User demand analysis method and device, electronic equipment, medium and program product

Abstract

This application provides a user demand analysis method, apparatus, electronic device, medium, and program product, relating to the application of large-scale models in the fintech field, and applicable to the fields of big data technology and artificial intelligence technology. The method includes: obtaining query input from a target user; extracting a target user subgraph from a pre-constructed user demand knowledge graph based on the query input, wherein the user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer, and the nodes of the behavior layer include user behavior nodes; and performing demand analysis on the target user based on the target user subgraph, outputting the demand analysis results.

Description

Technical Field

[0001] This application relates to the application of large models in the field of financial technology, and to the fields of big data technology and artificial intelligence technology. More specifically, it relates to a user demand analysis method, apparatus, equipment, medium and program product. Background Technology

[0002] With the continuous evolution of big data and artificial intelligence technologies, accurately understanding user behavior and potential needs has become crucial for improving service quality and business efficiency. However, traditional demand mining methods, due to their limited data modalities and analytical dimensions, often remain at the level of surface-level behavioral identification, failing to delve into users' underlying emotions and motivations, thus affecting the accuracy and effectiveness of strategy formulation.

[0003] Specifically, existing multimodal analysis methods have failed to address the challenges of heterogeneity and fusion between modalities, lacking efficient cross-modal alignment and unified representation methods, resulting in an incomplete understanding of user behavior. Most knowledge graphs remain static, single-layered structures, lacking dynamic updates and multi-level causal representation capabilities, making it difficult to support the modeling of complex behaviors leading to demand chains. Summary of the Invention

[0004] In view of the above problems, this application provides user requirements analysis methods, apparatus, equipment, media and program products.

[0005] According to a first aspect of this application, a user requirements analysis method is provided, the method comprising: obtaining query input from a target user; extracting a target user subgraph from a pre-constructed user requirements knowledge graph based on the query input, wherein the user requirements knowledge graph includes a behavior layer, a requirements layer, and a causal layer, the nodes of the behavior layer include user behavior nodes, the edges of the behavior layer include temporal or logical dependencies between the user behavior nodes, the nodes of the requirements layer include user requirements tag nodes, the edges of the requirements layer include semantic relationships between the user requirements tag nodes, the causal layer is used to represent causal inference relationships between the user behavior nodes and the user requirements tag nodes, and the edges of the causal layer connect the user behavior nodes and the user requirements tag nodes based on target association weights; and performing user requirements analysis based on the target user subgraph, and outputting requirements analysis results.

[0006] According to an embodiment of this application, the step of extracting a target user subgraph from a pre-constructed user demand knowledge graph based on the query input includes: extracting behavior nodes and demand nodes corresponding to the target user from the behavior layer and the demand layer based on the query input; calculating the association weight between the behavior nodes and the demand nodes using a multi-head attention mechanism; obtaining the causal weights corresponding to the behavior nodes and the demand nodes in the causal layer; correcting the association weights based on the causal weights to obtain the target association weights; and obtaining the target user subgraph based on the behavior nodes, the demand nodes, and the target association weights.

[0007] According to an embodiment of this application, a user demand knowledge graph is pre-constructed, including: acquiring behavioral rule information and multi-source heterogeneous data based on user information; constructing a behavioral layer graph based on the multi-source heterogeneous data and the behavioral rule information; performing semantic analysis on the behavioral layer graph using a multimodal model to generate corresponding user demand tag nodes; constructing a demand layer graph based on the user demand tag nodes; and calculating the causal relationship strength between the user behavior nodes and the user demand tag nodes using a causal analysis model; and constructing a causal layer graph based on the user behavior nodes, the user demand tag nodes, and the causal relationship strength.

[0008] According to an embodiment of this application, pre-constructing a user demand knowledge graph includes: acquiring multi-source heterogeneous data based on user information; extracting features from the multi-source heterogeneous data to obtain a multimodal vector; aligning the features of the multimodal vector to obtain a multimodal embedding vector; and constructing the user demand knowledge graph based on the multimodal embedding vector.

[0009] According to an embodiment of this application, the method further includes: under a federated learning framework, multiple data providers perform model training operations based on local multimodal embedding vectors to generate local model gradients or update parameters; encrypt the local model gradients or update parameters and upload the encrypted results to a joint modeling server; and decrypt and aggregate the encrypted results based on the joint modeling server to generate global model parameters for constructing the user demand knowledge graph.

[0010] According to an embodiment of this application, the method further includes: performing incremental calculation of the causal relationship strength within a set time period based on the pairing relationship between the user behavior node and the user demand tag node; and performing a significance test on the corresponding behavioral demand causal path based on the result of the incremental calculation, retaining the behavioral demand causal path with a confidence level exceeding a target threshold to update the causal layer graph.

[0011] According to an embodiment of this application, the step of performing target user demand analysis based on the target user subgraph and outputting demand analysis results includes: sorting the behavioral demand causal paths in the target user subgraph according to the target association weights to obtain target paths; generating an interactive visualization graph based on the target paths, wherein the nodes and edges of the interactive visualization graph are displayed using color mapping based on the target association weights; and / or generating a natural language report using a language model based on the target paths and the target association weights.

[0012] According to an embodiment of this application, the method further includes: generating an intervention strategy for the target user based on the demand analysis results, executing the intervention strategy on the target user side; obtaining behavioral feedback data corresponding to the intervention strategy, uploading the behavioral feedback data as an intervention effect indicator to the federated learning framework to the joint modeling server, generating intervention strategy parameters, and updating the global model parameters based on the intervention strategy parameters.

[0013] According to an embodiment of this application, the method further includes: constructing a comparison model of the intervention group and the control group on the joint modeling server; calculating the average treatment effect of the intervention group and the control group based on the comparison model; and if the average treatment effect meets a preset condition, writing the updated global model parameters and the intervention strategy parameters into the user demand knowledge graph.

[0014] A second aspect of this application provides a user demand analysis apparatus, the apparatus comprising: a data acquisition module, configured to: acquire query input from a target user; a target user subgraph acquisition module, configured to: extract a target user subgraph from a pre-constructed user demand knowledge graph based on the query input, wherein the user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer, the nodes of the behavior layer include user behavior nodes, the edges of the behavior layer include temporal or logical dependencies between the user behavior nodes, the nodes of the demand layer include user demand tag nodes, the edges of the demand layer include semantic relationships between the user demand tag nodes, the causal layer represents causal inference relationships between the user behavior nodes and the user demand tag nodes, and the edges of the causal layer connect the user behavior nodes and the user demand tag nodes based on target association weights; and a demand analysis module, configured to: perform demand analysis on the target user based on the target user subgraph and output demand analysis results.

