps:

[0030] Obtain users' historical consumption information and extract their spending power and consumption preference characteristics;

[0031] The consumption capacity characteristics and consumption preference characteristics are input into the pre-trained consumption prediction model to generate predicted consumption information and predicted consumption preference information.

[0032] Based on the preset hierarchical matching rules, the predicted consumption information is compared and the corresponding predicted consumption hierarchical information is obtained.

[0033] By adopting the above technical solution, users' historical consumption information is obtained for feature extraction. The extracted consumption capacity and consumption preference features are then input into the consumption prediction model to generate predicted consumption information and predicted consumption preference information. Based on hierarchical matching rules, the predicted consumption information is compared to obtain predicted consumption hierarchy information. This enables the prediction of users' consumption capacity and consumption preferences based on historical consumption information, improving the accuracy of data delivery and indirectly increasing the conversion rate.

[0034] In a preferred example, this application can be further configured as follows: the step of optimizing the basic delivery strategy based on the bidirectional correlation features output by the social influence model to generate an optimized delivery strategy includes the following steps:

[0035] The social influence model uses user identity information to perform social graph analysis, determine the interaction frequency of the propagation relationship chain, and mine the second-degree propagation relationship chain to generate bidirectional correlation features.

[0036] Based on the user's predicted consumption level information, the basic targeting strategy is initially optimized to generate an initial targeting strategy.

[0037] The initial delivery strategy is optimized a second time based on bidirectional correlation features to generate an optimized delivery strategy.

[0038] By adopting the above technical solution, the social influence model performs social graph analysis on the communication relationship chain, determines its interaction frequency and mines the second-degree communication relationship chain, thereby generating corresponding bidirectional correlation features. Based on the user's corresponding predicted consumption level information, the basic delivery strategy is optimized to obtain the preliminary delivery strategy. Then, based on the bidirectional correlation features, the preliminary delivery strategy is further optimized to generate the optimized delivery strategy. This realizes the optimization of the delivery content of the basic delivery strategy based on the predicted consumption level information to obtain the preliminary delivery strategy. Then, based on the bidirectional correlation features, the delivery method of the preliminary delivery strategy is optimized to obtain the final optimized delivery strategy.

[0039] In a preferred embodiment, this application can be further configured as follows: the social influence model performs social graph analysis based on user identity information, determines the interaction frequency of the propagation relationship chain, and mines the second-degree propagation relationship chain to generate bidirectional association features, including the following steps:

[0040] Obtain historical interaction information between sharing users and receiving users, determine the frequency of interaction, and generate bidirectional correlation features corresponding to high-frequency interaction relationships and bidirectional correlation features corresponding to low-frequency interaction relationships respectively.

[0041] Mine second-degree propagation relationship chains formed based on indirect interactions and generate bidirectional association features corresponding to indirect interaction relationships.

[0042] By adopting the above technical solution, the historical interaction information between users in the propagation relationship chain is analyzed to determine the interaction frequency, and corresponding bidirectional correlation features are generated and second-degree propagation relationship chains are mined, thereby generating bidirectional correlation features corresponding to indirect interaction relationships. This enables frequency judgment and analysis of the propagation relationship chain, and thus differentiated processing.

[0043] In a preferred embodiment, this application can be further configured as follows: the basic delivery strategy includes delivery content and discount strategies; the step of initially optimizing the basic delivery strategy based on the user's corresponding predicted consumption level information to generate an initial delivery strategy includes the following steps:

[0044] Optimize the content of the basic advertising strategy based on predicted consumer tier information;

[0045] The discount strategy of the basic delivery strategy is optimized based on the predicted consumer preference information;

[0046] The optimized content and promotional strategies generate an initial campaign strategy.

[0047] By adopting the above technical solution, the content of the basic advertising strategy is optimized based on predicted consumption level information, and the discount strategy of the basic advertising strategy is optimized based on predicted consumption preference information. The optimized content and discount strategy are combined to generate a preliminary advertising strategy. This achieves matching the corresponding products as advertising content based on predicted consumption level information and generating corresponding discount strategies based on predicted consumption preference information. In this way, the preliminary advertising strategy is generated by combining the advertising content and discount strategy, ensuring that the advertising content and discount strategy are within the acceptable range of users and improving the actual conversion rate of data advertising.

