A method for allocating equity products based on real-time user interaction information

Abstract

The present invention provides a method for configuring equity products based on real-time user interaction information, which belongs to the technical field of interactive information data processing. The present invention first constructs a multidimensional user behavior feature vector, forms a complete behavior sequence by capturing user interaction data in real time and combining it with historical interaction records; then models the user interaction state through an interactive state neural network combined with multidimensional data such as user geographic location, timestamp and device information; then assigns differentiated weights to the behavioral features, and generates an intention intensity matrix through a multi-source data fusion algorithm; then inputs the weighted behavioral feature data into an interactive perception model of a deep bidirectional attention network architecture to generate an equity recommendation list; finally, dynamically adjusts the recommendation strategy according to the user's real-time response data, performs optimized allocation, and determines the final equity delivery plan, thereby achieving accurate identification of the user's actual needs and intentions and efficient configuration of equity products.

Description

Technical Field

[0001] The present invention belongs to the technical field of interactive information data processing, and in particular relates to a method for configuring equity products based on real-time user interactive information. Background Art

[0002] In e-commerce and digital marketing, the precise allocation of equity products (such as discount coupons, vouchers, discount coupons, and points coupons) is a crucial tool for improving user conversion rates and promoting consumption. Traditional equity product allocation methods primarily rely on historical behavioral data from individual users. By collecting information such as browsing history, click behavior, and purchase history, they construct static user profiles and preference models to match and promote equity products. These methods, such as collaborative filtering-based recommendation algorithms, rule-based matching systems, and simple user profile analysis, have been widely used in e-commerce platforms, membership points systems, and marketing campaigns. However, traditional methods analyzing single-user behavioral data have significant limitations. First, these methods rely too heavily on historical behavioral data and fail to capture current user intentions and shifting needs in real time. Second, static user profiles fail to reflect the diverse needs of users across different scenarios and time points. Third, simple behavior frequency statistics and category matching lack a deep and qualitative understanding of user behavior, resulting in a mismatch between equity allocation and actual user needs. Especially in today's increasingly complex and dynamic world of user behavior, single-dimensional, static user behavior analysis is no longer sufficient for precision marketing. In actual application scenarios, users' demand for equity products is often influenced by multiple factors, including situational factors such as time, location, and social environment, as well as deeper psychological needs and decision-making intentions. Existing technologies mainly focus on surface behavioral data and lack the ability to perceive the deeper intentions reflected in users' real-time interactions, making it difficult to accurately identify users' true needs. This leads to the mismatch and waste of equity resources, and the equity products obtained by users do not match their actual needs. This not only reduces user experience and conversion rates, but also affects merchants' marketing effectiveness and resource utilization efficiency. There is an urgent need for a new equity product configuration method that can effectively identify users' actual needs and intentions. Summary of the Invention

[0003] In view of this, the present invention provides a method for configuring equity products based on real-time user interaction information, which can solve the technical problem in the prior art that equity product configuration based on single user behavior data cannot effectively identify the user's actual needs and intentions.

[0004] The present invention is implemented as follows: The present invention provides a method for configuring equity products based on real-time user interaction information, including: constructing a multi-dimensional user behavior feature vector; inputting user behavior feature data into an interactive state neural network, generating a user association state score, constructing a user relationship graph, and calculating a peer user group association vector; inputting the user relationship graph into a minimum spanning tree algorithm, performing a structured analysis on the peer user group, identifying key influencing nodes and calculating the equity propagation path within the group, and outputting a user influence index; introducing a user intention recognition module, assigning differentiated weight values ​​to user behavior feature data, and forming weighted behavior feature data; establishing an intention strength scoring system, and generating an intention strength matrix; calculating the similarity between the peer user group association vector and the standard association vector, determining the input weight coefficient, and generating an equity recommendation list; executing a multi-objective optimization allocation strategy, and determining the final equity investment plan.

[0005] The step of constructing a multi-dimensional user behavior feature vector is to obtain user behavior feature data by capturing user interaction data in real time and combining it with historical interaction records to form a behavior sequence.

[0006] Among them, the step of inputting user behavior feature data into the interactive state neural network specifically combines user geographic location data, timestamp data, and device information data to generate a user association state score, constructs a user relationship graph based on the user association state score, and calculates the association vector of the peer user group.

[0007] Among them, the interactive state neural network refers to a feedforward neural network structure used to analyze the interactive associations between users. The input layer receives user behavior feature data, geographic location data, timestamp data and device information data. The hidden layer uses a fully connected structure for feature extraction, and the output layer generates a user association state score representing the strength of user association.

[0008] Among them, the interactive state neural network is trained using the cosine similarity loss function, and the number of network parameters is controlled within 1 million to ensure real-time inference performance.

[0009] Among them, the peer user group association vector refers to a multidimensional feature vector that characterizes the association pattern between users, including geographical distance component, time synchronization component, interaction frequency component, interest overlap component and social relationship strength component. The vector dimension is 32, and each component ranges from 0 to 1.

[0010] Among them, the standard association vector refers to the ideal association state vector obtained through the association pattern statistics of the historical high-conversion rate user group. It serves as a reference benchmark for evaluating the association state of the actual user group. The vector dimension is consistent with the association vector of the peer user group.

[0011] Among them, the similarity weight function refers to a nonlinear function that maps the cosine similarity between the peer user group association vector and the standard association vector to an input weight coefficient. The function expression is a sigmoid function, and the steepness parameter is determined by optimizing historical conversion data. The output range is 0.5 to 1.5.

[0012] Among them, the minimum spanning tree algorithm refers to an algorithm that finds the connected subgraph with the minimum total weight in the user relationship graph. It is used to identify the most cost-effective equity propagation path in the user group. The key influencing node is the node with the highest degree centrality in the minimum spanning tree.

[0013] Among them, the user influence index is calculated by the weighted sum of the weights of the connecting edges of the key influence nodes in the minimum spanning tree.

[0014] The step of entering the user intention recognition module is to assign differentiated weight values ​​to the user behavior feature data according to the interaction depth value, stay time value, and interaction frequency value to form weighted behavior feature data.

[0015] The steps of the neutral intention strength scoring system are to input weighted behavioral feature data into the multi-source data fusion algorithm, calculate the user's intention scores for different product categories, and generate an intention strength matrix.

[0016] The source data fusion algorithm refers to a computational method that integrates users' explicit behavior data with implicit interest signals. It maps heterogeneous data into a unified feature space through tensor decomposition technology to achieve dimensional consistency and semantic relevance.

[0017] The benefit recommendation list refers to a set of ranked benefit products generated based on the intention intensity matrix and product attribute matching, which includes key attribute information such as benefit type, usage conditions, validity period and expected conversion rate, and dynamically adjusts the recommendation priority of products in the benefit recommendation list based on real-time user response data.

[0018] The multi-objective optimization allocation strategy involves taking the recommended equity list and user influence index as input, comprehensively considering individual conversion probabilities and group influence values, and determining the final equity allocation plan. A multi-objective optimization allocation strategy is a resource allocation method that simultaneously maximizes individual conversion rates and group influence under the constraint of limited equity resources. This problem is modeled as a constrained multi-objective optimization problem, using a Pareto optimal solution set as candidate solutions and calculating the overall score through a weighted summation method.

