Method and system for predicting potential users of a service system based on interaction structure

Abstract

The present disclosure provides an interaction structure-based service system condition potential user prediction method and system, relates to the technical field of service systems and electronic commerce data mining, and comprises the following steps: acquiring user behavior data and constructing a user interaction behavior graph; based on the user interaction behavior graph, selecting a high-popularity service node in the service system, creating an interaction structure between the high-popularity service node and other service nodes and user nodes, forming an interaction structure optimization candidate set; inputting the interaction structure optimization candidate set into a proxy model to perform high-confidence interaction structure optimization, constructing a high-confidence interaction structure optimization set, and updating the user interaction behavior graph according to the high-confidence interaction structure optimization set; and inputting the updated user interaction behavior graph into a gold standard classifier to predict the service system condition potential user under a given interaction structure optimization mode.

Description

Technical Field

[0001] This disclosure relates to the fields of service systems and e-commerce data mining technology, specifically to a method and system for predicting potential users in service systems based on interactive structures. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] User prediction plays a crucial role in service systems, profoundly impacting their efficiency and healthy development. Recent advances in deep learning have significantly improved the accuracy and efficiency of user prediction models. However, traditional methods, relying solely on past user-generated data to identify potential users, are no longer sufficient to meet the growing demands of large-scale service systems such as e-commerce platforms. For instance, in real-world e-commerce promotional scenarios, certain strategies, such as offering coupons or product samples for trial, have been observed to stimulate interest among users initially categorized as non-potential. These users can subsequently be converted into potential customers, known as conditional potential users. Therefore, distinguishing these conditional potential users from non-potential users becomes a pressing challenge; however, existing user mining methods suffer from several problems in identifying conditional potential users.

[0004] First, current potential user prediction methods fail to identify conditional potential users. While they can distinguish between potential and non-potential users, they lack the ability to identify conditional potential users. For example, in an e-commerce scenario, suppose an e-commerce platform uses interaction structure optimization methods, excluding users who previously had no interaction with the target T-shirt from the potential user list. However, by introducing a proxy model, the system can identify these initially uninterested users and perform detailed analysis. After receiving a bundle of promotional offers from the seller, the proxy model discovers that these users' interaction structure has changed, showing a renewed interest in the target T-shirt. Therefore, these users are reclassified as potential users. Although service systems have a significant need to predict conditional potential users, the concept of conditional potential users is currently undefined, and methods for predicting them are clearly lacking. Therefore, defining and identifying conditional potential users based on these premises is crucial.

[0005] Furthermore, the lack of interpretability is another key challenge for user prediction models. Recent research has turned to advanced deep learning models to improve prediction efficiency. However, these deep learning models are often complex and lack interpretability, making it challenging to understand the reasoning behind their predictions. The inability to provide a clear and concise explanation for predictions can hinder practical application and potentially lead to significant business losses. These challenges are particularly critical when predicting conditional latent users, as this requires not only predicting these users but also developing corresponding optimization strategies. Without interpretability in user prediction models, understanding the optimization mechanisms driving conditional latent users becomes elusive, ultimately leading to prediction failures. Therefore, improving the interpretability of user prediction methods, especially in the domain of conditional latent users, and providing a clear and coherent explanation of the factors influencing predictions, is crucial for accurately identifying the corresponding interaction-conditional optimization mechanisms. Summary of the Invention

[0006] To address the aforementioned issues, this disclosure proposes a method and system for predicting conditional potential users in service systems based on interaction structures. Conditional potential users are defined as individuals initially classified as non-potential users but who have the potential to become active potential users after optimizing their interactions with various services within the service system. A Conditional Potential User Mining (CPUM) framework is constructed, integrating a gold standard classifier and a proxy model to more accurately identify conditional potential users for a specific type of service within the service system and their interaction structure optimization methods, thereby improving the accuracy and interpretability of user behavior prediction.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] Methods for predicting conditional potential users in service systems based on interaction structures include:

[0009] Acquire user behavior data and construct user interaction behavior graphs;

[0010] Based on the user interaction behavior graph, select the service nodes with high popularity in the service system, create the interaction structure between the service nodes and user nodes, and form a candidate set for interaction structure optimization.