[0015] According to an embodiment of this application, the target user subgraph acquisition module can also be used to extract the behavior nodes and demand nodes corresponding to the target user from the behavior layer and the demand layer based on the query input; calculate the association weight between the behavior nodes and the demand nodes using a multi-head attention mechanism; obtain the causal weights corresponding to the behavior nodes and the demand nodes in the causal layer; correct the association weights based on the causal weights to obtain the target association weights; and obtain the target user subgraph based on the behavior nodes, the demand nodes, and the target association weights.

[0016] According to embodiments of this application, the requirements analysis module can also be used to sort the causal paths of behavioral requirements in the target user subgraph according to the target association weights to obtain target paths; and to generate an interactive visualization graph based on the target paths, wherein the nodes and edges of the interactive visualization graph are displayed using color mapping based on the target association weights; and / or to generate a natural language report using a language model based on the target paths and the target association weights.

[0017] According to embodiments of this application, the user demand analysis device may further include a graph construction module. The graph construction module can be used to acquire behavioral rule information and multi-source heterogeneous data based on user information; construct a behavioral layer graph based on the multi-source heterogeneous data and the behavioral rule information; perform semantic analysis on the behavioral layer graph using a multimodal model to generate corresponding user demand tag nodes; construct a demand layer graph based on the user demand tag nodes; and calculate the causal relationship strength between the user behavioral nodes and the user demand tag nodes using a causal analysis model; construct a causal layer graph based on the user behavioral nodes, the user demand tag nodes, and the causal relationship strength.

[0018] According to embodiments of this application, the graph construction module can also be used to acquire multi-source heterogeneous data based on user information, extract features from the multi-source heterogeneous data to obtain multimodal vectors, perform feature alignment on the multimodal vectors to obtain multimodal embedding vectors, and construct the user demand knowledge graph based on the multimodal embedding vectors.

[0019] According to embodiments of this application, the graph construction module can also be used in a federated learning framework to have multiple data providers perform model training operations based on local multimodal embedding vectors to generate local model gradients or update parameters; encrypt the local model gradients or update parameters and upload the encrypted results to a joint modeling server; and decrypt and aggregate the encrypted results based on the joint modeling server to generate global model parameters for constructing the user demand knowledge graph.

[0020] According to an embodiment of this application, the graph construction module can also be used to perform incremental calculation of the causal relationship strength within a set time period based on the pairing relationship between the user behavior node and the user demand tag node; and based on the result of the incremental calculation, perform a significance test on the corresponding behavioral demand causal path, and retain the behavioral demand causal path with a confidence level exceeding a target threshold to update the causal layer graph.

[0021] According to embodiments of this application, the user needs analysis device may further include an intervention module. The intervention module can be used to generate an intervention strategy for the target user based on the needs analysis results, execute the intervention strategy on the target user side; and acquire behavioral feedback data corresponding to the intervention strategy, upload the behavioral feedback data as an intervention effect indicator to the federated learning framework, generate intervention strategy parameters, and update the global model parameters based on the intervention strategy parameters.

[0022] According to an embodiment of this application, the intervention module can also be used to construct a comparison model between the intervention group and the control group on the joint modeling server; calculate the average treatment effect of the intervention group and the control group based on the comparison model; and if the average treatment effect meets a preset condition, write the updated global model parameters and the intervention strategy parameters into the user demand knowledge graph.

[0023] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0024] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0025] According to embodiments of this application, the layered design of the behavior layer, demand layer, and causal layer decouples low-level atomic operations, abstract demand semantics, and causal logic from the user, enabling the system to maintain event-level granularity of temporal accuracy while achieving intent aggregation and causal reasoning at a higher level, significantly improving the accuracy and interpretability of demand identification. Simultaneously, the query input-driven on-demand truncation mechanism for the target user subgraph instantly prunes massive global graphs into small-scale analytical subgraphs, reducing real-time computational overhead and ensuring the contextual integrity of personalized analysis. Furthermore, the integrated subgraph weights provide a unified and quantifiable metric basis, enabling the system to support real-time, interpretable demand diagnosis and strategy generation end-to-end, thereby technically achieving high throughput, high accuracy, and high interpretability in user demand analysis. Attached Figure Description

[0026] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0027] Figure 1 The illustrations depict application scenarios of user demand analysis methods, apparatus, devices, media, and program products according to embodiments of this application.

[0028] Figure 2 A flowchart illustrating a user requirements analysis method according to an embodiment of this application is shown schematically.

[0029] Figure 3 A flowchart illustrating a method for extracting a target user subgraph according to some exemplary embodiments of this application is shown schematically.

[0030] Figure 4 The flowchart illustrates a method for pre-constructing a user demand knowledge graph according to some exemplary embodiments of this application.

[0031] Figure 5 This schematically illustrates a structural block diagram of a user requirements analysis apparatus according to an embodiment of this application; and

[0032] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a user requirements analysis method according to an embodiment of this application. Detailed Implementation

[0033] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0034] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0035] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0036] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0037] First, the technical terms used in this article are explained and clarified as follows.

[0038] Bayesian networks: a probabilistic graphical modeling technique that analyzes historical data (such as the temporal correlation between customer service call sentiment and return rate) to calculate the strength of causal relationships and incrementally trains every hour to adjust weights, retaining only salient paths to ensure the timeliness and accuracy of the knowledge graph.

[0039] Federated learning framework: A distributed machine learning technique that supports joint modeling with cross-platform data privacy protection. Participants (such as e-commerce platforms and customer service systems) only share encrypted model gradient parameters; the original data remains within the domain, ensuring user privacy compliance.

[0040] Counterfactual analysis is an assessment technique based on causal inference. Its core idea is to generate a "not-occurred but possible" control scenario for a "fact" that has already occurred, in order to measure the true impact of a treatment (intervention) on the outcome variable.

[0041] Multi-source heterogeneity refers to data sets that come from different business systems, devices, or sensing media and are inconsistent in terms of data structure, modality, temporal granularity, and semantic level. These include structured logs, semi-structured text, unstructured audio and video, sensor streams, etc. The technical challenges are concentrated in cross-modal feature extraction, multi-dimensional temporal alignment, and consistent labeling.

[0042] Multi-head attention is the core operator of the Transformer architecture. Its idea is to compute attention scores in parallel in multiple learnable query-key-value subspaces, and then concatenate the outputs of each subspace before linear transformation to capture dependencies in different dimensions (position, semantics, context).

[0043] With the continuous development of big data and artificial intelligence technologies, the financial industry has entered a new era driven by data. In this environment, it is essential to effectively understand user needs and provide precise services. However, traditional demand mining methods often only scratch the surface of user behavior, failing to delve into potential needs and emotional changes, and often lacking explanations of the deeper motivations behind those needs, thus affecting the accurate implementation of strategies.