[0048] In a preferred embodiment, this application can be further configured as follows: the step of performing secondary optimization on the initial delivery strategy based on bidirectional correlation features to generate an optimized delivery strategy includes the following steps:

[0049] Grouping is based on the interaction frequency corresponding to bidirectional correlation features, and the groups include priority push group and interaction incentive group;

[0050] The propagation relationship chains with high-frequency interaction are grouped into priority push groups, and priority push optimization is carried out based on the corresponding initial delivery strategy to generate corresponding optimized delivery strategies.

[0051] The propagation relationship chains with low-frequency interaction and indirect interaction are grouped into interaction incentive groups. Incentive optimization is performed based on the corresponding initial deployment strategy to generate corresponding optimized deployment strategies.

[0052] By adopting the above technical solution, propagation relationship chains with high-frequency interaction are grouped based on the interaction frequency corresponding to bidirectional correlation characteristics. These chains are categorized into a priority push group, and priority push optimization is performed based on the corresponding initial delivery strategy to generate an optimized delivery strategy. Similarly, propagation relationship chains with low-frequency or indirect interaction are categorized into an interaction incentive group, and incentive optimization is performed based on the corresponding initial delivery strategy to generate an optimized delivery strategy. By grouping propagation relationship chains with different interaction frequencies and applying different optimization schemes to the initial delivery strategies corresponding to different groups, propagation relationship chains with high-frequency interaction continue to maintain high-frequency interaction, while interaction incentives are applied to the remaining propagation relationship chains, thereby activating the bidirectional transmission effect between users.

[0053] The second objective of this invention is achieved through the following technical solution:

[0054] A data delivery optimization system based on multimodal data includes:

[0055] The data information acquisition module is used to acquire user identity information and multimodal data associated with the user identity information as input for the pre-trained profile building model. The multimodal data includes historical user behavior information, corresponding content feature information, and platform traffic type information.

[0056] The user profile building module is used to adjust the analysis weight of multimodal data based on platform traffic type information, and to build user profile information by analyzing the historical user behavior information and the content feature information.

[0057] The strategy generation module is used to generate a basic delivery strategy based on the user profile information output by the profile building model.

[0058] The social influence module is used to determine the propagation relationship chain based on user identity information, and the propagation relationship chain includes sharing users and receiving users;

[0059] The social analytics module is used to acquire historical sharing information and social network structure information between sharer and receiver users in the propagation relationship chain, and to build a social influence model.

[0060] The strategy optimization module is used to optimize the basic delivery strategy based on the bidirectional correlation features output by the social influence model, and generate an optimized delivery strategy.

[0061] By adopting the above technical solution, the data information acquisition module is used to acquire user identity information and multimodal data associated with the user identity information as input to the pre-trained profile building model. The multimodal data includes historical user behavior information, corresponding content feature information, and platform traffic type information. The profile building module is used to adjust the analysis weights of the multimodal data based on the platform traffic type information and to build user profile information by analyzing the historical user behavior information and the content feature information. The strategy generation module is used to generate a basic delivery strategy based on the user profile information output by the profile building model when it receives the user profile information. The social influence module is used to determine the propagation relationship chain based on the user identity information. The propagation relationship chain includes sharer users and recipient users. The social analysis module is used to acquire historical sharing information and social network structure information between sharer users and recipient users in the propagation relationship chain and to construct a social influence model. The strategy optimization module is used to optimize the basic delivery strategy based on the bidirectional correlation features output by the social influence model and generate an optimized delivery strategy.

[0062] The beneficial effects of the data delivery optimization method and system based on multimodal data of the present invention are as follows:

[0063] 1. By analyzing platform traffic type information, the profiling model adjusts the analysis weights of multimodal data based on the analysis results, and then uses the profiling model with adjusted analysis weights to analyze the multimodal data, thereby creating user profile information. This method achieves the adjustment of multimodal data analysis weights according to the mainstream types of the platform, solving the problem of unreasonable capacity allocation caused by the average allocation method of traditional data delivery, which leads to low accuracy of user profiles and low computing efficiency.

[0064] 2. By determining the propagation relationship chain based on user identity information and outputting two-way correlation features based on the social influence model constructed based on the propagation relationship chain, the basic delivery strategy is optimized based on the two-way correlation features. This achieves the effect of optimizing the basic delivery strategy based on the two-way transmission relationship between users, which solves the problem of low accuracy of user profiles caused by the traditional data delivery method using a one-way propagation model to process social network data between users. It also has the advantage of improving the conversion rate of data delivery. Attached Figure Description

[0065] Figure 1 This is a flowchart of an embodiment of a data delivery optimization method based on multimodal data according to this application;

[0066] Figure 2 This is a flowchart illustrating step S20 in an embodiment of a data delivery optimization method based on multimodal data according to this application.