[0019] This invention achieves precise identification of users' actual needs and intentions through the construction of multidimensional user behavior feature vectors, real-time interaction state modeling, the introduction of a user intent recognition module, and multi-source data fusion. This approach not only focuses on historical user behavior patterns but also prioritizes the underlying intent signals expressed during real-time interactions, providing a more accurate basis for decision-making regarding equity product allocation. Compared to traditional technologies, this invention overcomes the limitations of analyzing single-user behavior data. By capturing user interaction information in real time and combining it with historical records to form behavioral sequences, it accurately grasps users' dynamic intentions. The system assigns differentiated weights to user behavior characteristics based on metrics such as interaction depth, dwell time, and interaction frequency, providing a more nuanced understanding of users' true needs. Furthermore, a multi-source data fusion algorithm integrates explicit user behavior with implicit interest signals to generate an intent intensity matrix that comprehensively reflects user intent. The interaction perception model employed in this invention utilizes a deep bidirectional attention network architecture, effectively capturing the complex interaction patterns between user behavior and product attributes, enabling precise matching of user intent with equity products. The system dynamically adjusts recommendation strategies based on real-time user response data, ensuring the timeliness and adaptability of equity allocation. Through these innovative technical means, the present invention successfully solves the core technical problem of the existing technology that is unable to effectively identify the actual needs and intentions of users, and significantly improves the accuracy of equity product configuration and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] like Figure 1 FIG. 1 is a flow chart of a method for configuring equity products based on real-time user interaction information provided by the present invention. The method includes the following steps:

[0023] S01. Construct a multi-dimensional user behavior feature vector by capturing user interaction data in real time and combining it with historical interaction records to form a behavior sequence, thereby obtaining user behavior feature data.

[0024] S02. Input the user behavior feature data into the interaction state neural network, combine the user's geographic location data, timestamp data, and device information data to generate a user association state score, construct a user relationship graph based on the user association state score, and calculate the association vector of the peer user group;

[0025] S03. Input the user relationship graph into a minimum spanning tree algorithm, perform a structured analysis on the peer user group, identify key influence nodes, calculate the equity propagation path within the group, and output the user influence index;

[0026] S04. Introducing a user intention recognition module to assign differentiated weights to the user behavior feature data according to the interaction depth value, the stay duration value, and the interaction frequency value, thereby forming weighted behavior feature data;

[0027] S05. Establish an intention strength scoring system, input the weighted behavioral feature data into a multi-source data fusion algorithm, calculate the user's intention scores for different product categories, and generate an intention strength matrix;

[0028] S06. Calculate the similarity between the peer user group association vector and the standard association vector, determine an input weight coefficient using a similarity weight function, multiply the intention intensity matrix and the user influence index by the input weight coefficient, and input the resultant weight coefficient into a pre-trained interaction perception model to generate a benefit recommendation list. Dynamically adjust the recommendation priority of products in the benefit recommendation list based on real-time user response data.

[0029] S07. Execute a multi-objective optimization allocation strategy, take the equity recommendation list and the user influence index as input, comprehensively consider the individual conversion probability value and the group influence value, and determine the final equity allocation plan.

[0030] Among them, the interactive state neural network specifically refers to a feedforward neural network structure used to analyze interactive associations between users. The input layer receives user behavior feature data, geographic location data, timestamp data and device information data. The hidden layer uses a fully connected structure for feature extraction. The output layer generates a user association state score representing the strength of user association. The interactive state neural network is trained using a cosine similarity loss function, and the number of network parameters is controlled within 1 million to ensure real-time inference performance.

[0031] Among them, the peer user group association vector specifically refers to a multidimensional feature vector that characterizes the association pattern between users, including a geographical distance component, a time synchronization component, an interaction frequency component, an interest overlap component and a social relationship strength component. The vector dimension is 32, and each component ranges from 0 to 1, which is calculated through the user association status score.

[0032] The standard association vector specifically refers to an ideal association state vector obtained through association pattern statistics of historical high-conversion rate user groups, which serves as a reference benchmark for evaluating the association state of actual user groups. The vector dimension is consistent with the association vector of the peer user group.

[0033] Among them, the similarity weight function specifically refers to a nonlinear function that maps the cosine similarity between the peer user group association vector and the standard association vector to an input weight coefficient. The function expression is a sigmoid function, and the steepness parameter is determined by optimizing historical conversion data. The output range is 0.5 to 1.5. The higher the similarity, the larger the output weight coefficient.

[0034] Among them, the minimum spanning tree algorithm specifically refers to an algorithm for finding a connected subgraph with the minimum total weight in the user relationship graph, which is used to identify the most cost-effective equity propagation path in the user group. The edge weight is used to represent the strength of the influence relationship between users, and the node weight represents the user's influence on group decision-making. The key influence node is the node with the highest degree centrality in the minimum spanning tree. The user influence index is calculated by the weighted sum of the weights of the edges connecting the key influence node in the minimum spanning tree.

[0035] Among them, the multi-source data fusion algorithm specifically refers to a computational method for integrating user explicit behavior data with implicit interest signals, mapping heterogeneous data to a unified feature space through tensor decomposition technology to achieve dimensional consistency and semantic relevance, inputting the weighted behavior feature data, and outputting the intention intensity matrix.

[0036] Among them, the rights recommendation list specifically refers to a set of ranked rights products generated based on the intention intensity matrix and the matching degree of product attributes, which contains key attribute information such as rights type, usage conditions, validity period and expected conversion rate. Rights types include discount coupons, vouchers, full-reduction coupons, and points coupons. Usage conditions include the scope of applicable products, minimum consumption amount, and usage time limit. The validity period includes start time, end time, and valid days after activation. The expected conversion rate is obtained based on statistical analysis of historical data.

[0037] Among them, the multi-objective optimization allocation strategy specifically refers to a resource allocation method that maximizes both individual conversion rate and group influence under the constraint of limited equity resources. The problem is modeled as a multi-objective optimization problem with constraints, and the Pareto optimal solution set is used as the candidate solution. The comprehensive score is calculated by the weighted sum method, and the solution with the highest comprehensive score is selected as the final equity investment solution.

[0038] Among them, the specific structure of the interactive perception model is a deep bidirectional attention network architecture, which consists of a user behavior encoding layer, a product attribute encoding layer, a cross-modal interaction layer and a multi-head self-attention layer. The core mechanism is a bidirectional attention mechanism and an adaptive feature fusion module. The number of attention heads is dynamically adjusted according to the dimension of the association vector of the peer user group. The feature fusion depth is determined by the user association status score, and the activation function slope parameter is determined by the mean of the intention intensity matrix.

[0039] The steps of establishing the training data set of the interactive perception model specifically include collecting multi-scenario user interaction behavior data including browsing trajectory data, click stream data, dwell time data, interactive operation data and conversion event data, extracting individual user characteristics and group association characteristics, constructing a positive sample set including successful rights conversion cases and corresponding pre-interaction sequences, constructing a negative sample set including failed rights push cases and corresponding pre-interaction sequences, dividing the training set, validation set and test set in chronological order, cleaning, standardizing and enhancing the training data, and generating a multimodal training sample library containing sequence features, graph structure features and label information.

[0040] The steps of training the interactive perception model specifically include randomly initializing the model network parameters, setting the initial values ​​of the weights using the Xavier initialization method, designing a joint loss function to simultaneously optimize the two goals of individual conversion prediction and group influence prediction, adopting a hierarchical learning rate strategy to apply differentiated update strengths to different network layers, implementing a batch gradient descent algorithm for parameter optimization, applying an early stopping strategy to avoid overfitting, periodically evaluating the validation set performance indicators including individual conversion rate prediction accuracy and group response consistency, dynamically adjusting the attention mechanism parameters and feature fusion strength, freezing the best-performing model parameters as a pre-trained model for online deployment, and designing a model compression strategy to achieve lightweightness to meet real-time inference requirements.

