Data processing method and device, nonvolatile storage medium and electronic equipment

Abstract

The invention discloses a data processing method and device, a nonvolatile storage medium and electronic equipment. The method comprises the steps of obtaining business condition data, extracting entities and relationships from the business condition data, and constructing a business condition knowledge graph based on the entities and the relationships; mapping entities and relationships in the business condition knowledge graph into semantic vectors to obtain a business condition semantic vector set; aligning and aggregating the business condition semantic vector set based on the time dimension, and constructing a node state of each entity in the business condition knowledge graph on continuous time steps in combination with the user behavior data and the event data to obtain a node state sequence; generating an initial demand prediction vector, and establishing a hypergraph structure; and based on the hypergraph structure, optimizing the initial demand prediction vector to obtain a target demand prediction vector. According to the method and the device, the technical problem that the demand prediction precision is relatively low due to the fact that a related demand prediction model is difficult to effectively integrate the business condition data and the user behavior information is solved.

Description

Technical Field

[0001] This application relates to the field of big data technology, and more specifically, to a data processing method, apparatus, non-volatile storage medium, and electronic device. Background Technology

[0002] With the booming development of the digital economy, competition in the B2B (Business-to-Business) market is becoming increasingly fierce, and accurate forecasting of customer demand has become one of the core competencies for enterprises. However, in traditional demand forecasting methods, although historical transaction data and user behavior logs are used for model training, these single-data-source methods often show limitations in predictive accuracy when faced with complex market changes and diversified user needs.

[0003] On the one hand, business intelligence data, including but not limited to industry news, policy announcements, competitor activities, and supply chain information, provides a valuable external perspective for understanding the needs of B-end users. This data can reveal macroeconomic factors such as market trends, policy guidance, and industry cycles, which have a significant indirect impact on potential user demand. However, due to the unstructured nature of business intelligence data and the difficulty in establishing correlations with user behavior information, existing predictive models struggle to effectively integrate and utilize this information, thus reducing the comprehensiveness and accuracy of predictions.

[0004] On the other hand, user behavior information, such as B-end users' access records, inquiry behavior, and purchasing history, is an important basis for predicting their specific needs. Although this type of data can directly reflect user preferences and demand patterns, due to the complexity and diversity of user behavior, relying solely on historical behavior data often makes it difficult to capture rapid changes in demand and potential growth points. Especially in the case of new customers or drastic changes in the market environment, the lack of effective external information supplementation leads to blind spots in the predictive models' ability to capture demand.

[0005] Furthermore, many related predictive models employ static modeling, meaning that during the model training phase, only the performance of a fixed historical dataset is considered, failing to account for the constantly updating nature of data over time. This static modeling approach makes the predictive results lagging and inflexible in the face of sudden market events, industry trend changes, and policy adjustments, hindering real-time optimization of the predictive model to adapt to dynamic market changes.

[0006] There is currently no effective solution to the above problems. Summary of the Invention

[0007] This application provides a data processing method, apparatus, non-volatile storage medium, and electronic device to at least solve the technical problem of low accuracy in demand forecasting caused by the difficulty in effectively integrating market intelligence data and user behavior information in relevant demand forecasting models.

[0008] According to one aspect of this application, a data processing method is provided, comprising: acquiring business intelligence data; extracting entities and relationships from the business intelligence data; and constructing a business intelligence knowledge graph based on the entities and relationships; mapping the entities and relationships in the business intelligence knowledge graph to semantic vectors to obtain a set of business intelligence semantic vectors; aligning and aggregating the set of business intelligence semantic vectors based on the time dimension, and combining user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps to obtain a node state sequence; generating an initial demand prediction vector based on the node state sequence, and constructing a hypergraph structure based on the initial demand prediction vector, wherein the hyperedges in the hypergraph structure are used to represent business behaviors; and optimizing the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector.

[0009] Optionally, entities and relations in the business intelligence knowledge graph are mapped to semantic vectors to obtain a set of business intelligence semantic vectors. This includes: performing natural language processing on the business intelligence data to obtain target triples, where each target triple includes a subject, relation, and object; mapping the target triples to the demand semantic space using a semantic representation learning function to obtain event vectors, and calculating the semantic similarity between the event vectors and multiple predefined demand category vectors; selecting target event vectors from the event vectors whose semantic similarity to any demand category vector exceeds a preset threshold; and performing linear mapping on the target event vectors using a mapping matrix. The transformation yields a demand feature representation; based on the topological connections of the business intelligence knowledge graph, local topological information of the neighboring nodes and relationships of each node is extracted; the local topological information and the demand feature representation are concatenated to obtain initial features; the initial features are iteratively updated using a graph neural network model to obtain node vectors, wherein, during the iterative update process, the demand category vector is used as a priori guiding signal for the graph attention mechanism, and semantic similarity is used to weight the importance of message passing between nodes to guide the node representation to align with the semantic space related to user needs; based on the node vectors, a set of business intelligence semantic vectors is constructed.

[0010] Optionally, an initial demand prediction vector is generated based on the node state sequence, including: using a temporal graph neural network to simultaneously extract the spatial topological features and temporal dynamic features of the node states to obtain a target spatiotemporal representation; processing the target spatiotemporal representation using a fully connected layer to obtain the initial demand prediction vector; and optimizing the initial demand prediction vector based on the hypergraph structure to obtain the target demand prediction vector, including: using the initial demand prediction vector, the hypergraph structure, and the incentive signals of business behavior as inputs, and using neural differential equations to drive the evolution of node states in the hypergraph structure to optimize the initial demand prediction vector and obtain the target demand prediction vector.