[0044] Although some existing technologies have begun to incorporate cutting-edge technologies such as artificial intelligence, federated learning, and knowledge graphs, there are still some technical defects and limitations.

[0045] First, current multimodal data analysis techniques often suffer from insufficient modality fusion. Specifically, traditional systems typically process data from each modality in isolation (such as behavioral logs, speech, and images), failing to effectively address the heterogeneity between modalities and lacking efficient cross-modal feature alignment and unified representation methods. This technical deficiency results in a poor comprehensive understanding of multimodal information, thereby affecting the more accurate identification and inference of user behavior and needs.

[0046] Secondly, current mainstream technologies are limited to static knowledge graphs and shallow reasoning, lacking multi-granular, multi-layered graph structures and dynamic update capabilities. Existing technologies typically employ single-dimensional static knowledge graphs (such as domain-specific knowledge graphs), failing to construct effective graph systems for complex, multi-layered relationships between behavior, needs, and causality, and lacking real-time updates to causal relationship weights. These technological limitations severely restrict the flexibility, interpretability, and real-time performance of knowledge graphs in real-world business scenarios.

[0047] Furthermore, current technologies generally lack effective causal reasoning and interpretability mechanisms, especially in causal relationship identification and dynamic modeling. They are unable to establish robust causal chains in response to dynamic changes in user behavior and needs, and it is even more difficult to achieve adaptive adjustment and iterative optimization of causal weights, which affects the transparency and credibility of the decision-making process.

[0048] Finally, while existing technologies have addressed the implementation of privacy protection frameworks for federated learning, they generally remain at the level of general privacy protection measures and lack customized federated learning mechanisms for multimodal financial data scenarios. Consequently, they cannot effectively address privacy, security, and efficiency issues in cross-platform data collaboration.

[0049] Based on this, embodiments of this application provide a user demand analysis method, comprising: obtaining query input from a target user; extracting a target user subgraph from a pre-constructed user demand knowledge graph based on the query input, wherein the user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer, the nodes of the behavior layer include user behavior nodes, the edges of the behavior layer include temporal or logical dependencies between the user behavior nodes, the nodes of the demand layer include user demand tag nodes, the edges of the demand layer include semantic relationships between the user demand tag nodes, the causal layer is used to represent causal inference relationships between the user behavior nodes and the user demand tag nodes, and the edges of the causal layer connect the user behavior nodes and the user demand tag nodes based on target association weights; and performing demand analysis on the target user subgraph and outputting demand analysis results. According to embodiments of this application, the layered design of the behavior layer, demand layer, and causal layer decouples low-level atomic operations, abstract demand semantics, and causal logic from the user, enabling the system to maintain event-level granularity of temporal accuracy while achieving intent aggregation and causal reasoning at a higher level, significantly improving the accuracy and interpretability of demand identification. Simultaneously, the query input-driven on-demand truncation mechanism for the target user subgraph instantly prunes massive global graphs into small-scale analytical subgraphs, reducing real-time computational overhead and ensuring the contextual integrity of personalized analysis. Furthermore, the integrated subgraph weights provide a unified and quantifiable metric basis, enabling the system to support real-time, interpretable demand diagnosis and strategy generation end-to-end, thereby technically achieving high throughput, high accuracy, and high interpretability in user demand analysis.

[0050] It should be noted that the user demand analysis methods, apparatus, devices, media, and program products defined in this application relate to the application of large-scale models in the fintech field. They can be used in the fields of big data technology and artificial intelligence technology, as well as in the fintech field and various other fields besides big data technology, artificial intelligence technology, and fintech. The application fields of the user demand analysis methods, apparatus, devices, media, and program products provided in the embodiments of this application are not limited.

[0051] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0052] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0053] Figure 1 The illustration shows an application scenario diagram of the user demand analysis method, apparatus, device, medium, and program product according to embodiments of this application.

[0054] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0055] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0056] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0057] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0058] It should be noted that the user demand analysis method provided in this application embodiment can generally be executed by server 105. Correspondingly, the user demand analysis device provided in this application embodiment can generally be located in server 105. The user demand analysis method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the user demand analysis device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0059] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0060] The following will be based on Figure 1 The described scene, through Figures 2-4 The user requirements analysis method of the disclosed embodiments is described in detail.

[0061] Figure 2 A flowchart illustrating a user requirements analysis method according to an embodiment of this application is shown.

[0062] like Figure 2 As shown, the user requirements analysis method 200 of this embodiment includes operations S210 to S230.

[0063] In operation S210, the query input of the target user is obtained.

[0064] In this embodiment, query input can be received in real time through a front-end interactive interface. The front-end interactive interface can include various forms such as desktop web pages, mobile applications, and intelligent customer service mini-programs. The system can pre-configure a unified query protocol for each form, supporting users to submit query requests using natural language, keyword combinations, or template-based dropdown options. The back-end can use a unified parsing module to perform semantic parsing and parameter extraction on the input content, generating structured query objects. Taking natural language queries as an example, a target user can input "view user behavior and potential needs of competitor pages over the past week." The parsing module will identify the time window (the past week), target scenario (competitor pages), and scope of needs (potential needs), and encapsulate it into a query object in a data exchange format before passing it to the next processing stage.

[0065] In some embodiments, the query input of the target user can also be obtained by calling an open interface from an enterprise-level business system. For example, in a bank customer service agent system, when an agent enters a customer's account number or drags a customer card to the "Requirements Analysis" plugin area during a conversation, the system sends a request to the requirements sourcing engine containing the user identifier, session context, and optional filtering conditions (such as transaction category and session start and end time). The backend gateway component first performs permission verification and signature verification on the request, and then maps the extracted parameters to a unified query object, maintaining interface consistency with the graph retrieval module and ensuring that calls from different systems do not disrupt the decoupling of the overall process.

[0066] In some embodiments, query input can also be automatically generated by a scheduled task triggered in the background for offline batch analysis or strategy inspection. When the scheduled task is triggered, user segmentation tags, time windows, and business scenarios can be dynamically combined according to preset rules to form query conditions, enabling periodic monitoring of the needs of key customer groups. For example, the system can automatically generate query input for "high-risk customer groups / behavior in the last 24 hours" at midnight every day, allowing the operations team to view the latest demand trends.

[0067] To accommodate multimodal interaction needs, voice query processing can also be supported. When a user submits a command via the voice interface, the voice recognition service can be called first to transcribe the audio into text, and then the same semantic parsing module can generate a query object. If the voice contains explicit emotional features (such as anger or anxiety), the parsing module can add an emotion tag field to the query object, so that subsequent subgraph extraction and causal inference can take into account the emotional dimension, further improving the accuracy of requirements analysis.