[0067] Figure 3 This is a flowchart illustrating step S01 in an embodiment of a data delivery optimization method based on multimodal data according to this application.

[0068] Figure 4 This is a flowchart illustrating step S60 in an embodiment of a data delivery optimization method based on multimodal data according to this application.

[0069] Figure 5 This is a flowchart illustrating step S602 in an embodiment of a data delivery optimization method based on multimodal data according to this application.

[0070] Figure 6 This is a flowchart illustrating step S603 in an embodiment of a data delivery optimization method based on multimodal data according to this application. Detailed Implementation

[0071] The following is in conjunction with the appendix Figure 1-6 This application will be described in further detail.

[0072] In one embodiment, such as Figure 1 As shown, this application discloses a data delivery optimization method based on multimodal data, which specifically includes the following steps:

[0073] S10: Obtain user identity information and multimodal data associated with the user identity information as input for the pre-trained profile building model. The multimodal data includes historical user behavior information, corresponding content feature information, and platform traffic type information.

[0074] In this embodiment, user identity information is the basic data that uniquely identifies a user within the platform; multimodal data is composite information that integrates various data formats such as text, images, audio, and video; the profile building model is generated by integrating user identity information and multimodal behavioral data to create a model that includes dimensions such as interests and preferences; historical user behavior information is the user's interaction trajectory on the platform, including clicks, browsing, dwell time, purchase records, etc.; content feature information is information that represents the correlation between the content corresponding to historical user behavior information and user interests; and platform traffic type information is the traffic proportion information of each modality of information on the platform, such as text, images, and videos.

[0075] Specifically, user identity information is obtained, and their corresponding historical user behavior information, content feature information corresponding to historical user behavior information, and traffic proportion information of each modal information type on the platform are captured and input into the profile to build the model.

[0076] S20: The user profile model is established by adjusting the analysis weight of multimodal data based on platform traffic type information, and by analyzing the historical user behavior information and the content feature information, user profile information is established;

[0077] In this embodiment, the analysis weights are different importance coefficients assigned to data such as text, images, and behaviors, and the user profile information includes information such as the user's basic attributes, interest tags, and behavioral characteristics.

[0078] Specifically, the analysis weight of multimodal data in the profile building model is adjusted based on the platform's traffic type information. For example, for platforms where short videos are the mainstream, the analysis weight of short video data in the corresponding multimodal data is increased. The profile building model then analyzes historical user behavior information and content feature information based on the adjusted analysis weight to build user profile information.

[0079] S30: When receiving user profile information output from the profile building model, generate a basic delivery strategy based on the user profile information;

[0080] In this embodiment, the basic delivery strategy is the initial execution plan for data delivery.

[0081] Specifically, effective tags are extracted from user profile information, and content that matches the effective tags is selected to generate corresponding basic delivery strategies.

[0082] S40: Determine the propagation relationship chain based on user identity information, wherein the propagation relationship chain includes sharing user and receiving user;

[0083] In this embodiment, the propagation relationship chain is the path structure of information flow in the user network. It is a chain or network relationship formed by sharing users (information sending nodes) and receiving users (information receiving nodes) through social interaction. The sharing user is the starting node of information propagation, which transmits information to other users through active behaviors such as forwarding. The receiving user is the ending or intermediate node of information propagation, which participates in propagation through passive reception such as browsing and liking, or active interaction such as commenting and secondary forwarding.

[0084] Specifically, node identification is performed based on user identity information, including sharer users and recipient users, and the topology of the relationship chain corresponding to the node is expanded to establish the corresponding propagation relationship chain.

[0085] S50: Obtain historical sharing information and social network structure information between sharer users and recipient users in the propagation relationship chain, and construct a social influence model;

[0086] In this embodiment, historical sharing information is a record of a user's past dissemination behavior, including data such as sharing time and content type. Social network structure information is a topological feature describing the user's social relationships, including indicators such as node degree and community clustering. The social influence model is a model that quantifies a user's dissemination ability and predicts the diffusion effect of information in the social network through node influence weights.