[0011] The candidate set of interaction structure optimization is input into the proxy model for high-confidence interaction structure optimization, a high-confidence interaction structure optimization set is constructed, and the user interaction behavior graph is updated according to the high-confidence interaction structure optimization set.

[0012] The updated user interaction behavior graph is input into the gold standard classifier to predict potential users of the service system under a given interaction structure optimization method.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A service system based on an interaction structure, including a conditional potential user prediction system, comprising:

[0015] The data acquisition module is used to acquire user behavior data and construct user interaction behavior graphs;

[0016] The interaction structure construction module is used to select highly popular service nodes in the service system based on the user interaction behavior graph, create interaction structures between other service nodes and user nodes, and form a candidate set for interaction structure optimization.

[0017] The interaction structure optimization module is used to input the interaction structure optimization candidate set into the proxy model for high-confidence interaction structure optimization, construct a high-confidence interaction structure optimization set, and update the user interaction behavior graph based on the high-confidence interaction structure optimization set.

[0018] The prediction module is used to input the updated user interaction behavior graph into the gold standard classifier to predict potential users under the service system conditions given the interaction structure optimization method.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned service system conditional potential user prediction method based on an interactive structure.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the conditional potential user prediction method for the service system based on the interactive structure.

[0023] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0024] This disclosure presents a conditional potential user prediction method for service systems based on interaction structure. Conditional potential users are defined as individuals initially classified as non-potential users but who, after optimizing their interactions with various services within the service system, have the potential to become active potential users. Traditional user prediction models typically only distinguish between potential and non-potential users, neglecting user groups that may become potential users after changes in the interaction structure. The Conditional Potential User Prediction (CPUM) framework of this disclosure effectively identifies and predicts these conditional potential users by integrating a gold-standard classifier and a surrogate model. The CPUM framework not only distinguishes the current state of users but also predicts which individuals who were not originally potential users may become potential users after changes in certain interaction conditions. This significantly improves the flexibility and application scope of user prediction models, providing service systems with more comprehensive user insights.

[0025] This disclosure presents a method for predicting potential users in a service system based on an interaction structure. Traditional deep learning user prediction models often lack interpretability, making it difficult for marketing decision-makers to understand the logic behind the predictions. This disclosure provides a clear explanation for user classification decisions by incorporating an attention-based agent model. This improvement enables marketers and decision-makers not only to understand users' potential interests but also to comprehend the specific interactive factors influencing changes in user interests, such as promotional activities and product features. This enhanced interpretability is crucial for developing precise marketing strategies and improving user satisfaction.

[0026] This disclosure addresses the limitations of traditional user prediction models by introducing innovative methods, particularly achieving significant progress in identifying conditional potential users and improving prediction interpretability. These improvements not only enhance the accuracy of user predictions but also provide e-commerce platforms with deeper user insights, thereby helping businesses develop more effective marketing strategies and improve user experience. Attached Figure Description

[0027] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0028] Figure 1 The training process of the Conditional Potential User Prediction (CPUM) framework in this embodiment of the disclosure;

[0029] Figure 2 This is a flowchart illustrating the potential user prediction process under certain conditions, as described in this disclosure. Detailed Implementation

[0030] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0032] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0033] Example 1

[0034] This disclosure presents a method for conditional potential user mining in a service system based on an interaction structure, comprising:

[0035] Step 1: Obtain user behavior data and construct a user interaction behavior graph;

[0036] Step 2: Based on the user interaction behavior graph, select high-popularity service nodes in the service system, create interaction structures with other service nodes and user nodes, and form a candidate set for interaction structure optimization;

[0037] Step 3: Input the candidate set of interaction structure optimization into the proxy model to perform high-confidence interaction structure optimization, construct a high-confidence interaction structure optimization set, and update the user interaction behavior graph according to the high-confidence interaction structure optimization set;

[0038] Step 4: Input the updated user interaction behavior graph into the gold standard classifier to predict potential users under the service system conditions given the interaction structure optimization method.