[0041] The specific implementation of the above steps is described in detail below.

[0042] The specific implementation method of step S01 is to obtain user interaction behavior data through multiple acquisition channels and construct a multi-dimensional behavior feature vector. First, collect explicit interaction data such as user clicks, slides, and stays, and record raw data including interaction type, interaction object, interaction duration, and interaction frequency; then, combine the user's historical interaction records to construct a behavior sequence according to the time series, and the sequence length is set to the interaction behavior within the last 48 hours; then perform feature extraction on the behavior sequence, including behavior frequency statistics, time interval analysis, and behavior conversion path identification; then apply the sequence embedding algorithm to convert the behavior sequence into a fixed-dimensional vector, and the vector dimension is set to 128 dimensions; finally, perform normalization processing to unify the feature value range of each dimension to 0~1, forming standardized user behavior feature data. The sequence embedding algorithm used in this step is based on the long short-term memory network, which can effectively capture the temporal dependency of user behavior, extract feature representations that contain the user's potential intentions, and lay the data foundation for subsequent analysis.

[0043] The specific implementation of step S02 involves constructing a user relationship graph and calculating group association vectors. First, user behavior feature data is concatenated with geographic location data, timestamp data, and device information data to form an input vector. This input vector is then fed into an interactive state neural network, which consists of an input layer, three fully connected hidden layers, and an output layer. The number of hidden layer neurons is 512, 256, and 128, respectively, and uses a ReLU activation function. User association status scores are then obtained from the network output, ranging from 0 to 100 and representing the strength of associations between users. A user relationship graph is then constructed based on the user association status scores, where nodes represent users and edges represent user associations. Edge weights are determined by the association status scores, and only edges with a score greater than 30 are retained. Finally, the peer user group association vector is calculated, extracting a 32-dimensional vector consisting of geographic distance, temporal synchrony, interaction frequency, interest overlap, and social relationship strength. This step leverages the powerful expressive power of neural networks to effectively mine implicit association patterns between users, providing key data support for subsequent group impact analysis.

[0044] The specific implementation of step S03 is to apply the minimum spanning tree algorithm to analyze the user group structure. First, the user relationship graph constructed in step S02 is used as input. The edge weight is defined as 100 minus the user association status score, ensuring that the higher the association strength, the smaller the weight. Then, the Kruskal algorithm or Prim algorithm is used to construct a minimum spanning tree to find the subgraph connecting all users with the minimum total weight. Next, the degree centrality of each node in the spanning tree is calculated. The degree centrality calculation formula is the number of edges connecting the node divided by the maximum possible number of connections. The node with the highest degree centrality is then identified as the key influence node. The threshold is set to 0.6, that is, nodes with a degree centrality greater than 0.6 are considered key influence nodes. Finally, the user influence index is calculated by weighted summing the weights of the edges connecting the key influence nodes. The weight coefficient is inversely proportional to the level of the edge in the spanning tree. The higher the level, the lower the weight coefficient. The influence index range is standardized to 0-100. This step effectively identifies key influence nodes in the group through graph theory algorithms, quantitatively assesses user influence, and lays the foundation for the subsequent optimization of equity resource allocation.

[0045] The specific implementation method of step S04 is to introduce a user intention recognition module and perform weighted processing on the user behavior feature data. First, three key indicators, namely interaction depth value, dwell time value, and interaction frequency value, are extracted from the user behavior feature data; then a weight distribution model is designed, with the interaction depth weight range being 1.2 to 1.8, the dwell time weight range being 0.8 to 1.5, and the interaction frequency weight range being 1.0 to 1.6; then specific weight values ​​are set based on the behavior type, with higher weights given to direct conversion behaviors such as adding to a shopping cart or adding to favorites, and lower weights given to browsing behaviors; then corresponding weight values ​​are applied to the behavioral feature data of each dimension, and the weighting formula is the original value of the feature multiplied by the corresponding weight value; finally, the weighted results are normalized to ensure the rationality of the proportions of different feature dimensions, thereby forming weighted behavioral feature data. This step highlights users' high-value interactive behaviors through differentiated weight distribution, enhances the ability to identify users' true intentions, and improves the accuracy of subsequent recommendations.

[0046] The specific implementation of step S05 is to establish an intent strength scoring system. First, the weighted behavioral feature data obtained in step S04 is used as input, combined with multi-source data such as the user's historical purchase history, search keywords, and favorite products. A multi-source data fusion algorithm is then applied. This algorithm, based on tensor decomposition technology, maps heterogeneous data into a unified 64-dimensional feature space. Reliability coefficients are then set for data from different sources: the coefficient for historical purchase behavior data is 0.8-1.0, the coefficient for browsing behavior data is 0.3-0.6, and the coefficient for search behavior data is 0.5-0.8. The user's intent score for each product category is then calculated, ranging from 0 to 10. The scoring formula is the sum of the contribution score of each data source multiplied by the reliability coefficient. Finally, an intent strength matrix is ​​generated. The matrix dimension is the number of users multiplied by the number of product categories, and the matrix elements are the intent scores of the corresponding user for the corresponding product category. This step uses multi-source data fusion technology to comprehensively evaluate the strength of user intent and generate a structured intent representation, providing data support for accurate rights and interests matching.

[0047] The specific implementation of step S06 is to generate and dynamically adjust the equity recommendation list. First, calculate the cosine similarity between the peer user group association vector and the standard association vector. The cosine similarity calculation formula is the inner product of the two vectors divided by the product of the two vectors' modulus lengths; then determine the input weight coefficient through the similarity weight function. The function adopts the sigmoid function, and the expression is: , where k is the steepness parameter, ranging from 5 to 10, and x is the cosine similarity, with an output range of 0.5 to 1.5. The intent strength matrix and user influence index are then multiplied by the input weight coefficients. The weighted data is then fed into a pre-trained interaction perception model, which utilizes a deep bidirectional attention network architecture. Finally, a list of equity recommendations is generated, and the priority of recommendations is adjusted every 30 minutes based on real-time user response data, such as click-through rate, redemption rate, and sharing rate. The priority is increased proportionally to the degree of positive response, with a maximum increase of 30%. This step enhances the model's sensitivity to high-quality groups through a similarity-driven weight adjustment mechanism. It also dynamically optimizes the model based on real-time feedback, improving the adaptability of the recommendation system.

[0048] The specific implementation method of step S07 is to execute a multi-objective optimization allocation strategy. First, the equity recommendation list and user influence index are used as input data; then the optimization objectives are set, including maximizing the individual conversion rate and maximizing the group influence diffusion; then the constraints are defined, including the total equity quantity limit, the maximum equity quantity limit for a single user, the equity type diversity requirement, etc.; then the problem is modeled as a multi-objective optimization problem with constraints, and converted into a single-objective problem using the weighted summation method, with the weight ratio of individual conversion rate 0.6 and group influence 0.4; finally, the optimal allocation plan is solved by the genetic algorithm, with the population size set to 100, the number of iterations to 50 to 100, the roulette wheel method used for the selection operation, the crossover probability to 0.8, the mutation probability to 0.1, and finally the individual with the highest fitness is selected as the equity allocation plan. This step uses multi-objective optimization technology to balance the two dimensions of individual conversion and group influence, achieve efficient resource allocation, and improve the overall marketing effect.