[0087] Specifically, data such as the sharing time and content type between the sharer and the recipient are obtained, as well as social network structure information describing the users' social relationships, and a social influence model is constructed based on this.

[0088] S60: Optimize the basic targeting strategy based on the bidirectional correlation features output by the social influence model to generate an optimized targeting strategy.

[0089] In this embodiment, the bidirectional correlation feature is a dynamic relationship indicator that reflects the mutual influence between users. It is divided into positive correlation features, which are the dissemination effect of the sharer on the recipient, such as the click rate of the recipient after forwarding, and negative correlation features, which are the feedback influence of the recipient on the sharer, such as the liking behavior increasing the sharer's subsequent activity. The optimized delivery strategy is an upgraded solution that is optimized by superimposing data such as social influence analysis results on the basic strategy.

[0090] Specifically, the bidirectional correlation features output by the social influence model are obtained, and optimization steps are performed on the basic delivery strategy to optimize priority delivery or incentives, thereby generating the final optimized delivery strategy.

[0091] In one embodiment, the profile building model includes a data classification layer, a weight adjustment layer, and a feature fusion layer, such as... Figure 2 As shown, step S20 includes the following steps:

[0092] S201: The data classification layer performs data modality classification on historical user behavior information and its corresponding content feature information to obtain feature tensors of different modalities;

[0093] S202: The weight adjustment layer analyzes the weight information of the feature tensors of different modalities based on the platform traffic type information and adjusts the weights accordingly;

[0094] S203: The feature fusion layer performs feature extraction and feature fusion on historical user behavior information and the content feature information based on the weight adjustment results, and outputs user profile information.

[0095] In this embodiment, the data classification layer is a technical module that divides multimodal data into different modalities according to their presentation forms, such as text, images, and user behavior sequences; the weight adjustment layer is a technical module that dynamically adjusts the importance weights of different modal features; and the feature fusion layer is the core module that integrates multimodal features and generates a unified user profile.

[0096] Specifically, the data classification layer of the profile building model performs data modality classification on multimodal data to obtain feature tensors of different modalities. The weight adjustment layer performs weight analysis and weight adjustment on the feature tensors of different modalities based on platform traffic type information. The feature fusion layer extracts and fuses features based on the weight adjustment results of historical user behavior information and content feature information, thereby outputting user profile information.

[0097] In one embodiment, step S50 includes the following steps:

[0098] S501: Obtain historical sharing information between the sharing user and the receiving user, and generate a low-dimensional vector of user nodes with each user as a node, thereby generating social network structure information;

[0099] S502: Construct a social influence model based on historical sharing information and social network structure information.

[0100] In this embodiment, the low-dimensional vector is a dense vector of fixed dimensions obtained by compressing high-dimensional user behavior data, such as historical sharing records and social relationships, through embedding technology.

[0101] Specifically, the system obtains historical sharing information between sharer and recipient users, extracts user interaction behavior from the historical sharing information, and compresses high-dimensional user behavior data into low-dimensional vectors with each user as a node, thereby generating social network structure information. The system then integrates the low-dimensional user vectors with structural features to construct a social influence model.

[0102] In one embodiment, prior to step S60, the following step is also included:

[0103] S01: Obtain the user's historical consumption information as input to the pre-trained consumption prediction model, and match it with the generated predicted consumption information to obtain the corresponding predicted consumption level information.

[0104] In this embodiment, historical consumption information refers to the user's past consumption behavior data, including purchase time, product category, consumption amount, etc. The consumption prediction model is a model used to predict future consumption trends or amounts. The predicted consumption information is the quantitative result output by the model, including the future consumption amount range, consumption frequency, etc. The predicted consumption level information is the consumption level divided according to the prediction results, including mass consumer groups, middle class, high net worth, etc.

[0105] Specifically, historical consumption information corresponding to users is extracted from transaction records, the preprocessed data is input into a pre-trained consumption prediction model to generate predicted consumption information for future periods, and hierarchical matching is performed to obtain the corresponding predicted consumption level information.

[0106] In one embodiment, such as Figure 3 As shown, step S01 includes the following steps:

[0107] S011: Obtain users' historical consumption information and extract their spending power and consumption preference characteristics;

[0108] S012: Input the consumption capacity characteristics and consumption preference characteristics into the pre-trained consumption prediction model to generate predicted consumption information and predicted consumption preference information;

[0109] S013: Based on the preset hierarchical matching rules, compare and predict consumption information to obtain the corresponding predicted consumption hierarchical information.