[0011] Optionally, after obtaining the target demand prediction vector, the method further includes: calculating the prediction deviation vector between the target demand prediction vector and the initial demand prediction vector; performing weighted calibration on the prediction deviation vector based on the optimized hyperedge weights obtained after the evolution of the neural differential equation to obtain the calibrated deviation vector; and adding the calibrated deviation vector to the target demand prediction vector to obtain the optimized target demand prediction vector.

[0012] Optionally, the method further includes: performing dimensionality reduction and visualization rendering on the hypergraph structure, the initial demand prediction vector, and the node state evolution path during the optimization of the initial demand prediction vector to obtain a multi-dimensional interactive interface; receiving adjustment instructions sent by the user through the interactive interface, wherein the adjustment instructions include adjusting the weights of the specified hyperedge or the evolution parameters of the neural differential equation; in response to the adjustment instructions, using the initial demand prediction vector as a fixed initial value, re-executing the evolution process of the neural differential equation based on the adjusted hyperedge weights or evolution parameters to generate an updated target demand prediction vector, and updating the visualization display in the interactive interface.

[0013] Optionally, after updating the visualization in the interactive interface, the method further includes: constructing a training dataset from the interaction logs, using the adjustment instructions in a single interaction record as labels for supervised learning, and concatenating and standardizing the initial hypergraph structure representation, initial demand prediction vector, and user historical behavior embedding corresponding to the adjustment instructions as input features of the deep learning model; supervising the training of the deep learning model with the training objective of minimizing the loss between the predicted instructions generated by the deep learning model and the real instructions, wherein the trained deep learning model is used to predict the user's adjustment instructions and sends the prediction results as adjustment strategies to the user.

[0014] Optionally, the method further includes: determining the top N key entities in the ranking of predicted values ​​of each entity in the initial demand prediction vector, and obtaining the semantic vectors associated with the key entities, where N is a positive integer; constructing a prompt message by combining the top N key entities, the semantic vectors associated with the key entities, and historical business behavior data, and inputting the prompt message into a generative model; using the generative model to reason about the prompt message, and constraining the generative model to output generated content in a predefined structured data format, wherein the generated content includes at least: a list of entity combinations with synergistic or causal potential in the prediction scenario, and the natural language explanation and confidence level generated for each entity combination; locating the corresponding entity in the business intelligence knowledge graph based on the entity identifier in the generated content, and selecting the target entity combination whose confidence level meets the preset conditions from the entity combination list; dynamically generating or optimizing the hyperedges in the hypergraph structure based on the target entity combination, and storing the natural language explanations associated with the target entity combination as the attributes of the hyperedges associated with the target entity combination.

[0015] According to another aspect of this application, a data processing apparatus is also provided, comprising: an acquisition module for acquiring business intelligence data, extracting entities and relationships from the business intelligence data, and constructing a business intelligence knowledge graph based on the entities and relationships; a mapping module for mapping entities and relationships in the business intelligence knowledge graph into semantic vectors to obtain a set of business intelligence semantic vectors; a construction module for aligning and aggregating the set of business intelligence semantic vectors based on the time dimension, and combining user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps to obtain a node state sequence; a generation module for generating an initial demand prediction vector based on the node state sequence, and constructing a hypergraph structure based on the initial demand prediction vector, wherein the hyperedges in the hypergraph structure are used to represent business behaviors; and an optimization module for optimizing the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector.

[0016] According to another aspect of this application, a non-volatile storage medium is also provided, the storage medium including a stored program, wherein the program, when running, controls the device where the storage medium is located to execute the above data processing method.

[0017] According to another aspect of this application, an electronic device is also provided, comprising: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the program executes the above-described data processing method during runtime.

[0018] According to another aspect of this application, a computer program is also provided, wherein the computer program, when executed by a processor, implements the above-described data processing method.

[0019] According to another aspect of this application, a computer program product is also provided, comprising a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-described data processing method.

[0020] In this application, a business intelligence data acquisition method is employed, from which entities and relationships are extracted, and a business intelligence knowledge graph is constructed based on these entities and relationships. Entities and relationships in the business intelligence knowledge graph are mapped to semantic vectors, resulting in a set of business intelligence semantic vectors. This set of semantic vectors is aligned and aggregated based on the time dimension, and combined with user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at consecutive time steps, resulting in a node state sequence. An initial demand prediction vector is generated based on the node state sequence, and a hypergraph structure is constructed based on this initial demand prediction vector. Hyperedges in the hypergraph structure represent business behaviors. The initial demand prediction vector is then optimized based on the hypergraph structure to obtain a target demand prediction vector. Through dynamic temporal knowledge graph fusion and hypergraph neural network prediction model optimization, this approach achieves the goal of efficiently integrating business intelligence data and user behavior information, and accurately capturing demand change trends. This significantly improves the accuracy of user demand prediction, thereby solving the technical problem of low demand prediction accuracy caused by the difficulty of effectively integrating business intelligence data and user behavior information in related demand prediction models. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 This is a flowchart of a data processing method according to an embodiment of this application;

[0023] Figure 2 This is a structural diagram of a data processing apparatus according to an embodiment of this application;

[0024] Figure 3 This is a hardware structure block diagram of a computer terminal according to an embodiment of the present application for a data processing method. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] According to an embodiment of this application, a method embodiment for data processing is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] Figure 1 This is a flowchart of a data processing method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0029] Step S101: Obtain business intelligence data, extract entities and relationships from the business intelligence data, and construct a business intelligence knowledge graph based on the entities and relationships.