[0049] It should be noted that the present invention has achieved a qualitative breakthrough in the configuration of equity products by deeply mining the interactive correlation information of user groups. The interactive state neural network established by the system not only analyzes the behavior of a single user, but also pays attention to the dynamic interaction pattern between users. It constructs a multi-dimensional peer user group correlation vector through multi-dimensional features such as geographic location, time synchronization, and interest overlap, and accurately depicts the influence transmission mechanism within the group. The combination of the user relationship graph and the minimum spanning tree algorithm enables the system to identify the key influencing nodes in the group decision-making process, depict the optimal propagation path of equity value within the group, and realize the transformation from point-to-point push to point-to-surface propagation. The similarity weight function innovatively compares the actual group correlation pattern with the historical high conversion standard correlation vector, dynamically adjusts the recommendation strategy strength, and enables the system to have the ability to adapt to changes in group structure. The bidirectional attention mechanism and adaptive feature fusion module of the interactive perception model can simultaneously process individual demand signals and group influence evaluation. The number of attention heads is flexibly adjusted according to the dimension of the group correlation vector to ensure that the influence path is accurately captured in complex social network structures. The multi-objective optimization allocation strategy takes individual conversion probability and group influence as dual objectives, finding the resource allocation equilibrium point through a Pareto optimal solution set to achieve the social network amplification effect of equity value. This group interaction-aware recommendation method significantly improves the accuracy and coverage efficiency of equity delivery, prioritizing the allocation of limited equity resources to key node users with fission propagation potential, forming a self-reinforcing group marketing ecosystem. It also significantly enhances recommendation accuracy, reduces ineffective equity delivery, and improves overall marketing efficiency. Furthermore, it maximizes the transfer of equity value through the diffusion of influence at key nodes, creating a geometric amplification effect for limited resources and forming a sustainable positive marketing cycle.

[0050] Specifically, the core technology behind this invention, which addresses the problem of identifying actual user needs and intent, lies in building a framework for allocating equity products that integrates multi-dimensional feature extraction, real-time interaction state modeling, and deep intent recognition. This framework, through systematic technical means, achieves breakthroughs in identifying user intent, moving from static to dynamic, and from superficial to deep.

[0051] At the user feature characterization level, the present invention breaks through the traditional isolated behavior analysis model. By capturing user interaction data in real time and combining it with historical records to form a behavior sequence, a multi-dimensional user behavior feature vector containing time series information is constructed. This dynamic feature characterization method can more comprehensively portray the trajectory of changes in user behavior patterns and provide a rich information basis for intent identification. Unlike traditional methods, the present invention not only focuses on the frequency of behavior, but also pays more attention to the quality characteristics of behavior. By assigning differentiated weights to indicators such as interaction depth, dwell time, and interaction frequency, the system can distinguish the differences in the intensity of intentions behind different behaviors and more accurately understand the real needs of users.

[0052] At the interaction state modeling level, this invention innovatively proposes an interaction state neural network that integrates user behavior characteristics with contextual data such as geographic location, timestamp, and device information, achieving a comprehensive perception of the user's interaction state. Through the training mechanism of the cosine similarity loss function, the network can effectively capture the differences in user intentions in different contexts, overcoming the shortcomings of traditional methods that lack consideration of contextual factors. This interaction state modeling method that integrates contextual information enables the system to understand the changes in user needs under certain temporal and spatial conditions, significantly improving the accuracy of intent recognition.

[0053] At the level of intention recognition and matching, the present invention establishes an intention strength scoring system, inputs weighted behavioral feature data into a multi-source data fusion algorithm, and generates an intention strength matrix that characterizes the user's interest in different product categories. The multi-source data fusion algorithm uses tensor decomposition technology to map the user's explicit behavioral data and implicit interest signals to a unified feature space, achieving dimensional consistency and semantic relevance, and can fully capture the user's multi-level intention information. The interactive perception model adopts a deep bidirectional attention network architecture. Through the synergy of the user behavior encoding layer, the product attribute encoding layer, the cross-modal interaction layer, and the multi-head self-attention layer, it accurately analyzes the complex interaction pattern between user behavior and product attributes, and achieves the optimal match between user intention and equity products.

[0054] In terms of a dynamic adjustment mechanism, this invention dynamically adjusts the recommendation priority of items in the equity recommendation list based on real-time user response data, forming a closed-loop feedback system. This dynamic adjustment mechanism enables the system to promptly correct deviations in intent recognition and continuously optimize equity allocation strategies, ensuring the system's adaptability and robustness. By designing a joint loss function and applying a hierarchical learning rate strategy, the interactive perception model can continuously improve the accuracy of intent recognition during a continuous learning process.

[0055] In summary, the present invention organically combines technical means such as multi-dimensional feature representation, real-time interaction state modeling, deep intent recognition and dynamic adjustment mechanism to construct a rights and interests product configuration system that can effectively identify users' actual needs and intentions. It fundamentally solves the core problems in the existing technology and realizes the precision and personalization of rights and interests product configuration.

[0056] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0057] The specific implementation of step S01 is to obtain user interaction behavior data through multiple acquisition channels and construct a multi-dimensional behavior feature vector. First, collect explicit interaction data such as user clicks, slides, and stays, and record original data including interaction type, interaction object, interaction duration, interaction frequency, etc.; then combine the user's historical interaction records to construct a behavior sequence according to the time series, and the sequence length is set to the interaction behavior within the last 48 hours; then extract features from the behavior sequence, including behavior frequency statistics, time interval analysis, and behavior conversion path identification; then apply the sequence embedding algorithm to convert the behavior sequence into a fixed-dimensional vector. The sequence embedding algorithm is based on the long short-term memory network, and for the input behavior sequence ,in Representative behavioral events, and the characteristics of each behavioral event are expressed as ,in Indicates the interaction type code, Represents the interaction object encoding, Indicates the interaction time. Represents the interaction frequency. The sequence embedding calculation formula is as follows:

[0058] ;

[0059] ;

[0060] Where, Represents the long short-term memory network at time step The hidden state of , has a dimension of 128; Represents the computational function of the long short-term memory network; Represents the final behavior sequence embedding vector, with a dimension of 128; Represents the mapping function of the fully connected layer, which maps the hidden state after average pooling to the feature space. Finally, normalization is performed, and the normalization formula is:

[0061] ;

[0062] Where, Represents the normalized eigenvector, with a value range of 0 to 1; and Represents vectors This step uses sequence embedding technology to convert temporal behavior data into fixed-dimensional feature representations, effectively capturing the temporal dependencies of user behavior and laying the data foundation for subsequent analysis.