[0068] It should be noted that the system can attach tenant identifiers, data access levels, and audit information to all query inputs. If the query involves cross-regional data or restricted scenarios, the gateway can block or anonymize relevant fields according to permission policies; in this way, even if the S210 operation is triggered by a third-party system, it can be ensured that subsequent graph retrieval and requirements analysis are conducted only within the scope of compliance.

[0069] In operation S220, a target user subgraph is extracted from a pre-constructed user demand knowledge graph based on the query input. The user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer. The nodes of the behavior layer include user behavior nodes, and the edges of the behavior layer include the temporal sequence relationship or logical dependency relationship between the user behavior nodes. The nodes of the demand layer include user demand label nodes, and the edges of the demand layer include the semantic relationship between the user demand label nodes. The causal layer is used to represent the causal inference relationship between the user behavior nodes and the user demand label nodes, and the edges of the causal layer connect the user behavior nodes and the user demand label nodes based on the target association weight.

[0070] In this embodiment, during the graph construction phase, the system encodes the behavioral events of different users into user behavior nodes and constructs a behavioral layer graph structure based on information such as timestamps and operation sequences. Simultaneously, various user demand tag nodes can be mined from user behavior nodes, and semantic hierarchical or similarity connections can be established. Furthermore, causal modeling methods can be used to identify significant causal paths between user behavior and demand, establish directional edges between user behavior nodes and user demand tag nodes, and assign a causal strength score (target association weight) to each edge to support subsequent reasoning and interpretability analysis.

[0071] It should be noted that the graph construction method is not limited to a specific form. For example, user behavior nodes can be constructed not only based on user operation events, but also generated through behavior session clustering or behavior pattern mining; user demand tags can be obtained directly by a rule engine combined with behavior keywords, or generated through user similarity tag co-occurrence modeling and inference; causal relationships can be obtained by Bayesian network modeling, or generated through graph neural network attention mechanisms, etc. Through these diverse alternative methods, the graph structure generation strategy can be flexibly adjusted according to system deployment capabilities, data scale, and computing resources, improving the model's adaptability and deployment efficiency.

[0072] According to the embodiments of this application, a multi-level index structure can be established in the graph service layer, which is oriented towards the behavior layer, the demand layer, and the causal layer: the behavior layer adopts an inverted index of the event time axis, the demand layer adopts a semantic vector nearest neighbor index, and the causal layer uses a sparse matrix to cache the behavior-demand causal weights.

[0073] In this embodiment, after receiving a query object, the system can locate the corresponding set of behavioral nodes in the user demand knowledge graph based on the target user's user identification information (such as user ID, group tags, behavioral trajectories, etc.), and simultaneously retrieve its upstream and downstream time-related events to form a structured behavioral trajectory subgraph. For example, in the behavioral layer, nodes may represent operation events such as "clicking the price comparison button," "browsing competitor pages for more than 60 seconds," and "actively searching for refund policies," while edges represent the order of occurrence of these behaviors in the time series or the logically related triggering relationships. The system can also filter and limit the range of nodes according to time windows to ensure that the extracted behavioral trajectory subgraph has contextual integrity and avoids redundant interference.

[0074] Subsequently, demand tag nodes that have semantic connections or historical statistical correlations with the aforementioned behavioral nodes can be retrieved from the demand layer. For example, if a user has repeatedly engaged in the "contact customer service" behavior and their voice emotion is marked as dissatisfaction, demand tag nodes such as "after-sales concerns" and "price complaints" may be associated. In the demand layer, edges are used to express the semantic similarity, hierarchical relationship, or co-occurrence probability between tags. The edges in the demand layer can be used to further expand the range of candidate demand tags and improve the coverage and generalization ability of the tags.

[0075] Furthermore, corresponding causal edges can be retrieved in the causal layer to determine whether there is a statistically significant causal inference path between user behavior nodes and user demand nodes. For example, if the behavior path "Contact customer service within 30 minutes after viewing the promotion policy" has a high-confidence causal edge (e.g., causal weight > 0.75, p < 0.05) with the demand tag "Question about discount conditions", the system will prioritize retaining this path and label it as a high-impact path.

[0076] In some embodiments, if the query conditions include upper-level semantic filtering items such as "specific scenario" or "emotion tag", semantic filtering can be performed first on the demand layer index, and then the behavior nodes corresponding to the filtered demand tag nodes can be searched downwards. For long-tail behavior nodes missed during the filtering process, a two-level cache compensation strategy can be provided: if the causal layer weight shows that the weight of the node and the core demand tag is higher than the dynamic threshold, then it is forcibly added to the final subgraph to ensure the integrity of the causal chain.

[0077] In operation S230, the target user's needs are analyzed based on the target user subgraph, and the needs analysis results are output.

[0078] In the embodiments of this application, to improve the readability and business operability of the analysis results, the aforementioned critical paths can be output in a visual form, constructing an interactive graph interface. The thickness and color of the path edges can dynamically change according to weights, and the color or label of the nodes reflects dimensional information such as the time, frequency, or context of the behavior. Users can view the contextual details of each behavioral node (such as event time, device type, geographical location, etc.) and the corresponding demand label nodes in the interface, supporting the rapid identification of key demand points such as potential churn risks, pre-sales hesitation, or post-sales expectations. Furthermore, to accommodate the usage habits of non-technical users, the system also supports converting the above graph analysis results into natural language reports. The report content includes summary diagnostic conclusions, key behavioral evidence, inferred demand labels and corresponding confidence levels, and explanations of typical causal paths, facilitating direct use by business personnel or embedding into customer operation processes.

[0079] In some embodiments, the system can also support automatically matching strategy templates and outputting recommended strategies based on the results of demand analysis. For example, if a high-confidence causal path in the target user subgraph involves the pattern of "repeatedly jumping to competitor pages and price sensitivity", the system can match the strategy template of "tentatively issuing coupons" and provide additional explanations such as the reasoning basis and success rate prediction in the output.

[0080] To meet the diverse needs of different users, demand analysis can also be performed on user groups. In such scenarios, user queries input customer segmentation criteria (e.g., "female users who have filed frequent complaints within the last 7 days"), and the subgraphs extracted by the system can be merged subgraphs or aggregated graphs of that group. The results of the demand analysis include group-level tag distribution, key behavioral patterns, and typical causal chain analysis. The group analysis output can provide statistical views, such as high-frequency demand tag heatmaps, key behavior trigger path distributions, and demand and behavior clustering results, to support higher-level applications such as market insights, product redesign, and strategy formulation.

[0081] It should be noted that the embodiments of this application not only support structured graph presentation and interpretable path annotation, but also support linkage interfaces with downstream decision-making systems or operation platforms, such as writing the output demand tags into the customer profile system, or triggering the real-time marketing engine to push corresponding preferential strategies.

[0082] The user requirements analysis method of this application will be specifically described below by way of preferred embodiments.