[0030] Business intelligence data includes, but is not limited to: news reports, industry reports, policy announcements, bidding information, and social media activity.

[0031] In step S101, natural language processing technology is used to extract entities (such as companies, products, locations, times, etc.) and relationships (such as cooperative relationships, competitive relationships, supply relationships, etc.) from the business intelligence data. Based on the entities and relationships, a business intelligence knowledge graph is constructed. The business intelligence knowledge graph not only includes all the extracted entity nodes, but also connects the entity nodes through relationship edges to form a semantic network that reflects business intelligence events and industry dynamics.

[0032] Step S102: Map the entities and relationships in the business intelligence knowledge graph into semantic vectors to obtain a set of business intelligence semantic vectors.

[0033] Step S102 utilizes deep learning models, such as a Transformer-based semantic encoder, to convert each entity and relation in the graph into a high-dimensional semantic vector. This transformation process ensures that the complex semantics of business intelligence information are preserved in the digital space, thereby forming a set of business intelligence semantic vectors, laying the foundation for subsequent analysis and prediction.

[0034] Step S103: Align and aggregate the business intelligence semantic vector set based on the time dimension, and combine user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps, thus obtaining the node state sequence.

[0035] In step S103, the semantic vector set of business intelligence is time-aligned to ensure that business intelligence information from different time periods is correctly associated and compared. Aggregation functions, such as average pooling or attention mechanisms, are used to merge the semantic vectors of the same entity at different time points to reflect the entity's state trend over time. Furthermore, user behavior data (such as access records, inquiry requests, purchase orders, etc.) and external event data (such as policy changes, industry news, etc.) are integrated to construct a comprehensive node state for each entity node at consecutive time steps, generating a node state sequence. Each node state in the node state sequence includes business intelligence information, user behavior, and the impact of external events at that time point.

[0036] Step S104: Generate an initial demand prediction vector based on the node state sequence, and construct a hypergraph structure based on the initial demand prediction vector, wherein the hyperedges in the hypergraph structure are used to represent business behaviors.

[0037] In step S104, an initial demand prediction vector is generated based on time series analysis techniques and the constructed node state sequence. This initial demand prediction vector represents the user's demand patterns at different times. A hypergraph structure is then constructed based on this initial demand prediction vector.

[0038] Hypergraphs are an extension of standard graph structures. In traditional graph structures, edges (or arcs) connect two nodes to represent binary relationships. Hypergraphs allow edges to connect more than two nodes; these edges are called hyperedges. Hyperedges can represent more complex and higher-level relationships, and are particularly suitable for describing business activities involving multiple parties or multiple node interactions.

[0039] Step S104 uses hyperedges to represent business activities, meaning it can capture complex business activities involving multiple entities, such as group buying, supply chain cooperation, multi-supplier procurement, and industry alliances. These business activities are often not simple interactions between two entities, but involve collaborative or competitive relationships among multiple entities. For example, a hyperedge can represent "bidding behavior of multiple suppliers for the same product," involving multiple supplier nodes, a buyer node, and a product node. The weight of the hyperedge may reflect the intensity of the bidding activity or its potential impact on the buyer's demand. In this way, the hypergraph structure can more realistically reflect the multilateral interaction patterns in the B2B market, enabling the demand forecasting model to consider not only user behavior and market intelligence data, but also the effects of higher-order relationships between multiple entities, thereby improving the accuracy and comprehensiveness of the forecast.

[0040] Step S105: Based on the hypergraph structure, optimize the initial demand prediction vector to obtain the target demand prediction vector.

[0041] Step S105 employs a hypergraph neural network algorithm to iteratively optimize the node states and hyperedge connections within the hypergraph structure. This optimization process essentially refines the initial demand prediction vector, reflecting the direct impact of changes in business behavior and the market environment on demand prediction by updating node states. The hypergraph neural network algorithm dynamically adjusts the demand prediction vector using the weights of hyperedges and the interactions between nodes until optimal prediction performance is achieved. The generated target demand prediction vector accurately reflects users' potential demand trends and priorities, thereby significantly improving the accuracy and practicality of predictions.

[0042] The above steps involve acquiring business intelligence data, extracting entities and relationships from the data, and constructing a business intelligence knowledge graph based on these entities and relationships. Entities and relationships in the knowledge graph are mapped to semantic vectors, resulting in a set of business intelligence semantic vectors. This set is then aligned and aggregated based on the time dimension, and combined with user behavior data and event data to construct the node state of each entity in the knowledge graph at consecutive time steps, resulting in a node state sequence. An initial demand prediction vector is generated based on this sequence, and a hypergraph structure is constructed based on it, where hyperedges represent business behaviors. Finally, the initial demand prediction vector is optimized based on the hypergraph structure to obtain the target demand prediction vector. Through dynamic temporal knowledge graph fusion and hypergraph neural network prediction model optimization, this approach achieves the goal of efficiently integrating business intelligence data and user behavior information, accurately capturing demand change trends, and significantly improving the accuracy of user demand prediction.