[0063] The specific implementation of step S02 is to construct a user relationship graph and calculate the group association vector. First, the user behavior feature data is combined with the geographic location data, timestamp data, and device information data to form an input vector , specifically expressed as:

[0064] ;

[0065] Where, is the input vector after feature concatenation, with a dimension of 162; The normalized user behavior feature vector obtained in step S01 has a dimension of 128; is the geographic location data vector with a dimension of 10; is the timestamp data vector with a dimension of 8; is the device information data vector with a dimension of 16. Then the input vector The input interactive state neural network is a feedforward neural network structure consisting of an input layer, three fully connected hidden layers, and an output layer. The network forward propagation calculation process is as follows:

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] Where, 、 、 、 are the weight matrices of each layer respectively; 、 、 、 are the bias vectors of each layer respectively; are the linear transformation outputs of each layer respectively; are the outputs of each hidden layer after the activation function; is the rectified linear unit activation function, defined as ; is the S-type activation function, defined as ; Score the user association status from 0 to 100. Then build a user relationship graph based on the user association status score. ,in Represents a set of user nodes, represents the set of user association edges, Represents the edge weight set. The construction conditions are:

[0075] ;

[0076] Where, Represents a user and users The threshold value of the association status score between them is set to 30, that is, only the edges with an association status score greater than 30 are retained. Finally, the association vector of the peer user group is calculated. , with a dimension of 32 and containing five components, is calculated as follows:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] Where, is the geographical distance component, with a dimension of 8; Represents a user and users The geographical distance between them, in meters; Indicates the maximum considered distance, set to 10,000 meters; is the temporal synchronization component, with a dimension of 6; and Represents users and users The activity timestamp; is the time decay coefficient, set to 3600 seconds; is the interaction frequency component, with a dimension of 6; Represents a user Frequency of interaction; is the frequency threshold, set to 10 times / hour; is the interest overlap component, with a dimension of 6; and Represents users and users interest collection; is the social relationship strength component, with a dimension of 6; Represents a user and users This step calculates the user association status score through a neural network, constructs a user relationship graph, and extracts multi-dimensional group association feature vectors, providing key data support for subsequent user group impact analysis.

[0084] The specific implementation of step S03 is to apply the minimum spanning tree algorithm to analyze the user group structure. First, the user relationship graph constructed in step S02 is As input, define the edge weight conversion function:

[0085] ;

[0086] Where, is the edge weight after conversion; For users and users This transformation ensures that the higher the association strength, the smaller the corresponding edge weight. Then the Kruskal algorithm is applied to construct the minimum spanning tree. ,in and , the algorithm process is as follows:

[0087] (1) All edges are weighted Sort from smallest to largest;

[0088] (2) Initially, each node constitutes an independent connected component;

[0089] (3) Traverse each edge from small to large according to weight , if the node and nodes If the edge is not in the same connected component, add it to the minimum spanning tree. , and merge these two connected components;

[0090] (4) Repeat step (3) until the minimum spanning tree contains Edge.

[0091] Then calculate the degree centrality of each node in the spanning tree , the calculation formula is:

[0092] ;

[0093] Where, Representation node degree centrality of ; Representation node In the minimum spanning tree The degree of the node The number of connected edges; Represents the total number of nodes. The node with the highest degree centrality is then identified as the key influencing node, and the judgment conditions are:

[0094] ;

[0095] Where, Represents the set of key influential nodes; the threshold is set to 0.6, that is, nodes with degree centrality greater than 0.6 are considered key influential nodes. Finally, the user influence index is calculated , the calculation formula is:

[0096] ;

[0097] ;

[0098] Where, Represents a user The influence index of All directly connected edges; is the weight coefficient, and the node Level in the spanning tree Related, the higher the level, the lower the weight coefficient; is the edge weight after conversion. Finally, the influence index is standardized to the range of 0 to 100:

[0099] ;

[0100] Where, For standardized users The influence index of and The minimum and maximum values ​​of the influence index among all users are represented respectively. This step effectively identifies key influential nodes in the group through graph theory algorithms, quantitatively assesses user influence, and lays the foundation for the subsequent optimization of equity resource allocation.

[0101] The specific implementation of step S04 is to introduce a user intention recognition module to perform weighted processing on the user behavior feature data. First, three key indicators, namely, interaction depth value, stay time value, and interaction frequency value, are extracted from the user behavior feature data to form a feature vector. :

[0102] ;

[0103] Where, Represents the interaction depth feature subvector; Represents the characteristic subvector of the length of stay; Represents the interaction frequency feature vector. Then design the weight distribution vector :

[0104] ;

[0105] Where, Represents the interaction depth weight sub-vector, ranging from 1.2 to 1.8; Represents the length of stay weight sub-vector, with a value range of 0.8 to 1.5; Represents the interaction frequency weight subvector, with a value range of 1.0 to 1.6. The specific weight value is determined by the behavior type, and the calculation formula is:

[0106] ;

[0107] ;

[0108] ;

[0109] Where, Indicates the The importance score of each behavior type ranges from 0 to 10, with higher scores for direct conversion behaviors such as adding to a shopping cart and adding to favorites, and lower scores for browsing behaviors; Indicates the The standard score of the duration of stay of each behavior ranges from 0 to 10; Indicates the The frequency standard score of each behavior ranges from 0 to 10. Then, the corresponding weight value is applied to the behavioral feature data of each dimension to calculate the weighted feature vector :

[0110] ;

[0111] Where, Represents the Hadamard product operator, that is, multiplication of corresponding elements; Represents the weighted eigenvector. Finally, the weighted result is normalized and the calculation formula is:

[0112] ;

[0113] Where, Represents the normalized weighted feature vector, i.e., weighted behavioral feature data; and Represents vectors This step highlights users’ high-value interactive behaviors through differentiated weight allocation, enhances the ability to identify users’ true intentions, and improves the accuracy of subsequent recommendations.

[0114] The specific implementation of step S05 is to establish an intention intensity scoring system. First, the weighted behavior feature data obtained in step S04 is As input, combined with multi-source data such as user historical purchase records, search keywords, favorite products, etc., a multi-source feature matrix is ​​formed :

[0115] ;

[0116] Where, represents the multi-source feature matrix; Represents the historical purchase behavior feature vector; Represents the search behavior feature vector; represents the feature vector of the collection behavior. Then, a multi-source data fusion algorithm is applied. This algorithm is based on tensor decomposition technology to map heterogeneous data into a unified feature space. The tensor decomposition process is expressed as:

[0117] ;

[0118] Where, Represents the original multi-source data tensor; represents the core tensor; 、 、 Represent the factor matrices of three dimensions respectively; Indicates that along By tensor decomposition, heterogeneous data is mapped to a unified 64-dimensional feature space to obtain a unified representation vector Then set the reliability coefficient vector for data from different sources :

[0119] ;

[0120] Where, Represents the reliability coefficient of historical purchase behavior data, ranging from 0.8 to 1.0; Represents the reliability coefficient of browsing behavior data, ranging from 0.3 to 0.6; Represents the reliability coefficient of search behavior data, ranging from 0.5 to 0.8. Then calculate the user's intention score matrix for each product category , the calculation formula is:

[0121] ;

[0122] Where, Represents a user For product categories intention score, ranging from 0 to 10; Represents users in the unified representation space For product categories In the data source The eigenvalues ​​on ; Represents a data source Reliability coefficient; Indicates the total number of data sources. Finally, the intent strength matrix is ​​generated , which is the normalized intention score matrix:

[0123] ;

[0124] Where, Represents the intention intensity matrix; and Represents matrices respectively This step uses multi-source data fusion technology to comprehensively evaluate the strength of user intent and generate a structured intent representation, providing data support for accurate rights matching.