[0083] Figure 3 The flowchart illustrating a method for extracting a target user subgraph according to some exemplary embodiments of this application is shown schematically.

[0084] like Figure 3As shown, the method for extracting the target user subgraph includes operations S310 to S340.

[0085] In operation S310, based on the query input, the behavior node and demand node corresponding to the target user are extracted from the behavior layer and the demand layer.

[0086] In the embodiments of this application, behavior nodes can be filtered based on atomic operation events generated by the user within a specific time window, including behavior records such as page clicks, function calls, and abnormal dwell time; demand nodes can be filtered based on tag nodes that have high semantic relevance or historical co-occurrence with the target user, such as semantic tags such as "price sensitive", "after-sales demand", and "purchase hesitation".

[0087] In operation S320, the association weight between the behavior node and the demand node is calculated using a multi-head attention mechanism.

[0088] In the embodiments of this application, a multi-head attention mechanism can be used to calculate the semantic association weights between all connectable pairs of the extracted behavior nodes and demand nodes. Specifically, the system can employ an attention module based on the Transformer architecture, encoding each behavior node and its contextual behavior trajectory into a query vector, and representing each demand tag node as a key-value vector. Attention scores are calculated in parallel using multiple attention heads to capture the degree of attention paid to demand tags by different behavioral dimensions (such as temporal position, semantic intent, and historical conversion rate), thereby obtaining the initial association weights between behavior-demand node pairs.

[0089] For example, for each demand layer node Calculate its relationship with all behavior layer nodes. Association weight :

[0090] (1)

[0091] in, and These represent the query vector and key vector for the demand node and the behavior node, respectively. For vector dimensions.

[0092] In operation S330, the causal weights corresponding to the behavior nodes and the demand nodes in the causal layer are obtained, and the association weights are corrected based on the causal weights to obtain the target association weights.

[0093] In embodiments of this application, causal weights corresponding to behavior-demand node pairs can be further retrieved from the causal layer. These causal weights represent the statistical strength of the causal influence of behavioral events on demand tags. Subsequently, causal corrections can be applied to the association weights, for example, by calculating the target association weight using a weighted formula. This ensures that the final subgraph not only reflects the correlation between behavior and demand but also exhibits more explanatory causal guidance characteristics, thereby effectively improving the credibility and controllability of subsequent analysis results.

[0094] For example, the association weights can be adjusted based on the causal strength index obtained by training a Bayesian network model in the causal layer. The Bayesian network weight can be represented as the probability of the causal influence of a certain behavior node on a certain demand node, with a value range of [0,1]. For example, the causal strength of "clicking the return rules page" on the "after-sales intent" tag is 0.8, indicating that the behavior triggered this type of demand expression with a high probability.

[0095] Specifically, combining the Bayesian network weights of the causal layer The adjustment of attention weights can be expressed as:

[0096] (2)

[0097] In the embodiments of this application, in order to avoid the contamination of the graph structure by paths with insignificant causality or abnormal fluctuations in attention, the system can perform a confidence filtering operation after fusion to remove paths with target association weights lower than a set threshold (such as 0.5), ensuring that only key behavior-requirement pairs with actual business interpretation value are retained in the final subgraph.

[0098] In operation S340, the target user subgraph is obtained based on the behavior node, the demand node, and the target association weight. The target user subgraph can be a locally directed weighted graph, where edges represent inference paths between behavior nodes and demand nodes, and edge weights are the target association weights. Nodes retain attribute information from the original graph, including behavior occurrence time, tag confidence, and contextual sentiment.

[0099] Figure 4 The flowchart illustrating a method for pre-constructing a user demand knowledge graph according to some exemplary embodiments of this application is shown schematically.

[0100] like Figure 4 As shown, the method for pre-constructing a user demand knowledge graph includes operations S410 to S430.

[0101] In operation S410, behavioral rule information and multi-source heterogeneous data based on user information are acquired, and a behavioral layer graph is constructed based on the multi-source heterogeneous data and the behavioral rule information.

[0102] In the embodiments of this application, behavioral rule information may include operation event definitions, behavior classification criteria, event triggering logic, and upstream and downstream dependencies, etc., to guide the standardization of subsequent graph nodes and edge construction methods. Multi-source heterogeneous data can be data sets collected from multiple data sources that differ in data type, modal form, and structural organization. Specifically, it may include: structured click behavior logs, page dwell time, and geographic location information; semi-structured customer service call transcripts and user comment content; and unstructured audio files, interface operation videos, and device interaction trajectories, etc. Under different modalities, the expression and semantic carrier of user behavior differ significantly, therefore, targeted feature extraction methods need to be used for encoding.

[0103] According to embodiments of this application, the system can use multimodal modeling components to perform feature extraction processing for multi-source heterogeneous data types. For example, for image and video modalities, a visual model can be called to extract static frame features or time-dimensional interaction trajectory information, such as mouse hotspots, button focus frequencies, page scrolling directions, etc.; for voice data, an emotion recognition model can be used to output emotion tags (such as anger values, anxiety values) and perform text transcription processing; for text modalities, such as user comments, question and answer records, a semantic encoder can extract contextual embedding representations; and structured data can be directly mapped to behavioral event attribute vectors (such as device type, city code, operation time, etc.).

[0104] After feature extraction, multimodal vectors can be aligned to obtain multimodal embedding vectors in a unified format. The alignment process can be synchronized based on timestamps, user identifiers, and behavioral context to ensure that data from different modalities correctly correspond to the same behavioral event nodes. For example, the system can align user clicks, screen recordings of interface operations, and simultaneous voice conversations by time window using session numbers or event sequence identifiers, forming a complete behavioral context group. Subsequently, projection transformations or cross-modal fusion models (such as gated attention mechanisms) can be used to map different modal vectors to the same feature space, achieving semantic consistency and temporal uniformity between modalities.

[0105] In operation S420, a multimodal model is used to perform semantic analysis on the behavior layer graph to generate corresponding user demand tag nodes, and a demand layer graph is constructed based on the user demand tag nodes.

[0106] Specifically, user behavior node sequences and their corresponding multimodal context vectors can be input into a fusion model, such as a multimodal Transformer or graph neural network, to encode the node sequences and infer intent, generating demand label nodes. Subsequently, semantic association edges between labels can be constructed based on semantic hierarchical relationships between labels, label co-occurrence patterns, or domain ontology knowledge, thereby building a complete demand layer graph.

[0107] In operation S430, the causal relationship strength between the user behavior node and the user demand label node is calculated using a causal analysis model, and a causal layer graph is constructed based on the user behavior node, the user demand label node, and the causal relationship strength.