[0039] As one embodiment, the specific implementation process of the service system conditional potential user mining method based on interaction structure disclosed herein is as follows:

[0040] Step 1: Construct a user interaction behavior graph based on the user's interaction behavior with the services provided by the service system. Including user node o u Service Node o p and the edges e representing their interactions up ;

[0041] Specifically, constructing user interaction behavior graphs The process includes:

[0042] (a) Determine the user node o u Service node op The original data is in the form of a multidimensional dataset, containing user-service interaction information such as clicks, likes, and ratings. For each user behavior data point, key features such as user ID, service ID, and interaction type are extracted. These features are used to construct a user interaction behavior graph, where users and services are nodes, and the interactions between users and services are edges. Each node contains attribute information about the user or service, and interactions include information such as the user's historical service records or information on joint services provided by different services.

[0043] (b) Create edges based on user interactions with services (such as purchasing, browsing, experiencing, etc.). up ;

[0044] (c) Initialize the graph using user historical interaction data and service information, and extract o using user historical interaction data and service information. p o u and e up Construct graph data to ensure that the nodes and edges in the graph accurately reflect user behavior and service characteristics.

[0045] In this system, users and service objects are connected by edges, with the edge weights representing the interaction strength. Noise and inconsistencies in the original data are cleaned and normalized to improve data quality.

[0046] Step 2: Construct a candidate set for interaction structure optimization based on the service popularity calculator ξ provided in the service system. t By selecting the service node with the highest popularity And create relationship edges between it and other services and user nodes in the user interaction behavior graph;

[0047] Specifically, based on the popularity of services, a candidate set ζ for optimizing user interaction structure is constructed. t The process includes:

[0048] (a) Selecting highly popular service nodes in the service system Based on historical data and statistical results, high-prevalence service nodes in the service system were selected.

[0049] (b) Create highly popular service nodes Interactions with other service nodes and user nodes; specifically, through building highly popular service nodes on the graph. Edges between nodes represent service nodes with high popularity. Interactions with other service nodes and user nodes;

[0050] (c) Form an interaction structure optimization candidate set ζ based on the created interaction structure. tIt contains all the interactive structures that may inspire users to engage with the services they are offered.

[0051] Step 3: Introduce an agent model based on attention mechanism Identify key user interaction structures;

[0052] The candidate set for optimizing the interaction structure is input into the proxy model. The process involves optimizing high-confidence interaction structures, constructing a set of optimized high-confidence interaction structures, and updating the user interaction behavior graph based on the set of optimized high-confidence interaction structures.

[0053] Specifically, constructing a high-confidence interaction structure optimization set includes:

[0054] (a) Using the proxy model Calculate the confidence score w for each interaction structure candidate. set Using the proxy model Input a graph and a candidate set of interaction structures, and output the confidence score w for each candidate interaction structure. set ;

[0055] (b) From w set Select the interaction structure with the highest confidence level to form a high-confidence interaction structure optimization set.

[0056]

[0057] (c) Optimize the set using a high-confidence interaction structure Add it to the user interaction behavior graph In the simulation service system, the optimized user interaction diagram is shown. For further analysis.

[0058] Furthermore, the proxy model It is a neural network structure based on the attention mechanism, containing two graph attention layers, used to explain the gold standard classifier. The model's prediction results are obtained by analyzing the gold standard classifier. The model's decision-making process provides an explanation of the importance of each interaction (such as clicks and experiences). Utilizing an attention mechanism, the model learns from extensive historical data which interactions are most important for user classification and generates importance weights to improve interpretability. During training, the model minimizes the differences between the interaction and the user's experience. The parameters are optimized based on the predicted discrepancies.

[0059] During training, the following loss function is defined to optimize the model so that the surrogate model fits the prediction results of the gold standard classifier, which can provide an important explanation for different user interaction behaviors.

[0060]

[0061] The surrogate model is trained iteratively until the predetermined performance metrics are met.

[0062] Step 4: Use Graph Convolutional Network (GCN) as the gold standard classifier Input the updated user interaction behavior graph Predicting user preferences for target category services The underlying intention;

[0063] Gold Standard Classifier A graph convolutional network (GCN) was chosen as the infrastructure to process user interaction behavior graphs. A multi-layer graph convolutional structure was designed to capture the complex interaction relationships between users and services. The GCN was trained to learn user interactions with specific service categories. Potential use cases; optimizing GCN performance using cross-validation and hyperparameter tuning.