[0125] The specific implementation of step S06 is to generate and dynamically adjust the equity recommendation list. First, calculate the peer user group association vector With standard correlation vector Cosine similarity of , the calculation formula is:

[0126] ;

[0127] Where, Represents a vector With vector The cosine similarity of represents the inner product of two vectors; and Represents vectors and vector of Norm. Then the input weight coefficient is determined by the similarity weight function , the function adopts the sigmoid function, and the expression is:

[0128] ;

[0129] Where, Indicates the input weight coefficient, the value range is 0.5~1.5; Represents the steepness parameter, with a value of 5 to 10, which is determined based on historical conversion data; represents the cosine similarity. Then the intention intensity matrix and user influence index Multiply by the input weight coefficient :

[0130] ;

[0131] ;

[0132] Where, represents the weighted intention intensity matrix; Represents the weighted user influence index. The weighted data is then input into the pre-trained interactive perception model to obtain the initial equity recommendation list Each benefit item contains the following attributes: benefit type, usage conditions, validity period, expected conversion rate, etc. Finally, the recommendation priority is dynamically adjusted based on the user's real-time response data. The adjustment formula is:

[0133] ;

[0134] Where, Indicates equity items Adjusted priority of Indicates the initial priority; Indicates the adjustment coefficient, the maximum value is 0.3; Indicates equity items Real-time response indicators, such as click-through rate, conversion rate, sharing rate, etc. represents the average response index of all equity items; Represents the standard deviation of the response metric. Adjustments are made every 30 minutes, with the priority increase proportional to the degree of positivity. This step uses a similarity-driven weight adjustment mechanism to enhance the model's sensitivity to high-quality groups. It also dynamically optimizes the model based on real-time feedback, improving the adaptability of the recommendation system.

[0135] The specific implementation of step S07 is to execute a multi-objective optimization allocation strategy. First, the equity recommendation list and user influence index As input data; then set the optimization goal, including maximizing the individual conversion rate objective function and maximize the group influence diffusion objective function :

[0136] ;

[0137] ;

[0138] Where, Indicates the equity distribution plan, Indicates whether the equity Assign to user , the value is 0 or 1; Represents a user Equity Expected conversion rate; Represents a user The influence index of Expressing Rights The propagation coefficient is related to the equity type and attributes; Indicates the number of users; Indicates the number of equity types. Then define the constraints:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] Where, Indicates the total equity quantity limit; Indicates the maximum number of stakes per user, the default value is 3; Indicates the maximum number of users to which a single equity can be allocated; Indicates the The minimum number of equity types to be allocated is determined to ensure the diversity of equity types. The problem is then modeled as a multi-objective optimization problem with constraints and converted into a single-objective problem using the weighted summation method:

[0144] ;

[0145] Satisfy all the above constraints

[0146] Where, represents the comprehensive objective function; and They represent the weights of individual conversion rate target and group impact target respectively, with default values ​​of 0.6 and 0.4 respectively. Finally, the optimal allocation scheme is solved by genetic algorithm, and the algorithm parameters are set as follows: the population size is 100, the chromosome encoding adopts binary encoding, and each position represents the corresponding The selection operation uses the roulette wheel method, and the selection probability is proportional to the fitness. The crossover operation uses single-point crossover with a crossover probability of 0.8. The mutation operation uses bit reversal mutation with a mutation probability of 0.1. The termination condition is to reach the maximum number of iterations of 50 to 100 or the optimal solution has not improved for 10 consecutive generations. The fitness function is defined as:

[0147] ;

[0148] Where, Represents an individual The fitness value of This represents a penalty function for violating a constraint, proportional to the degree of violation. Ultimately, the individual with the highest fitness is selected as the equity investment plan. This step uses multi-objective optimization techniques to balance individual conversion and group influence, achieving efficient resource allocation and improving overall marketing effectiveness.

[0149] In this embodiment, the detailed structure of the interactive state neural network is as follows: The network is a feedforward neural network used to analyze the interactive association pattern between users. The input layer consists of 4 parts: 128 user behavior feature data input units, 10 geographic location data input units, 8 timestamp data input units, 16 device information data input units, and the total number of neurons in the input layer is 162. The hidden layer adopts a fully connected structure and contains 3 layers. The number of neurons in the first hidden layer is 512, the second hidden layer is 256, and the third hidden layer is 128. The layers are fully connected, and the activation function adopts the ReLU function. The number of neurons in the output layer is 1, which is used to generate a user association state score ranging from 0 to 100. The output layer activation function adopts the sigmoid function and is linearly scaled to match the target range. The network is trained using the cosine similarity loss function, which is expressed as ,in is the predicted value, The actual value is . To ensure real-time inference performance, the number of network parameters is strictly controlled within 1 million. The specific parameters are calculated as follows: the first-layer weight matrix is ​​162×512 = 82,944 parameters, the first-layer bias is 512 parameters, the second-layer weight matrix is ​​512×256 = 131,072 parameters, the second-layer bias is 256 parameters, the third-layer weight matrix is ​​256×128 = 32,768 parameters, the third-layer bias is 128 parameters, the output layer weight matrix is ​​128×1 = 128 parameters, and the output layer bias is 1 parameter, for a total of 247,809 parameters. The network is trained using the batch gradient descent algorithm, with an initial learning rate of 0.001 and a cosine annealing strategy for dynamic learning rate adjustment. The batch size is set to 64, and the number of training epochs is 100.

[0150] In this embodiment, the detailed steps for establishing the training data set of the interactive state neural network are as follows: first, determine the target user group, screen users with high activity and rich behavioral data as the initial sample set, the number of users is not less than 50,000, covering people of different age groups, regions, and consumption abilities, to ensure sample diversity; then collect user interactive behavior data, including at least 3 months of original data such as click behavior, browsing behavior, purchase behavior, social behavior, etc., and the number of behavior records for each user is not less than 100; then collect user geographic location data, including permanent location, activity trajectory, travel radius, etc., to ensure that the location accuracy is within 100 meters; then collect timestamp data, record the time point, duration, periodic characteristics and other time series information of the user activity, and the time accuracy is controlled at the second level; at the same time, collect device information data, including technical parameters such as device type, operating system, network environment, application version, etc.; then clean and preprocess the collected original data, remove outliers, missing values ​​and noise data, and the abnormal judgment criteria Filling or removal operations are performed for data points that are beyond 3 standard deviations of the normal distribution and features with a missing rate of more than 20%; then label data is constructed, and user association status scores are automatically generated through expert annotation or rule-based automatic generation. The annotation rules include weighted calculation of multi-dimensional indicators such as interaction frequency, content similarity, and spatiotemporal overlap to generate association strength scores in the range of 0 to 100; then feature engineering is performed to reduce the dimension, combine and transform the original features to generate a more expressive feature set. The dimensionality reduction method uses principal component analysis to retain the principal components that can explain 85% of the variance. The feature combination uses cross-feature generation technology, and the feature transformation uses normalization and standardization processing; then the dataset is divided into training set, validation set and test set in chronological order, with a ratio of 8:1:1, to avoid data leakage problems caused by random partitioning; finally, data enhancement is performed to expand the training samples through interpolation, sampling, adding slight noise, etc. to enhance the generalization ability of the model. The size of the enhanced training set is expanded to 1.5 times the original data. The entire training dataset construction process strictly adheres to data quality control standards, and regularly conducts data consistency checks and distribution deviation monitoring to ensure that the data foundation for model training is reliable and effective.