[0108] To achieve an interpretable connection between behavior and demand, the strength of the causal relationship between behavior nodes and demand label nodes can be calculated based on causal analysis models (such as Bayesian networks). For example, on a node-by-node basis, the posterior co-occurrence probability of recurring patterns in user behavior trajectories and demand labels can be evaluated. Causal inference can then be performed using control variables, outputting causal strength indices such as average treatment effect, conditional treatment effect, or causal probability. Based on the inference results, the system establishes directed edges between behavior nodes and demand nodes, attaching causal weights to construct a causal layer graph.

[0109] Furthermore, to achieve cross-platform, high-quality construction of user demand knowledge graphs while protecting user privacy, a federated learning framework can be introduced to coordinate the collaborative modeling process among multiple data providers. Specifically, the federated learning sub-nodes deployed locally on each data holder (such as e-commerce platforms, customer service systems, mobile terminal operators, etc.) can receive and process locally generated multimodal embedding vectors.

[0110] During the local modeling phase, each data provider performs a predefined model training task based on its local multimodal embedding vectors, generating local model gradients or updated model parameters. This model can be a graph neural network for behavior-demand mapping, a Bayesian factor graph model for causal relationship evaluation, or a Transformer structure. The training process is completed locally, and the original user data remains on the local device or data center, avoiding the privacy risks associated with cross-platform data movement.

[0111] To ensure the security of model parameters during the upload process, encryption mechanisms such as homomorphic encryption, differential privacy, or secure multi-party computation can be used to encrypt the local model gradients or parameter update results. The encrypted local model gradients or parameter update results are then uploaded to a central joint modeling server. The system performs decryption and parameter aggregation operations on the server side, such as integrating model contributions from different data providers based on algorithms like federated averaging or weighted aggregation.

[0112] According to embodiments of this application, global model parameters for constructing a user demand knowledge graph can be obtained without accessing the original data. The global model possesses cross-domain behavior modeling capabilities and stronger generalization performance, and can be further used for key tasks such as semantic extraction of behavior nodes, demand label prediction, and causal edge weight updates, significantly improving the accuracy and adaptability of the knowledge graph.

[0113] In some embodiments, to ensure the timeliness, dynamism, and reliability of the causal layer structure in the user demand knowledge graph, the system supports periodic incremental calculations of causal relationship strength and graph updates based on the pairing relationship between user behavior nodes and user demand tag nodes. Specifically, by monitoring user behavior data streams and demand tag identification results, the system automatically extracts the latest paired behavior node and demand tag node pairs within a set time period (such as hourly, daily, or by user behavior batch) as a set of candidate behavioral demand causal paths for behavioral demand causal analysis.

[0114] For each candidate path pair, a causal analysis sample set can be constructed based on historical behavior sequences and label feedback results, and incremental updates can be performed using causal inference methods. This incremental approach means the system does not reconstruct the entire causal model, but only adjusts parameters and corrects relationship strength for newly added behavior samples or changed label states, thus significantly reducing computational overhead and supporting online graph maintenance. This calculation process outputs the latest causal relationship strength index for each candidate path.

[0115] After obtaining the updated causal strength, statistical significance tests can be performed on these paths to assess the credibility of their causal impact. The system can determine the statistical validity of behavioral demand causal paths based on p-value tests (e.g., p < 0.05), confidence interval analysis, or hypothesis testing methods. Only when the confidence level of a behavioral demand causal path exceeds a preset threshold (e.g., significance passed, weight greater than 0.6) will the system retain the path and use it to update the causal layer map. Paths that do not meet the statistical significance standard are automatically filtered or downweighted to avoid introducing spurious causal interference.

[0116] Through the incremental update and saliency filtering mechanism of this application embodiment, the system can continuously optimize the edge structure and weight configuration in the causal layer graph, ensuring that the graph has rapid response capability and long-term stability when facing changes in user behavior.

[0117] In the embodiments of this application, requirement interpretation and result output operations can be further performed based on the causal paths of behavioral requirements in the target user subgraph to facilitate visualization and understanding by business personnel. In this process, the established causal paths of behavioral requirements between all behavioral nodes and requirement label nodes in the target user subgraph can first be sorted. The sorting is based on the target association weight corresponding to each path. The target association weight combines the semantic relevance weight generated by the multi-head attention mechanism in the previous stage with the causal strength output by the Bayesian network or other causal analysis models in the causal layer, thereby reflecting the explanatory power and credibility of each path for requirement reasoning in the current analysis context.

[0118] After sorting, the system can extract the top few high-weight paths as "critical paths" and generate an interactive visualization graph based on these paths. The interactive visualization graph uses user behavior nodes and requirement label nodes as vertices, paths as directed edges, with the edge direction pointing from behavior to requirement. The edge weights are used to control the style mapping during visualization; for example, the importance of a path can be represented by edge thickness, color gradient, or line style. The color of a node can be encoded based on its aggregation weight, trigger frequency, or semantic category. The interactive visualization graph allows users to click on nodes to view details of the original behavioral data, the meaning of requirement labels, and the reasoning process, helping to quickly identify key events and the source of core requirements in complex behavioral contexts.

[0119] In embodiments of this application, to lower the understanding threshold for non-technical users, the system can also generate structured natural language reports based on the aforementioned key paths. These natural language reports can be driven by a pre-trained language model, combining structural information from the graph with path semantic content to generate diagnostic text. For example, the report content may include: a summary section, which briefly lists the core demand tags associated with the current user or user group, and an overview of the key behavioral-demand causal path supporting the conclusion, including path weight ranking and path description; a behavioral layer evidence section, which displays a detailed list of high-weight operation events, with each event record including contextual information such as operation type, timestamp, and page or device type, to reproduce the user's operation trajectory; a demand layer inference section, which lists the demand tags obtained through inference and their confidence scores, and clearly indicates which behavioral nodes are associated with the tag or the contextual patterns that co-occur; and a causal layer verification section, which provides the weight values ​​of key causal edges based on the Bayesian network modeling results, and performs statistical comparisons with business indicators (such as the return rate, churn rate, or complaint frequency of the user's corresponding customer group) to verify the actual business relevance and causal significance of the inference path, thereby making the report readable, traceable, and decision-making.

[0120] In the embodiments of this application, after completing the needs analysis of the target user, the system can automatically generate a personalized intervention strategy based on the output needs tags, key behavioral paths, and their target association weights, and implement the strategy on the user side. The intervention strategy may include various forms such as coupon distribution, proactive customer service follow-up, pop-up guidance, product recommendations, and service prompts. The specific content is dynamically matched based on the needs tag type (such as "price sensitive", "after-sales concerns", "hesitation and comparison", etc.) and the historical strategy response effects.