[0064] Gold Standard Classifier By learning the structure and node features of user interaction behavior graphs, the model predicts users' potential intentions. The model's input is the node features of the user interaction behavior graph, and the output is the probability that each user of a service system will become a potential user of a certain type of service. By optimizing the cross-entropy loss function, the model can effectively learn from large-scale user behavior data and make accurate predictions.

[0065] After selecting high-confidence interaction structure optimization strategies, the gold standard classifier Using the updated user interaction behavior graph As input, predict the user's potential intent towards the target type of service; record this in a high-confidence interaction structure. User u i And related optimization methods; through this process, we can accurately identify potential users under the conditions of optimizing specific interaction structures in the service system.

[0066] This process includes applying the optimized interaction structure to the user interaction behavior graph. and utilize The updated graph is analyzed. If a user is predicted to be a potential user under specific interaction structure optimizations, this information will be recorded for further user behavior analysis and marketing strategy development.

[0067] Example 2

[0068] One embodiment of this disclosure provides a method for conditional potential user mining in a service system based on an interaction structure. This embodiment takes an e-commerce system as an example, where the service provided by the system is goods. The specific implementation process is as follows:

[0069] (1) Construction and preprocessing of user interaction behavior graphs;

[0070] The process of constructing a user behavior graph is based on the interaction data between users (u) and products (o). The specific steps are as follows: extract the interaction history between users and products, such as clicks or purchase behaviors, and construct a user behavior graph. In the diagram, user u and product o are connected by edge e. u,o Connected edges, with weight w u,o This represents the intensity of interaction. Noise and inconsistencies in the original data are cleaned and normalized to improve data quality.

[0071] (2) Construction and training of the gold standard classifier;

[0072] Gold Standard Classifier This is a deep learning model used to predict users' purchase intentions for products. The training process is as follows: A suitable neural network architecture is selected, and the network parameters are initialized. A large amount of user behavior data {(u,o,y} is used. u,o Train a classifier, where y u,o This represents the purchase intent label of user u for product o. Optimize network parameters to improve the prediction accuracy of the classifier.

[0073] (3) Construction and training of the proxy model;

[0074] Proxy Model This is an attention-based explanatory model used to explain the decision-making process of the gold standard classifier. The training process includes: defining a simple network structure and initializing the model parameters. Using the same user behavior data and the output of the gold standard classifier as the training objective, the surrogate model is trained by minimizing the prediction error.

[0075] (4) Construction of the candidate set for interaction structure optimization;

[0076] Interaction structure optimization candidate set ζ t It includes all the interactive structures that might potentially motivate users to engage with the service. The construction process includes: first, selecting highly popular service nodes within the service system. Then create Interactions with other service nodes and user nodes create an interaction structure that forms a candidate set ζ. t .

[0077] (5) Screening of high-confidence interaction structure optimization methods;

[0078] The screening process involves analyzing the impact of each optimization method on user purchase intent. Specific steps include: inputting the optimization methods and graph data into the proxy model. Output the score(e) of each optimization method's impact on user purchase intent. From the candidate set ζ t Select the optimization method with the highest score to form a high-confidence interaction structure optimization set.

[0079] (6) Prediction of potential users based on conditions;

[0080] Using the gold standard classifier and proxy model Predict users who are likely to become potential users under a given optimization method. The prediction process includes: inputting combinations of user u and item o and their interaction data, as well as a high-confidence interaction structure optimization set. The algorithm is adapted to a gold-standard classifier and a proxy model. The output is the probability that user u becomes a potential user under simulated optimization.

[0081] Example 3

[0082] One embodiment of this disclosure provides a service system conditional potential user mining system based on an interaction structure, comprising:

[0083] The data acquisition module is used to acquire user behavior data and construct user interaction behavior graphs;

[0084] The interaction structure construction module is used to select highly popular service nodes in the service system based on the user interaction behavior graph, create interaction structures between other service nodes and user nodes, and form a candidate set for interaction structure optimization.