[0151] In this embodiment, the detailed structure of the interactive perception model is as follows: The model adopts a deep bidirectional attention network architecture, and its core design concept is to capture the complex interactive relationship between user behavior and product attributes. The model consists of four main components: user behavior encoding layer, product attribute encoding layer, cross-modal interaction layer, and multi-head self-attention layer. The user behavior encoding layer adopts a bidirectional long short-term memory network structure with 256 hidden units. The input is the user behavior sequence and the output is the user behavior representation vector. The product attribute encoding layer adopts a multi-layer perceptron structure with 3 layers. The number of neurons in each layer is 128, 64, and 32 respectively. The input is the product attribute feature and the output is the product attribute representation vector. The cross-modal interaction layer realizes bidirectional attention interaction between user behavior representation and product attribute representation. The number of attention heads is dynamically adjusted according to the dimension of the association vector of the peer user group, and is basically set to 8 attention heads. The multi-head self-attention layer further extracts the interaction features. The number of layers is 2, and the number of attention heads is the same as that of the cross-modal interaction layer. The model also includes an adaptive feature fusion module. The fusion depth is determined by the user's association status score. A score range of 0 to 30 corresponds to one-level fusion, 31 to 70 corresponds to two-level fusion, and 71 to 100 corresponds to three-level fusion. The model employs a multi-task learning framework to simultaneously optimize both individual conversion prediction and group influence prediction. The loss function is a weighted sum of the two task losses, with a weight ratio of 0.7 for individual conversion prediction and 0.3 for group influence prediction. The total number of model parameters is approximately 5 million, which has been reduced to less than 1 million through model compression techniques such as knowledge distillation and weight quantization, meeting the requirements of real-time inference.

[0152] In this embodiment, the detailed steps for establishing the training data set of the interactive perception model are as follows: first, collect multi-scenario user interaction behavior data, including browsing trajectory data, click stream data, dwell time data, interactive operation data and conversion event data for at least 6 months, with a data volume of no less than 1 million interaction records and covering no less than 100,000 users; then clean the original data, remove invalid records, outliers and duplicate data, and retain more than 90% of valid data after cleaning; then extract individual user characteristics, including demographic characteristics, behavioral habit characteristics, consumption capacity characteristics, etc., with a feature dimension of no less than 50 dimensions; then construct a user relationship network, and calculate based on indicators such as interaction co-occurrence, content overlap, and spatiotemporal consistency. The correlation strength between users is analyzed, and user pairs with correlation strength greater than the threshold of 0.3 are screened to form relationship edges; then, positive and negative sample sets are constructed according to the interaction results. Positive samples are successful equity conversion cases and corresponding pre-interaction sequences, and negative samples are failed equity push cases and corresponding pre-interaction sequences. The ratio of positive and negative samples is controlled at 1:3; then, the dataset is divided into training set, validation set and test set in chronological order, with ratios of 70%, 15% and 15%; then, the training data is enhanced, including sequence truncation, noise addition, feature masking and other methods to increase sample diversity; finally, a multimodal training sample library containing sequence features, graph structure features and label information is generated. The sample format is structured tensor, which is convenient for batch training of the model. The training process adopts a hierarchical learning rate strategy, setting different learning rates for different network layers. The learning rate of the underlying feature extraction layer is relatively low, at 0.0005, and the learning rate of the high-level interaction layer is relatively high, at 0.001. An early stopping strategy is also implemented, and training is stopped if there is no improvement in the performance of the validation set for five consecutive rounds. The performance indicators of the validation set are periodically evaluated, including an individual conversion rate prediction accuracy target of no less than 85% and a group response consistency target of no less than 80%. Finally, the model parameters with the best overall performance on the validation set are selected as the pre-trained model for online deployment.

[0153] In order to better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: A research and development team has developed an interactive writing application that integrates psychological analysis and health services. The application appears to be for practicing calligraphy, but in fact it infers the user's psychological state and living habits by analyzing the user's behavioral patterns during the writing process, and then configures personalized health product rights. The team selected 50,000 active users for a three-month test. The users' ages ranged from 18 to 65 years old, covering students, working people, housewives and retirees. The system records the user's multi-dimensional behavioral data such as stroke characteristics, pause duration, writing speed, modification frequency, completion, and font style preferences during the writing process. These data have been proven to be highly correlated with the user's stress level, concentration state, emotional fluctuations and decision-making style. Each user completed at least 100 writing exercises, with each exercise lasting 5 to 15 minutes. After cleaning the raw behavioral data collected by the system, step S01 is applied to construct a 128-dimensional user behavior feature vector. Different dimensions in the vector represent different psychological state characteristics, such as dimensions 1-20 representing stress level, dimensions 21-40 representing concentration ability, dimensions 41-60 representing emotional stability, dimensions 61-80 representing decision-making style, dimensions 81-100 representing daily routine, and dimensions 101-128 representing social preferences. Table 1 shows some key features extracted from user writing behavior data and their psychological health implications:

[0154] Table 1 Correspondence between writing behavior characteristics and mental health

[0155]

[0156] Applying step S02, the system combines the user's geographic location data, usage time period data, and device information data to calculate the association status score between users through the interactive state neural network. In actual applications, the association status scores between users with similar work and life rhythms and similar mental health conditions are generally higher, with an average score of 72.8, while the average association score between random user pairs is 25.3. Based on these scores, the system constructs a user relationship graph containing 50,000 nodes and approximately 350,000 edges, and calculates a 32-dimensional association vector for peer user groups. Table 2 shows the average values ​​of the association vector components for user groups with different mental states:

[0157] Table 2 Main components of correlation vectors of user groups with different psychological states

[0158]

[0159] In step S03, the system uses a minimum spanning tree algorithm to identify key influence nodes within each user group. In the high-stress group, key influence nodes are primarily users with a strong sense of health and who actively seek improvement. In the mood-fluctuating group, key influence nodes are primarily users who are socially active and willing to share their experiences. The system calculates an influence index for each user, revealing that approximately 8.3% of users have an influence index exceeding 75, indicating that these users play a role in leading healthy lifestyles within their respective groups. In steps S04 and S05, the system weights user behavioral feature data and establishes an intent strength scoring system. Based on users' writing patterns, the system infers the level of demand for different health products, including stress-relieving products, sleep-enhancing products, attention-enhancing products, and mood-regulating products. The intent strength matrix shows that the high-stress group's average intent score for stress-relieving products is 8.9, and for sleep-regulating products is 7.8. In contrast, the mood-fluctuating group's average intent score for mood-regulating products is 9.2, and for nutritional supplements is 6.5. Applying steps S06 and S07, the system generates personalized health product benefit recommendation lists for different users based on the user intention strength matrix and influence index, and implements a multi-objective optimization allocation strategy. Table 3 shows a comparison of benefit allocation effects for users with different psychological states:

[0160] Table 3 Effects of health product rights placement on users with different psychological states

[0161]

[0162] Compared with the traditional health product recommendation method based on direct expression of user needs and historical purchase analysis, the present invention has achieved significant improvement through the indirect method of analyzing psychological state through writing behavior. Traditional methods mainly rely on users' explicit searches, browsing records and questionnaires. Users are often unwilling or unable to accurately express their own mental health needs, resulting in limited recommendation accuracy, an average equity conversion rate of 31.5%, and a group influence diffusion rate of only 18.3%. However, the present invention uses the non-invasive method of analyzing writing behavior, combined with user relationship graphs and key influence node identification, to not only accurately capture users' potential psychological needs, but also fully consider the influence and propagation characteristics of user groups. The average equity conversion rate is increased to 47.9%, the group influence diffusion rate is increased to 38.3%, and the user health improvement feedback rate is as high as 64.9%. The present invention avoids the limitations of user self-reporting in traditional methods, and achieves a deep understanding of users' health needs through behavioral analysis. While meeting the needs of individual users, it effectively activates the group dissemination of healthy living concepts and realizes the accurate and efficient allocation of health product rights and interests resources.