[0110] In this embodiment, the consumption capacity feature is an indicator that quantifies the user's economic strength, such as average monthly consumption amount and single transaction limit. The consumption preference feature is the user's preference for promotional schemes such as consumption discounts and offers. The predicted consumption preference information is information on promotional schemes that the user may be willing to accept in the future, such as promotional schemes such as discounts for reaching a certain amount or half price for the second item.

[0111] Specifically, the system acquires users' historical consumption information, extracts consumption capacity and preference characteristics from it, and inputs them into a consumption prediction model to generate predicted consumption information, such as the remaining disposable consumption amount or the remaining monthly consumption amount within a specific time period, as well as predicted consumption preference information, such as the user's preference for purchasing goods to reach a certain spending threshold for discounts or to stock up on large quantities of goods with buy-one-get-one-free offers. Then, based on hierarchical matching rules, the predicted consumption information is compared to obtain the corresponding predicted consumption level information.

[0112] In one embodiment, such as Figure 4 As shown, step S60 includes the following steps:

[0113] S601: The social influence model performs social graph analysis based on user identity information, determines the interaction frequency of the propagation relationship chain, mines the second-degree propagation relationship chain, and generates bidirectional correlation features.

[0114] S602: Based on the user's predicted consumption level information, perform preliminary optimization of the basic delivery strategy to generate a preliminary delivery strategy;

[0115] S603: Based on bidirectional correlation features, perform secondary optimization on the initial delivery strategy to generate an optimized delivery strategy.

[0116] In this embodiment, the social graph is a networked data structure that describes the social relationships between users, the interaction frequency is the number of social interactions between users within a specific time period, and the second-degree propagation relationship chain is the path through which information spreads through two or more jumps.

[0117] Specifically, the social influence model uses users as nodes and interactive behaviors as edges to construct a directed graph, conducts social graph analysis, determines the interaction frequency of the propagation relationship chain, and mines the second-degree propagation relationship chain to generate bidirectional correlation features. Based on the predicted consumption level information, the basic targeting strategy is optimized to obtain the initial targeting strategy, and then the initial targeting strategy is further optimized based on the bidirectional correlation features to obtain the optimized targeting strategy.

[0118] In one embodiment, step S601 includes the following steps:

[0119] S6011: Obtain historical interaction information between the sharing user and the receiving user, determine the frequency of interaction, and generate bidirectional association features corresponding to high-frequency interaction relationships and bidirectional association features corresponding to low-frequency interaction relationships respectively.

[0120] S6012: Mine the second-degree propagation relationship chain formed by indirect interaction and generate bidirectional association features corresponding to the indirect interaction relationship.

[0121] In this embodiment, the interaction frequency is determined by counting the number of interactions between users within a specific time period, and then classifying them into high-frequency or low-frequency interaction relationships based on a threshold.

[0122] Specifically, historical interaction information of the propagation relationship chain is obtained to determine the frequency of interaction, and bidirectional association features corresponding to high-frequency interaction relationships and low-frequency interaction relationships are generated respectively. Secondary propagation relationship chains formed based on indirect interaction are mined, and bidirectional association features corresponding to indirect interaction relationships are generated.

[0123] In one embodiment, such as Figure 5 As shown, step S602 includes the following steps:

[0124] S6021: Optimize the content delivered by the basic delivery strategy based on predicted consumption level information;

[0125] S6022: Optimize the discount strategy of the basic delivery strategy based on the predicted consumer preference information;

[0126] S6023: Generate an initial campaign strategy based on the optimized campaign content and promotional strategies.

[0127] In this embodiment, the content delivered consists of related products whose prices are matched with predicted consumption levels, and the discount strategy is a promotional method adjusted according to the user's consumption level and preferences.

[0128] Specifically, based on predicted consumption level information, corresponding related products are matched to optimize the content of the basic advertising strategy. Based on predicted consumption preference information, the discount strategy of the basic advertising strategy is optimized. The optimized advertising content and discount strategy are combined to generate a preliminary advertising strategy.