[0043] The following are Figure 1 The steps shown are illustrated and explained by way of example.

[0044] According to some optional embodiments of this application, mapping entities and relations in a business intelligence knowledge graph to semantic vectors to obtain a set of business intelligence semantic vectors can be achieved through the following methods: performing natural language processing on the business intelligence data to obtain target triples, wherein the target triples include: subject, relation, and object; mapping the target triples to the demand semantic space through a semantic representation learning function to obtain event vectors, and calculating the semantic similarity between the event vectors and multiple predefined demand category vectors; selecting target event vectors from the event vectors whose semantic similarity to any demand category vector exceeds a preset threshold; and using a mapping matrix to... The target event vector is linearly transformed to obtain the demand feature representation. Based on the topological connections of the business intelligence knowledge graph, local topological information of the neighboring nodes and relationships of each node is extracted. The local topological information and the demand feature representation are concatenated to obtain the initial features. The initial features are iteratively updated using a graph neural network model to obtain node vectors. During the iterative update process, the demand category vector is used as a priori guiding signal for the graph attention mechanism, and semantic similarity is used to weight the importance of message passing between nodes to guide the node representation to align with the semantic space related to user needs. Based on the node vectors, a set of business intelligence semantic vectors is constructed.

[0045] In this embodiment, the business intelligence data is first subjected to natural language processing (NLP), which includes, but is not limited to, named entity recognition, syntactic analysis, and event relation extraction. The purpose of NLP is to accurately identify the subject, object, and the relationship between them from the text, thereby constructing a series of target triples. The subject, relation, and object refer to the key participants, actions or state transitions, and the recipients of those actions or state transitions in the business intelligence data, respectively. Each target triple is a structured description of a business intelligence event.

[0046] Secondly, the target triples are mapped to the demand semantic space using a semantic representation learning function. This function transforms the triples into event vectors within the demand semantic space, which is a vector space in which semantically similar business information events and user demand categories are close in vector distance. This step uses a deep learning model to capture the semantic connotations of the triples, ensuring that the event vectors accurately reflect the correlation between business information events and potential user needs.

[0047] Next, the semantic similarity between the event vector and a set of predefined demand category vectors is calculated. Cosine similarity, dot product, or other distance metrics are used to quantify the proximity of the event vector to each demand category vector. Target event vectors that have a semantic similarity exceeding a predetermined threshold with at least one demand category vector are selected from the event vector set. This selection process helps remove irrelevant or low-relevance business intelligence events, focusing on the business intelligence information most influential on predicting user demand.

[0048] Next, a linear transformation is performed on the selected target event vectors using a pre-defined mapping matrix. The purpose of this linear transformation is to map the event vectors to a demand feature space to obtain a demand feature representation. The mapping matrix represents the correlation between the original event semantics and the user's potential demand behavior features; the demand features represent the potential impact of the business intelligence events on user needs, providing rich semantic information for subsequent predictions.

[0049] Furthermore, based on the topological connections of the graph, each node in the graph is visited one by one, and its neighboring nodes and the relationships between them are collected. The local topological information is then concatenated with the previously obtained demand feature representation to generate the initial features of each node. The initial features include the node's own attributes and its position and role in the business intelligence network.

[0050] A graph neural network model is used to iteratively update the initial features to obtain more refined node vectors. During the iteration process, predefined demand category vectors are used as prior inputs to the graph attention mechanism, giving them a guiding role. The message passing weights between nodes are adjusted by calculating semantic similarity to ensure that the node vectors gradually evolve in a direction highly relevant to user needs. Each iteration is an optimization of the node representation, ensuring the quality and relevance of the node representation.

[0051] Finally, based on the updated node vectors, a set of business intelligence semantic vectors is constructed for the business intelligence knowledge graph. This set of business intelligence semantic vectors can reflect the dynamic semantic features of each entity in the business intelligence knowledge graph.

[0052] Overall, the aforementioned deep learning technology and graph neural network models have effectively bridged the gap between market intelligence data and demand forecasting, greatly improving the accuracy and timeliness of demand forecasting.

[0053] For example, let the input business intelligence data be... The event vector is obtained after processing. :

[0054]

[0055] in, For semantic representation learning function, For embedded dimensions, It is the set of real numbers.

[0056] In this way, unstructured business intelligence data is transformed into a structured vector representation.

[0057] For different demand categories (e.g., "expansion requirements", "security compliance requirements", "cost optimization requirements"), learn a category vector for each type of requirement. Business intelligence event vector The class vectors are matched using a softmax classifier to calculate the probability that they belong to a certain demand category:

[0058]

[0059] in: : Demand category vector; : Bias term; :event Category The probability of.

[0060] To enhance the ability to express semantics to requirements, this embodiment introduces a learnable mapping matrix. Mapping matrix Used to convert event vectors into demand feature representations :

[0061]

[0062] In this way, events not only retain their original semantic features, but also learn their correlation with user needs and behaviors through the matrix, thereby improving prediction accuracy.