[0085] The interaction structure optimization module is used to input the interaction structure optimization candidate set into the proxy model for high-confidence interaction structure optimization, construct a high-confidence interaction structure optimization set, and update the user interaction behavior graph based on the high-confidence interaction structure optimization set.

[0086] The prediction module is used to input the updated user interaction behavior graph into the gold standard classifier to predict potential users under the service system conditions given the interaction structure optimization method.

[0087] Example 4

[0088] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the conditional potential user mining method for a service system based on an interactive structure.

[0089] Example 5

[0090] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the method for mining potential users based on the interactive structure of the service system.

[0091] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for predicting potential users in a service system based on an interaction structure, characterized in that, include: Acquire user behavior data and construct a user interaction behavior graph; user interactions with services include purchasing, browsing, and experience. Based on the user interaction behavior graph, select the service nodes with high popularity in the service system, create the interaction structure between the service nodes and user nodes, and form a candidate set for interaction structure optimization. The candidate set of interaction structure optimization is input into the proxy model for high-confidence interaction structure optimization, a high-confidence interaction structure optimization set is constructed, and the user interaction behavior graph is updated according to the high-confidence interaction structure optimization set. The proxy model is a neural network structure built on an attention mechanism. During training, the decision-making process of the gold standard classifier is analyzed to interpret the importance of each interaction structure, calculate the confidence score of the candidate set for each interaction structure optimization, and identify the interaction structure that is most important to user classification. The updated user interaction behavior graph is input into the gold standard classifier to predict potential users of the service system under a given interaction structure optimization method. The gold standard classifier is a multi-layer neural network structure built on graph convolutional networks. It predicts the user's potential intention by recognizing the structure and node features in the user interaction behavior graph. Its input is the node features of the user interaction behavior graph, and its output is the probability that each service system user will become a potential user of a specific service.

2. The method for predicting potential users in a service system based on an interactive structure as described in claim 1, characterized in that, Constructing a user interaction behavior graph includes: obtaining the interaction behavior between the user and the services provided by the service system, determining user nodes and service nodes, creating edges based on the interaction behavior between the user and the service, constructing the user interaction behavior graph, and initializing the graph using historical user interaction data and service information.

3. The method for predicting potential users in a service system based on an interactive structure as described in claim 2, characterized in that, When the service provided is a product, in the user interaction behavior graph, the user and the product are connected by edges, and the weight of the edge represents the interaction intensity.

4. The method for predicting potential users in a service system based on an interactive structure as described in claim 1, characterized in that, Constructing an interaction structure optimization candidate set includes: selecting highly popular service nodes in the service system based on service popularity, creating interactions between highly popular service nodes and other service nodes and user nodes, and forming an interaction structure optimization candidate set that includes all interaction structures that may stimulate users' potential intention to provide services.

5. The method for predicting potential users in a service system based on an interaction structure as described in claim 1, characterized in that, Constructing a high-confidence interaction structure optimization set includes: using a proxy model to calculate the confidence score of each interaction structure optimization candidate set; selecting the interaction structure with the highest confidence score based on the confidence score to form a high-confidence interaction structure optimization set; and using the high-confidence interaction structure optimization set to update the user interaction behavior graph and simulate the optimized user interaction behavior of the service system.

6. A service system conditional potential user prediction system based on an interactive structure, characterized in that, Specifically, the method for predicting potential users in a service system based on an interaction structure, as described in any one of claims 1-5, includes: The data acquisition module is used to acquire user behavior data and construct user interaction behavior graphs; The interaction structure construction module is used to select highly popular service nodes in the service system based on the user interaction behavior graph, create interaction structures between other service nodes and user nodes, and form a candidate set for interaction structure optimization. The interaction structure optimization module is used to input the interaction structure optimization candidate set into the proxy model for high-confidence interaction structure optimization, construct a high-confidence interaction structure optimization set, and update the user interaction behavior graph based on the high-confidence interaction structure optimization set. The prediction module is used to input the updated user interaction behavior graph into the gold standard classifier to predict potential users under the service system conditions given the interaction structure optimization method.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the service system conditional potential user prediction method based on an interactive structure as described in any one of claims 1-5.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform the service system conditional potential user prediction method based on the interactive structure as described in any one of claims 1-5.

Images