[0129] In one embodiment, such as Figure 6 As shown, step S603 includes the following steps:

[0130] S6031: Grouping based on the interaction frequency corresponding to bidirectional correlation features, with groups including priority push group and interaction incentive group;

[0131] S6032: Group the propagation relationship chains with high-frequency interaction into priority push groups, optimize priority push based on the corresponding initial delivery strategy, and generate corresponding optimized delivery strategies.

[0132] S6033: Group the propagation relationship chains with low-frequency interaction and indirect interaction into interaction incentive groups, optimize the incentives based on the corresponding initial deployment strategy, and generate the corresponding optimized deployment strategy.

[0133] In this embodiment, the priority push group consists of user groups with high-frequency interaction relationships. Due to their strong social influence, they are reached first to maximize dissemination efficiency and improve conversion rates. The interaction incentive group consists of user groups with low-frequency or indirect interaction. Targeted incentives are used to enhance their participation and willingness to spread the word.

[0134] Specifically, based on the interaction frequency corresponding to the bidirectional association characteristics, the propagation relationship chains with high-frequency interaction are grouped into the priority push group. Priority push optimization is performed based on the corresponding initial delivery strategy to generate the corresponding optimized delivery strategy. Propagation relationship chains with low-frequency interaction and indirect interaction are grouped into the interaction incentive group. Incentive optimization is performed based on the corresponding initial delivery strategy to generate the corresponding optimized delivery strategy.

[0135] This data delivery optimization system based on multimodal data includes:

[0136] The data information acquisition module is used to acquire user identity information and multimodal data associated with the user identity information as input for the pre-trained profile building model. The multimodal data includes historical user behavior information, corresponding content feature information, and platform traffic type information.

[0137] The user profile building module is used to adjust the analysis weight of multimodal data based on platform traffic type information, and to build user profile information by analyzing the historical user behavior information and the content feature information.

[0138] The strategy generation module is used to generate a basic delivery strategy based on the user profile information output by the profile building model.

[0139] The social influence module is used to determine the propagation relationship chain based on user identity information, and the propagation relationship chain includes sharing users and receiving users;

[0140] The social analytics module is used to acquire historical sharing information and social network structure information between sharer and receiver users in the propagation relationship chain, and to build a social influence model.

[0141] The strategy optimization module is used to optimize the basic delivery strategy based on the bidirectional correlation features output by the social influence model, and generate an optimized delivery strategy.

[0142] For specific limitations regarding a data delivery optimization system based on multimodal data, please refer to the limitations of a data delivery optimization method based on multimodal data mentioned above, which will not be repeated here. Each module in the aforementioned data delivery optimization system based on multimodal data can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0143] The present invention and its embodiments have been described above. This description is not restrictive. The accompanying drawings are only one embodiment of the present invention. The actual structure is not limited to this. In short, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the present invention, such design should fall within the protection scope of the present invention.

Claims

1. A data delivery optimization method based on multimodal data, characterized in that: The method comprises the steps of: obtaining user identity information and obtaining multi-modal data associated with the user identity information as input of a pre-trained portrait establishment model, wherein the multi-modal data comprises historical user behavior information, corresponding content feature information and platform traffic type information; the portrait establishment model adjusts the analysis weight of the multi-modal data based on the platform traffic type information, analyzes the historical user behavior information and the content feature information, and establishes user portrait information; when receiving the user portrait information output by the portrait establishment model, a basic delivery strategy is generated based on the user portrait information; determining a propagation relationship chain based on the user identity information, wherein the propagation relationship chain comprises a sharer user and an accepter user; obtaining historical sharing information and social network structure information between the sharer user and the accepter user in the propagation relationship chain, and constructing a social influence model; optimizing the basic delivery strategy based on the bidirectional association features output by the social influence model to generate an optimized delivery strategy. 2.The method of claim 1, wherein: The portrait establishment model comprises a data classification layer, a weight adjustment layer and a feature fusion layer. The portrait establishment model adjusts the analysis weight of the multi-modal data based on the platform traffic type information, analyzes the historical user behavior information and the content feature information, and establishes user portrait information. The steps comprise the steps of: the data classification layer classifies the historical user behavior information and the corresponding content feature information according to different data modalities to obtain feature tensors of different modalities; the weight adjustment layer analyzes and adjusts the weight information of the feature tensors of different modalities based on the platform traffic type information; the feature fusion layer extracts and fuses the historical user behavior information and the content feature information based on the weight adjustment result, and outputs user portrait information. 3.The method of claim 1, wherein: The steps of obtaining historical sharing information between the sharer user and the accepter user in the propagation relationship chain and constructing a social influence model comprise the steps of: obtaining historical sharing information between the sharer user and the accepter user, and generating a low-dimensional vector of a user node based on each user as a node to generate social network structure information; constructing a social influence model based on the historical sharing information and the social network structure information. 4.The method of claim 1, wherein: Before the step of optimizing the basic delivery strategy based on the bidirectional association features output by the social influence model to generate an optimized delivery strategy, the following steps are further included: obtaining historical consumption information of the user as input of a pre-trained consumption prediction model, and matching the predicted consumption information generated based thereon to obtain corresponding predicted consumption level information.