[0063] Taking a customer's business intelligence information as an example: "This customer plans to expand its cloud storage capacity in East China within the next two months to meet the needs of launching new services." Semantic parsing: Extract the triple (customer—expansion—cloud storage capacity) to obtain the event vector. Demand matching: The probability that the event belongs to "expansion demand" is calculated using a softmax classifier. Significantly higher than other categories. Semantic enhancement: will Input mapping matrix ,get This information is then used as input to downstream demand forecasting models to drive updates to user demand profiles.

[0064] Through the above process, the system can automatically identify customers' upcoming needs from business intelligence semantics, rather than relying on historical tags, thereby achieving dynamic and forward-looking demand forecasting.

[0065] According to some alternative embodiments of this application, generating an initial demand prediction vector based on a node state sequence can be achieved by the following method: using a temporal graph neural network to simultaneously extract the spatial topological features and temporal dynamic features of the node states to obtain a target spatiotemporal representation; processing the target spatiotemporal representation using a fully connected layer to obtain an initial demand prediction vector; and optimizing the initial demand prediction vector based on a hypergraph structure to obtain a target demand prediction vector, including: taking the initial demand prediction vector, the hypergraph structure, and the incentive signal of business behavior as inputs, and using neural differential equations to drive the evolution of node states in the hypergraph structure to optimize the initial demand prediction vector and obtain the target demand prediction vector.

[0066] First, a Temporal Graph Neural Network (TGNN) is applied to process the data, including the node state sequences of the business intelligence knowledge graph, using deep learning. This process simultaneously captures the spatial topological features and temporal dynamic features of the node states. The spatial topological features represent the connection methods and relationship strengths between nodes in the graph structure, while the temporal dynamic features represent the patterns and trends of node state changes over time. Through multi-layer iterative computation of the TGNN, the spatial topological features and temporal dynamic features are fused to obtain a comprehensive spatiotemporal representation of the evolution of relationships between nodes and the temporal sequence evolution.

[0067] Secondly, the obtained spatiotemporal representation of the target is passed as input to a fully connected layer. The fully connected layer is used to perform further nonlinear transformations and feature extraction on the spatiotemporal representation. Specifically, through the calculation of the weight matrix and bias terms, the complex spatiotemporal features extracted by the temporal graph neural network can be transformed into a higher-dimensional feature space that is easier to understand and process. The output vector, i.e., the initial demand prediction vector, carries important information about the user's potential needs.

[0068] Then, based on the constructed hypergraph structure, an optimization operation is performed on the initial demand prediction vector. In this step, the initial demand prediction vector, the hypergraph structure, and the incentive signals from business activities are all input into the neural differential equation framework. The neural differential equation is used to describe the dynamic changes of node states in the hypergraph structure in the form of differential equations. The incentive signals from business activities, as external stimuli, influence the evolution path of node states. At each time step, by solving the neural differential equation, the node states are dynamically adjusted, thereby optimizing the initial demand prediction vector until the prediction vector fully reflects the incentive effect of business activities and the temporal changes of node states.

[0069] Finally, the optimized initial demand prediction vector is transformed into a more accurate target demand prediction vector. The target demand prediction vector not only incorporates the user demand trends from the initial prediction vector but is also corrected and enhanced based on the hypergraph structure and incentive signals from business behavior, enabling it to more effectively predict changes in user demand under specific business environments. The entire optimization process is dynamic and continuous, ensuring the prediction model's rapid response to and accurate capture of market dynamics, thus enhancing the application value of the prediction results in real-world business scenarios.

[0070] Optionally, after obtaining the target demand prediction vector, the following steps can also be performed: calculate the prediction deviation vector between the target demand prediction vector and the initial demand prediction vector; perform weighted calibration on the prediction deviation vector based on the optimized hyperedge weights obtained after the evolution of the neural differential equation to obtain the calibrated deviation vector; add the calibrated deviation vector to the target demand prediction vector to obtain the optimized target demand prediction vector.

[0071] In this embodiment, firstly, for each time step, the difference between the target demand prediction vector and the corresponding initial demand prediction vector is calculated. This difference is the prediction deviation vector. The prediction deviation vector is used to represent the specific change in demand prediction accuracy before and after the model is optimized by the hypergraph structure.

[0072] Then, the prediction bias vector is weighted using the optimized hyperedge weights obtained during the evolution of the neural differential equation. Weighted calibration is a quantitative adjustment of the model's prediction bias. The hyperedge weights represent the importance of specific business behaviors to demand forecasting. By assigning optimized hyperedge weights to the prediction bias vector, the impact of certain business behaviors on the prediction results can be amplified or reduced in a targeted manner, thereby achieving more precise control over the prediction bias.

[0073] Finally, the calibrated deviation vector is added to the original target demand prediction vector to generate the optimized target demand prediction vector. This addition operation is equivalent to an iterative correction based on the calibrated deviation information, making the final prediction result more consistent with reality and improving its accuracy and reliability. Through continuous deviation calculation, weighted calibration, and result iteration, these steps not only capture the basic trends in demand forecasting but also correct subtle deviations in the forecast in a timely manner, ensuring that the forecasting model can continuously adapt to market changes and make more accurate demand predictions.