[0163] Example 3: A research and development team applied the method of the present invention to configure equity products based on the interactive behavior of users in group chats. The team selected 2,000 active shopping exchange groups on the platform, with a total of about 150,000 users. The group sizes ranged from small interest groups of 20 people to large category groups of 5,000 people. The system collects behavioral data such as message sending, content interaction, product sharing, and evaluation feedback in group chats in real time through the API interface, and combines the user's browsing history, search history, and purchase behavior to construct a multi-dimensional user behavior feature vector. The data collection period is 6 months, and each user generates an average of about 800 interaction records. Table 4 shows the main group chat behavior characteristics collected by the system:

[0164] Table 4 Group chat interaction behavior characteristics

[0165]

[0166] Applying step S01, the system constructs a 128-dimensional behavioral feature vector based on the user's group chat behavior sequence, where each dimension represents the characteristics of different interaction types and interaction objects. Applying step S02, the system combines user geographic location data, active time data, and device information data to calculate the association status score between users through the interaction status neural network. Experimental data show that the average association status score between users who frequently interact in the same group chat is 83.2, while the average association score of random users across groups is 12.7. Based on these scores, the system constructs a user relationship graph and calculates the association vector of peer user groups. Table 5 shows the comparison of the main components of the group association vectors of different types of shopping groups:

[0167] Table 5 Comparison of correlation vector components of different types of shopping groups

[0168]

[0169] In step S03, the system uses a minimum spanning tree algorithm to analyze group structure, identify key influence nodes, and calculate user influence indexes. Results show that in the maternal and infant product group, active and experienced mothers are often key influence nodes, with influence indexes generally exceeding 85. In the digital product group, professional review contributors are key influence nodes, with an average influence index of 79.3. In steps S04 and S05, the system weights user behavioral feature data and establishes an intent strength scoring system. Experiments have found a strong correlation between user content in group chats and purchase intent. For example, users who frequently inquire about their experience with a particular product category typically score their intent for that category above 7.5. In step S06, the system calculates the similarity between the association vectors of each group type and the standard association vector and determines the input weight coefficients using a similarity weighting function. The similarity for the maternal and infant product group is 0.87, corresponding to a weight coefficient of 1.38; the similarity for the fresh food group is 0.62, corresponding to a weight coefficient of 0.95. The system generates a list of recommended benefits based on the weighted intention intensity matrix and user influence index, and dynamically adjusts it based on real-time user responses. Table 6 shows some examples of benefit recommendations generated by the system:

[0170] Table 6 Example of rights recommendation based on group chat behavior

[0171]

[0172] Applying step S07, the system executes a multi-objective optimization allocation strategy, comprehensively considering individual conversion probabilities and group influence to determine the final equity investment plan. During actual implementation, the system sets different weights for each group type. For example, the maternity and baby group and the beauty group have a higher group influence weight (0.5), while the digital group has a higher individual conversion weight (0.7). After a three-month experiment, comparative results show that the equity investment plan generated by the present method has achieved significant results across all groups. Table 7 shows the comparison of the results with the traditional method:

[0173] Table 7 Comparison of equity investment effects

[0174]

[0175] Traditional methods for allocating equity products primarily rely on individual user behavioral data and simple demographic characteristics to make recommendations, ignoring user interactions and influence differences within social networks. This approach has significant shortcomings in group chat scenarios, making it difficult to identify and leverage the influence of opinion leaders and information dissemination pathways within groups. The proposed method effectively captures the complex user interaction patterns and opinion dissemination mechanisms within group chat environments by constructing a user relationship graph, identifying key influence nodes, and calculating correlation vectors for peer user groups. Experimental results show that the proposed method significantly improves equity utilization, conversion rate, group activity, and return on investment (ROI) compared to traditional methods, with average increases of 58.7%, 74.3%, 84.6%, and 63.2%, respectively. This method not only improves the efficiency of equity product allocation but also amplifies the group dissemination effect of equity investment by accurately identifying and incentivizing key influencers, providing a powerful tool for precision marketing on social e-commerce platforms.

[0176] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 8, 9, 10 and 11 below.

[0177] Table 8 Variable Explanation Table (Part I)

[0178]

[0179] Table 9 Variable Explanation Table (Part II)

[0180]

[0181] Table 10 Variable Explanation Table (Part 3)

[0182]

[0183] Table 11 Variable Explanation Table (Part 4)

[0184]

[0185] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A method for configuring equity products based on real-time user interaction information, characterized in that: include: Construct a multi-dimensional user behavior feature vector; input user behavior feature data into the interactive state neural network to generate user association status scores, build a user relationship graph, and calculate the association vector of peer user groups; input the user relationship graph into the minimum spanning tree algorithm to perform structured analysis of peer user groups, identify key influence nodes, calculate the equity propagation path within the group, and output the user influence index; A user intention recognition module is introduced to assign differentiated weights to user behavior feature data to form weighted behavior feature data. An intention strength scoring system is established to input the weighted behavior feature data into a multi-source data fusion algorithm to calculate the user's intention scores for different product categories and generate an intention strength matrix. Calculate the similarity between the peer user group association vector and the standard association vector, determine the input weight coefficient through the similarity weight function, multiply the intention intensity matrix and the user influence index by the input weight coefficient respectively, and input them into the pre-trained interaction perception model to generate a rights recommendation list, and dynamically adjust the recommendation priority of the products in the rights recommendation list according to the real-time response data of the user; execute a multi-objective optimization allocation strategy, take the rights recommendation list and the user influence index as input, comprehensively consider the individual conversion probability value and the group influence value, and determine the final rights delivery plan; the peer user group association vector refers to a multidimensional feature vector that characterizes the association pattern between users, including a geographical distance component, a time synchronization component, an interaction frequency component, an interest overlap component, and a social relationship strength component. The vector dimension is 32, and each component ranges from 0 to 1; the minimum spanning tree algorithm refers to an algorithm for finding a connected subgraph with the minimum total weight in the user relationship graph, which is used to identify the most cost-effective rights propagation path in the user group. The key influence node is the node with the highest degree centrality in the minimum spanning tree; the user influence index is calculated by the weighted sum of the weights of the edges connecting the key influence node in the minimum spanning tree; The standard association vector refers to an ideal association state vector derived from the association patterns of historically high-conversion-rate user groups. It serves as a reference for evaluating the association state of actual user groups, and its dimensions are consistent with the association vector of peer user groups. The similarity weight function maps the cosine similarity between the peer user group association vector and the standard association vector to an input weight coefficient. The interaction perception model constructs training samples by collecting user interaction behavior data from multiple scenarios, and after training, it is used to generate a list of recommended benefits. The step of constructing a multi-dimensional user behavior feature vector is to obtain user behavior feature data by capturing user interaction data in real time and combining it with historical interaction records to form a behavior sequence; Among them, the step of inputting user behavior feature data into the interactive state neural network specifically combines user geographic location data, timestamp data, and device information data to generate a user association state score, constructs a user relationship graph based on the user association state score, and calculates the association vector of the peer user group.

2. The method according to claim 1, characterized in that The interactive state neural network refers to a feedforward neural network structure used to analyze the interactive associations between users. The input layer receives user behavior feature data, geographic location data, timestamp data, and device information data. The hidden layer uses a fully connected structure for feature extraction, and the output layer generates a user association state score that represents the strength of user association.

3. The method according to claim 2, characterized in that The interaction state neural network is trained using the cosine similarity loss function.

Images