5. The method of claim 4, wherein: The steps of obtaining historical consumption information of the user as input of a pre-trained consumption prediction model, and matching the predicted consumption information generated based thereon to obtain corresponding predicted consumption level information comprise the steps of: obtaining historical consumption information of the user, extracting consumption ability features and consumption preference features; inputting the consumption ability features and the consumption preference features into the pre-trained consumption prediction model to generate predicted consumption information and predicted consumption preference information; The predicted consumption information is compared based on a preset hierarchical matching rule to obtain corresponding predicted consumption hierarchical information.

6. The data delivery optimization method based on multi-modal data according to claim 5, characterized in that: The step of generating the optimized delivery strategy based on the bidirectional association feature output by the social influence model includes the steps of: The social influence model analyzes a social graph based on user identity information, determines an interaction frequency of a propagation relationship chain, and mines a second-degree propagation relationship chain to generate a bidirectional association feature. The step of generating the preliminary delivery strategy based on the predicted consumption hierarchical information corresponding to the user includes the steps of: The step of generating the optimized delivery strategy based on the bidirectional association feature includes the steps of:

7. The method of claim 6, wherein: The step of generating the bidirectional association feature based on the social influence model includes the steps of: The historical interaction information between the sharer user and the receiver user is obtained to determine the interaction frequency, and the bidirectional association feature corresponding to the high-frequency interaction relationship and the bidirectional association feature corresponding to the low-frequency interaction relationship are generated respectively. The second-degree propagation relationship chain formed based on indirect interaction is mined to generate the bidirectional association feature corresponding to the indirect interaction relationship. 8.The method of claim 6, wherein: The step of generating the preliminary delivery strategy based on the predicted consumption hierarchical information corresponding to the user includes the steps of: The delivery content of the basic delivery strategy is optimized based on the predicted consumption hierarchical information. The preferential strategy of the basic delivery strategy is optimized based on the predicted consumption preference information. The optimized delivery content and preferential strategy are combined to generate the preliminary delivery strategy. 9.The method of claim 6, wherein: The step of generating the optimized delivery strategy based on the bidirectional association feature includes the steps of: The interaction frequency corresponding to the bidirectional association feature is grouped into a priority push group and an interaction incentive group. The propagation relationship chain associated with the interaction frequency having a high-frequency interaction relationship is classified into the priority push group, and the corresponding preliminary delivery strategy is optimized for priority push to generate the corresponding optimized delivery strategy. The propagation relationship chain associated with the interaction frequency having a low-frequency interaction relationship and an indirect interaction relationship is classified into the interaction incentive group, and the corresponding preliminary delivery strategy is optimized for incentive to generate the corresponding optimized delivery strategy. 10.A data delivery optimization system based on multi-modal data, characterized in that: The data information acquisition module is configured to acquire user identity information and acquire multi-modal data associated with the user identity information as an input of the pre-trained portrait establishment model, wherein the multi-modal data includes historical user behavior information, corresponding content feature information, and platform traffic type information. The portrait establishment module is configured to adjust the analysis weight of the multi-modal data based on the platform traffic type information, analyze the historical user behavior information and the content feature information, and establish user portrait information. The strategy generation module is configured to generate a basic delivery strategy based on the user portrait information when receiving the user portrait information output by the portrait establishment model. The social influence module is configured to determine a propagation relationship chain based on user identity information, wherein the propagation relationship chain includes a sharer user and a receiver user. ​ The social analysis module is configured to acquire historical sharing information and social network structure information between a sharer user and a receiver user in a propagation relationship chain, and construct a social influence model. The strategy optimization module is configured to optimize a basic delivery strategy based on bidirectional association features output by the social influence model, and generate an optimized delivery strategy.

Images