[0074] On the other hand, generating an initial demand prediction vector based on the node state sequence can also be achieved through the following methods: Based on the topological structure of the business intelligence knowledge graph, perform a spectrogram wavelet transform on the node state sequence to obtain component sequences of multiple frequency bands; use a time series prediction model to predict the component sequences of each frequency band to obtain the state components at future times; perform an inverse spectrogram wavelet transform on the state components at future times to obtain the initial demand prediction vector; and optimize the initial demand prediction vector based on the hypergraph structure to obtain the target demand prediction vector, including: using neural differential equations to drive the evolution of the node state sequence in the hypergraph structure and optimizing the weights of the hyperedges to obtain the target demand prediction vector.

[0075] In some optional embodiments of this application, the above data processing method further includes the following steps: performing dimensionality reduction and visualization rendering on the hypergraph structure, the initial demand prediction vector, and the node state evolution path during the optimization of the initial demand prediction vector to obtain a multi-dimensional interactive interface; receiving adjustment instructions sent by the user through the interactive interface, wherein the adjustment instructions include adjusting the weights of the specified hyperedge or the evolution parameters of the neural differential equation; in response to the adjustment instructions, using the initial demand prediction vector as a fixed initial value, re-executing the evolution process of the neural differential equation based on the adjusted hyperedge weights or evolution parameters to generate an updated target demand prediction vector, and updating the visualization display in the interactive interface.

[0076] In this embodiment, the hypergraph structure, the initial demand prediction vector, and the node state evolution paths recorded during the optimization process are first used as input data. Dimensionality reduction techniques, such as t-SNE or PCA, are employed to convert this high-dimensional data into a low-dimensional representation, facilitating visualization. Subsequently, data visualization tools are used to render the converted data, constructing a multi-dimensional interactive interface. This interface intuitively presents the topological features of the hypergraph structure, the trend of demand prediction, and the dynamic changes in node states, providing users with a comprehensive visual window to understand the predictive model's operational mechanism.

[0077] Secondly, it allows users to directly send adjustment commands. These commands specifically include modifying the weights of specified hyperedges or the evolutionary parameters used in the neural differential equations. Users can manually adjust hyperedge weights to emphasize or weaken the impact of specific business behaviors on the prediction, or adjust evolutionary parameters to change the speed and direction of node state evolution, based on their understanding of the market environment, business behavior, or prediction results, thereby optimizing the prediction model.

[0078] Then, in response to the adjustment command sent by the user, the computation process of the neural differential equation is re-executed, starting from the initial demand prediction vector. Specifically, the adjusted hyperedge weights or evolution parameters are applied to the neural differential equation to guide the evolution of the node states. That is, even under the same initial conditions, the final evolution of the node states will differ due to changes in the hyperedge weights or evolution parameters, thereby generating an updated target demand prediction vector. This target demand prediction vector represents the user's personalized adjustments to the prediction results and their understanding of the market.

[0079] Finally, as the prediction vector is updated, the visualization in the multi-dimensional interactive interface is also adjusted accordingly. The updated interface not only displays the adjusted demand forecast trend but also reflects the specific impact of changes in hyperedge weights and evolution parameters on the prediction results. This enables real-time interaction between the prediction model and user decision-making, allowing users to intuitively see the effects of adjustment commands and iterate multiple times until a satisfactory prediction result is obtained. Through interactive visualization and parameter adjustment mechanisms, the transparency and adaptability of the prediction model can be improved, enabling business personnel to gain a deeper understanding and optimize the prediction model to cope with ever-changing market demands.

[0080] Optionally, after updating the visualization in the interactive interface, the above data processing method further includes the following steps: constructing a training dataset from the interaction logs, using the adjustment instructions in a single interaction record as labels for supervised learning, and concatenating and standardizing the initial hypergraph structure representation, initial demand prediction vector, and user historical behavior embedding corresponding to the adjustment instructions as input features of the deep learning model; supervising the training of the deep learning model with the goal of minimizing the loss between the predicted instructions generated for the deep learning model and the real instructions, wherein the trained deep learning model is used to predict the user's adjustment instructions and sends the prediction results as adjustment strategies to the user.

[0081] Specifically, data is extracted from stored interaction logs to construct a training dataset for supervised learning. Each individual interaction record explicitly identifies the user's adjustment instructions. These instructions include modifications to hyperedge weights or adjustments to the evolution parameters of the neural differential equations. These adjustment instructions are treated as labels in the training set to guide model learning.

[0082] The initial hypergraph structure representation, initial demand prediction vector, and user historical behavior embedding in each record are processed. Specifically, this includes merging information from different sources and standardization to ensure numerical comparability of features and prevent any single feature from dominating model training. The processed results serve as input features for the deep learning model, incorporating multi-dimensional information from the hypergraph structure, demand prediction state, and user behavior. A training objective is then set: minimizing the loss between the predicted instructions generated by the deep learning model and the actual adjustment instructions issued by the user. This objective can be achieved by calculating the distance or difference metric between the two, such as using mean squared error, cross-entropy loss, or edit distance as the loss function. During model training, model parameters are continuously adjusted to ensure that the generated predicted instructions closely approximate the user's true intent, reflecting the user's patterns and preferences for adjusting specific hyperedge weights or evolution parameters. After training, the deep learning model can predict the user's adjustment instructions based on the input features. In practical applications, the model receives real-time data containing the initial hypergraph structure representation, demand prediction vector, and user behavior embedding, and predicts possible user adjustment strategies based on internally learned patterns and rules. The prediction results are sent to users in the form of adjustment strategies, allowing them to quickly understand the model's suggestions and decide whether to adopt them or make further personalized adjustments. These steps not only reduce the uncertainty users encounter when adjusting the prediction model but also continuously optimize the model through the interaction between model predictions and user feedback, improving the accuracy and usability of its prediction instructions.