[0121] After the intervention strategy is executed on the target user's terminal or system side, the system continuously collects behavioral feedback data related to the strategy to evaluate the actual effect of the intervention. For example, behavioral feedback data may include: strategy reach response rate (such as click-through rate, view rate), subsequent behavioral changes (such as adding to cart, placing orders, changes in page dwell time), and final business conversion indicators (such as order rate, return rate, customer service follow-up success rate, etc.). All feedback data is collected in real time and corresponding intervention effect indicator vectors are generated locally for subsequent federated learning processing.

[0122] To ensure user data privacy and compliance with multi-platform joint training, the system employs a federated learning framework to model intervention feedback data. Locally, each data provider constructs a policy response model based on feedback metrics, outputting updated policy parameters or gradients. The local model update results are then encrypted using homomorphic encryption or differential privacy mechanisms. The encrypted parameters are subsequently uploaded to the joint modeling server, where they are uniformly decrypted and aggregated to form updated global model parameters. These updated global model parameters not only include the latest behavior-demand mapping but also dynamically incorporate intervention policy response effects, enabling the model to rapidly learn from changes in user behavior and policy adaptability.

[0123] In the embodiments of this application, to verify the effectiveness of the intervention strategy and guide the dynamic optimization of the user demand knowledge graph, a counterfactual inference mechanism can be introduced to construct a comparative model between the intervention group and the control group, thereby quantifying the real impact of the intervention measures on user behavior or business indicators. At the operational level, the system receives intervention behavior feedback data uploaded by various data providers on the joint modeling server and divides the samples into an "intervention group" (users who receive a specific strategy) and a "control group" (users who do not receive intervention but have similar characteristics) based on the implementation of the strategy. To ensure statistical comparability between the two groups, the system can use counterfactual modeling methods such as propensity score matching, inverse probability weighting, and difference within difference to construct a control system and balance and control potential confounding variables between groups.

[0124] After the comparative model is constructed, the system calculates the average treatment effect or conditional treatment effect of the intervention strategy based on the differences between the intervention group and the control group in target business indicators (such as conversion rate, return rate, and dwell time) to assess whether the strategy has a positive impact. For example, when the order conversion rate of the intervention group is significantly higher than that of the control group, and the difference meets the statistical significance level (e.g., p < 0.05), the intervention strategy can be considered to have a good response effect in the current context.

[0125] If the average processing effect meets the preset business or statistical threshold, the system will automatically write the global model parameters and intervention strategy parameters generated by this intervention learning into the user demand knowledge graph. The global model parameters are used to update the weight relationship or graph structure between behavior nodes and demand label nodes, while the intervention strategy parameters serve as extended attributes and path labels in the graph, recording the strategy type, effective scope, and historical response performance. This update mechanism allows the graph to not only reflect the current user behavior and demand status but also dynamically integrate response feedback after strategy intervention, thereby achieving the co-evolution of the graph structure and strategy logic, enhancing the model's adaptability and decision-making closed-loop capability.

[0126] By introducing controlled experiments and causal evaluation mechanisms, the embodiments of this application can effectively avoid the misconception that "correlation does not equal causation," ensuring that map updates are based only on effective and significant intervention behaviors, which helps to improve the accuracy of strategy recommendations and the stability of the system's long-term operation.

[0127] Corresponding to the above-described user demand analysis method, embodiments of this application also provide a user demand analysis device.

[0128] Figure 5 A schematic block diagram of a user requirements analysis apparatus according to an embodiment of this application is shown.

[0129] like Figure 5 As shown, the user demand analysis device 500 in this embodiment includes a data acquisition module 510, a target user subgraph acquisition module 520, and a demand analysis module 530.

[0130] The data acquisition module 510 can be used to acquire the query input of the target user. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0131] The target user subgraph acquisition module 520 can be used to extract a target user subgraph from a pre-constructed user demand knowledge graph based on the query input. The user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer. The nodes in the behavior layer include user behavior nodes, and the edges in the behavior layer include temporal or logical dependencies between the user behavior nodes. The nodes in the demand layer include user demand label nodes, and the edges in the demand layer include semantic relationships between the user demand label nodes. The causal layer represents the causal inference relationship between the user behavior nodes and the user demand label nodes, and the edges in the causal layer connect the user behavior nodes and the user demand label nodes based on target association weights. In one embodiment, the target user subgraph acquisition module 520 can be used to perform the operation S220 described above, which will not be repeated here.

[0132] The requirements analysis module 530 can be used to perform requirements analysis on the target user based on the target user subgraph and output the requirements analysis results. In one embodiment, the requirements analysis module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0133] According to an embodiment of this application, the target user subgraph acquisition module 520 can also be used to extract the behavior nodes and demand nodes corresponding to the target user from the behavior layer and the demand layer based on the query input; calculate the association weight between the behavior nodes and the demand nodes using a multi-head attention mechanism; obtain the causal weights corresponding to the behavior nodes and the demand nodes in the causal layer; correct the association weights based on the causal weights to obtain the target association weights; and obtain the target user subgraph based on the behavior nodes, the demand nodes, and the target association weights.

[0134] According to an embodiment of this application, the demand analysis module 530 can also be used to sort the causal paths of behavioral demands in the target user subgraph according to the target association weights to obtain target paths; and generate an interactive visualization graph based on the target paths, wherein the nodes and edges of the interactive visualization graph are displayed using color mapping based on the target association weights; and / or generate a natural language report using a language model based on the target paths and the target association weights.

[0135] According to embodiments of this application, the user demand analysis device 500 may further include a graph construction module. The graph construction module can be used to acquire behavioral rule information and multi-source heterogeneous data based on user information; construct a behavioral layer graph based on the multi-source heterogeneous data and the behavioral rule information; perform semantic analysis on the behavioral layer graph using a multimodal model to generate corresponding user demand tag nodes; construct a demand layer graph based on the user demand tag nodes; and calculate the causal relationship strength between the user behavior nodes and the user demand tag nodes using a causal analysis model; construct a causal layer graph based on the user behavior nodes, the user demand tag nodes, and the causal relationship strength.

[0136] According to embodiments of this application, the graph construction module can also be used to acquire multi-source heterogeneous data based on user information, extract features from the multi-source heterogeneous data to obtain multimodal vectors, perform feature alignment on the multimodal vectors to obtain multimodal embedding vectors, and construct the user demand knowledge graph based on the multimodal embedding vectors.

[0137] According to embodiments of this application, the graph construction module can also be used in a federated learning framework to have multiple data providers perform model training operations based on local multimodal embedding vectors to generate local model gradients or update parameters; encrypt the local model gradients or update parameters and upload the encrypted results to a joint modeling server; and decrypt and aggregate the encrypted results based on the joint modeling server to generate global model parameters for constructing the user demand knowledge graph.