[0083] As some optional embodiments of this application, the above data processing method further includes the following steps: determining the top N key entities in the ranking of the predicted values ​​of each entity in the initial demand prediction vector, and obtaining the semantic vectors associated with the key entities, where N is a positive integer; constructing prompt information by combining the top N key entities, the semantic vectors associated with the key entities, and historical business behavior data, and inputting the prompt information into a generative model; using the generative model to reason about the prompt information, and constraining the generative model to output generated content in a predefined structured data format, wherein the generated content includes at least: a list of entity combinations with synergistic or causal potential in the prediction scenario, and the natural language explanation and confidence level generated for each entity combination; locating the corresponding entity in the business intelligence knowledge graph according to the entity identifier in the generated content, and selecting the target entity combination whose confidence level meets the preset conditions from the entity combination list; dynamically generating or optimizing the hyperedges in the hypergraph structure based on the target entity combination, and storing the natural language explanations associated with the target entity combination as the attributes of the hyperedges associated with the target entity combination.

[0084] In this embodiment, firstly, the top N entities with the highest predicted values ​​from the initial demand prediction vector are selected as key entities, where N is a fixed positive integer representing the upper limit of the number of entities the model focuses on. Subsequently, the semantic vectors corresponding to these key entities are obtained. These semantic vectors include the entity's attributes, relationships, and behavioral characteristics, and are constructed based on business intelligence data and knowledge graph information, reflecting the specific connotations of the entities within the context of demand prediction.

[0085] Secondly, the prompt information is constructed. This prompt information consists of the first N key entities, the semantic vectors of the N key entities, and the user's historical business behavior data. The historical business behavior data includes the user's past decision-making patterns, preferences, and interaction history with specific entities, providing the generative model with background knowledge about user habits and the market environment. The constructed prompt information is then input into the generative model, which has the ability to reason from the input and create new content.

[0086] Then, a generative model is used to perform deep reasoning on the input prompts. Based on the entity relationships, behavioral patterns, and semantic features contained in the input information, the generative model identifies entity combinations with synergistic or causal potential in the prediction scenario. During the reasoning process, the generative model is constrained to output generated content in a predefined structured data format to ensure the consistency and readability of the output results. The generated content includes at least two key parts: a list of entity combinations, and a natural language explanation and confidence score generated for each entity combination. The entity combination list includes the set of entities that the model believes may produce synergistic effects or causal relationships in the prediction; the natural language explanation describes in detail the association logic between entity combinations; and the confidence score represents the model's confidence level in the association between entity combinations.

[0087] Furthermore, based on the entity identifiers in the generated content, the corresponding entities are precisely located in the business intelligence knowledge graph to ensure that the generated entity combinations match the actual entities existing in the graph. From the list of entity combinations, target entity combinations that meet preset conditions are selected. These preset conditions define the validity and reliability standards of the model output, and only entity combinations that meet these standards are adopted.

[0088] Finally, based on the selected target entity combinations, hyperedges in the hypergraph structure are dynamically generated or optimized. Hyperedges comprehensively consider the collaborative or causal relationships of entity combinations, as well as confidence scores, ensuring that the hypergraph structure accurately reflects the complex relationships between market dynamics and user demand predictions. Simultaneously, the natural language interpretations of the relationships generated for each target entity combination are stored as attributes of the corresponding hyperedges. These attributes not only enrich the hypergraph structure but also provide concrete evidence for subsequent model interpretation and decision support. This allows the hypergraph structure to not only reflect the technical relationships between business entities but also carry semantic interpretations and confidence information of prediction results, providing business personnel with a more comprehensive and understandable analytical perspective.

[0089] Figure 2 This is a structural diagram of a data processing apparatus according to an embodiment of this application, such as... Figure 2 As shown, the device includes:

[0090] The acquisition module 21 is used to acquire business intelligence data, extract entities and relationships from the business intelligence data, and construct a business intelligence knowledge graph based on the entities and relationships.

[0091] The mapping module 22 is used to map entities and relationships in the business intelligence knowledge graph into semantic vectors, thereby obtaining a set of business intelligence semantic vectors.

[0092] Module 23 is used to align and aggregate the set of business intelligence semantic vectors based on the time dimension, and combine user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps, thus obtaining the node state sequence.

[0093] The generation module 24 is used to generate an initial demand prediction vector based on the node state sequence, and to construct a hypergraph structure based on the initial demand prediction vector, wherein the hyperedges in the hypergraph structure are used to represent business behaviors.

[0094] The optimization module 25 is used to optimize the initial demand prediction vector based on the hypergraph structure to obtain the target demand prediction vector.