[0138] According to an embodiment of this application, the graph construction module can also be used to perform incremental calculation of the causal relationship strength within a set time period based on the pairing relationship between the user behavior node and the user demand tag node; and based on the result of the incremental calculation, perform a significance test on the corresponding behavioral demand causal path, and retain the behavioral demand causal path with a confidence level exceeding a target threshold to update the causal layer graph.

[0139] According to an embodiment of this application, the user demand analysis device 500 may further include an intervention module. The intervention module can be used to generate an intervention strategy for the target user based on the demand analysis results, execute the intervention strategy on the target user side; and obtain behavioral feedback data corresponding to the intervention strategy, upload the behavioral feedback data as an intervention effect indicator to the federated learning framework, generate intervention strategy parameters, and update the global model parameters based on the intervention strategy parameters.

[0140] According to an embodiment of this application, the intervention module can also be used to construct a comparison model between the intervention group and the control group on the joint modeling server; calculate the average treatment effect of the intervention group and the control group based on the comparison model; and if the average treatment effect meets a preset condition, write the updated global model parameters and the intervention strategy parameters into the user demand knowledge graph.

[0141] According to embodiments of this application, any multiple modules among the data acquisition module 510, target user subgraph acquisition module 520, and requirements analysis module 530 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, target user subgraph acquisition module 520, and requirements analysis module 530 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any appropriate combination of any of these three implementation methods. Alternatively, at least one of the data acquisition module 510, target user subgraph acquisition module 520, and requirements analysis module 530 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0142] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a user requirements analysis method according to an embodiment of this application.

[0143] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0144] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0145] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0146] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0147] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0148] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the user requirements analysis method provided in the embodiments of this application.

[0149] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0150] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0151] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0152] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0154] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0155] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A user requirements analysis method, characterized in that, The method includes: Obtain the query input from the target user; Based on the query input, a target user subgraph is extracted from a pre-constructed user demand knowledge graph. The user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer. The nodes in the behavior layer include user behavior nodes, and the edges in the behavior layer represent the temporal or logical dependencies between these user behavior nodes. The nodes in the demand layer include user demand tag nodes, and the edges in the demand layer represent the semantic relationships between these user demand tag nodes. The causal layer represents the causal inference relationship between the user behavior nodes and the user demand tag nodes, and the edges in the causal layer connect the user behavior nodes and the user demand tag nodes based on target association weights. Based on the target user subgraph, perform a requirement analysis of the target users and output the requirement analysis results.

2. The method according to claim 1, characterized in that, The step of extracting the target user subgraph from the pre-built user demand knowledge graph based on the query input includes: Based on the query input, extract the behavior nodes and demand nodes corresponding to the target user from the behavior layer and the demand layer; The association weights between the behavior nodes and the demand nodes are calculated using a multi-head attention mechanism. Obtain the causal weights corresponding to the behavior nodes and the demand nodes in the causal layer, and correct the association weights based on the causal weights to obtain the target association weights; and Based on the behavior node, the demand node, and the target association weight, the target user subgraph is obtained.

3. The method according to claim 1, characterized in that, Pre-construct a knowledge graph of user needs, including: Acquire behavioral rule information and multi-source heterogeneous data based on user information, and construct a behavioral layer graph based on the multi-source heterogeneous data and the behavioral rule information; Semantic analysis of the behavior layer graph is performed using a multimodal model to generate corresponding user demand tag nodes, and a demand layer graph is constructed based on these user demand tag nodes; and The causal relationship strength between the user behavior node and the user demand tag node is calculated using a causal analysis model, and a causal layer graph is constructed based on the user behavior node, the user demand tag node, and the causal relationship strength.

4. The method according to claim 1 or 3, characterized in that, Pre-construct a knowledge graph of user needs, including: Acquire multi-source heterogeneous data based on user information, extract features from the multi-source heterogeneous data, and obtain multimodal vectors; The multimodal vectors are feature aligned to obtain multimodal embedding vectors; and The user demand knowledge graph is constructed based on the multimodal embedding vectors.

5. The method according to claim 4, characterized in that, The method further includes: In the federated learning framework, multiple data providers perform model training operations based on local multimodal embedding vectors to generate local model gradients or update parameters. The local model gradient or the update parameters are encrypted, and the encrypted result is uploaded to the joint modeling server; and The encrypted results are decrypted and aggregated based on the joint modeling server to generate global model parameters for constructing the user demand knowledge graph.

6. The method according to claim 3, characterized in that, The method further includes: Based on the pairing relationship between the user behavior nodes and the user demand tag nodes, the strength of the causal relationship is incrementally calculated within a set time period; and Based on the results of incremental calculation, a significance test is performed on the corresponding behavioral demand causal path, and behavioral demand causal paths with confidence exceeding the target threshold are retained to update the causal layer map.

7. The method according to claim 2, characterized in that, The requirement analysis of the target user based on the target user subgraph, and the output of the requirement analysis results, include: The behavioral demand causal paths in the target user subgraph are sorted according to the target association weights to obtain the target paths; and An interactive visualization graph is generated based on the target path, wherein the nodes and edges of the interactive visualization graph are displayed using color mapping based on the target association weights; and / or a natural language report is generated using a language model based on the target path and the target association weights.

8. The method according to claim 5, characterized in that, The method further includes: Based on the results of the requirements analysis, an intervention strategy for the target user is generated, and the intervention strategy is executed on the target user side; and Obtain behavioral feedback data corresponding to the intervention strategy, upload the behavioral feedback data as an intervention effect indicator to the federated learning framework and the joint modeling server, generate intervention strategy parameters, and update the global model parameters based on the intervention strategy parameters.

9. The method according to claim 8, characterized in that, The method further includes: A comparative model between the intervention group and the control group was constructed on the joint modeling server; The average treatment effect of the intervention group and the control group was calculated based on the comparison model; and If the average treatment effect meets the preset conditions, the updated global model parameters and the intervention strategy parameters are written into the user demand knowledge graph.

10. A user demand analysis device, characterized in that, The device includes: The data acquisition module is used to: acquire the query input of the target user; The target user subgraph acquisition module is used to: extract a target user subgraph from a pre-constructed user demand knowledge graph based on the query input, wherein the user demand knowledge graph includes a behavior layer, a demand layer, and a causal layer; the nodes of the behavior layer include user behavior nodes, and the edges of the behavior layer include temporal or logical dependencies between the user behavior nodes; the nodes of the demand layer include user demand tag nodes, and the edges of the demand layer include semantic relationships between the user demand tag nodes; the causal layer is used to represent the causal inference relationship between the user behavior nodes and the user demand tag nodes; and the edges of the causal layer connect the user behavior nodes and the user demand tag nodes based on target association weights; and The requirements analysis module is used to: perform requirements analysis on the target users based on the target user subgraph, and output the requirements analysis results.

11. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

Images