[0095] Optionally, entities and relations in the business intelligence knowledge graph are mapped to semantic vectors to obtain a set of business intelligence semantic vectors. This involves the following steps: performing natural language processing on the business intelligence data to obtain target triples, where each target triple includes a subject, relation, and object; mapping the target triples to the demand semantic space using a semantic representation learning function to obtain event vectors, and calculating the semantic similarity between the event vectors and multiple predefined demand category vectors; selecting target event vectors from the event vectors whose semantic similarity to any demand category vector exceeds a preset threshold; and applying a mapping matrix to the target event vectors... Linear transformation is performed to obtain the demand feature representation; based on the topological connections of the business intelligence knowledge graph, local topological information of the neighboring nodes and relationships of each node is extracted; the local topological information and the demand feature representation are concatenated to obtain the initial features; the initial features are iteratively updated using a graph neural network model to obtain node vectors, wherein, during the iterative update process, the demand category vector is used as the prior guiding signal of the graph attention mechanism, and semantic similarity is used to weight the importance of message passing between nodes to guide the node representation to align with the semantic space related to user needs; based on the node vectors, a set of business intelligence semantic vectors is constructed.

[0096] Optionally, an initial demand prediction vector is generated based on the node state sequence, specifically including the following steps: using a temporal graph neural network to simultaneously extract the spatial topological features and temporal dynamic features of the node states to obtain a target spatiotemporal representation; processing the target spatiotemporal representation using a fully connected layer to obtain an initial demand prediction vector; and optimizing the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector, including: using the initial demand prediction vector, the hypergraph structure, and the incentive signals of business behavior as inputs, and using neural differential equations to drive the evolution of node states in the hypergraph structure to optimize the initial demand prediction vector and obtain the target demand prediction vector.

[0097] Optionally, after obtaining the target demand prediction vector, the following steps can also be performed: calculate the prediction deviation vector between the target demand prediction vector and the initial demand prediction vector; perform weighted calibration on the prediction deviation vector based on the optimized hyperedge weights obtained after the evolution of the neural differential equation to obtain the calibrated deviation vector; add the calibrated deviation vector to the target demand prediction vector to obtain the optimized target demand prediction vector.

[0098] Optionally, the data processing device is further configured to perform the following steps: dimensionality reduction and visualization rendering of the hypergraph structure, the initial demand prediction vector, and the node state evolution path during the optimization of the initial demand prediction vector to obtain a multi-dimensional interactive interface; receiving adjustment instructions sent by the user through the interactive interface, wherein the adjustment instructions include adjusting the weights of the specified hyperedge or the evolution parameters of the neural differential equation; in response to the adjustment instructions, using the initial demand prediction vector as a fixed initial value, re-executing the evolution process of the neural differential equation based on the adjusted hyperedge weights or evolution parameters to generate an updated target demand prediction vector, and updating the visualization display in the interactive interface.

[0099] Optionally, after updating the visualization in the interactive interface, the following steps can also be performed: construct a training dataset from the interaction log, use the adjustment instructions in a single interaction record as labels for supervised learning, and concatenate and standardize the initial hypergraph structure representation, initial demand prediction vector, and user historical behavior embedding corresponding to the adjustment instructions, and use them as input features for the deep learning model; train the deep learning model under supervision with the training objective of minimizing the loss between the predicted instructions generated for the deep learning model and the real instructions, wherein the trained deep learning model is used to predict the user's adjustment instructions and send the prediction results as adjustment strategies to the user.

[0100] Optionally, the data processing device is further configured to perform the following steps: In the ranking of predicted values ​​of each entity in the initial demand prediction vector, determine the top N key entities and obtain the semantic vectors associated with the key entities, where N is a positive integer; construct a prompt message by combining the top N key entities, the semantic vectors associated with the key entities, and historical business behavior data, and input the prompt message into a generative model; use the generative model to reason about the prompt message and constrain the generative model to output generated content in a predefined structured data format, wherein the generated content includes at least: a list of entity combinations with synergistic or causal potential in the prediction scenario, and natural language explanations and confidence levels generated for each entity combination; locate the corresponding entity in the business intelligence knowledge graph based on the entity identifier in the generated content, and select target entity combinations from the entity combination list whose confidence levels meet preset conditions; dynamically generate or optimize hyperedges in the hypergraph structure based on the target entity combinations, and store the natural language explanations associated with the target entity combinations as attributes of the hyperedges corresponding to the target entity combinations.

[0101] It should be noted that the above Figure 2 The modules in can be program modules (e.g., a set of program instructions that implements a specific function) or hardware modules. For the latter, they can be represented in the following forms, but are not limited to these: each of the above modules is represented by a processor, or the functions of each of the above modules are implemented by a processor.

[0102] It should be noted that, Figure 2 Preferred embodiments of the shown examples can be found in [reference needed]. Figure 1 The relevant descriptions of the embodiments shown will not be repeated here.

[0103] Figure 3 A hardware structure block diagram of a computer terminal for implementing a data processing method is shown. Figure 3 As shown, the computer terminal 30 may include one or more processors 302 (shown as 302a, 302b, ..., 302n in the figure) 302 (processor 302 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 304 for storing data, and a transmission module 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 3 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 30 may also include... Figure 3 The more or fewer components shown, or having the same Figure 3 The different configurations shown.

[0104] It should be noted that the aforementioned one or more processors 302 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 30. As involved in the embodiments of this application, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0105] The memory 304 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the data processing method in this embodiment. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, thereby implementing the aforementioned data processing method. The memory 304 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 304 may further include memory remotely located relative to the processor 302, and these remote memories can be connected to the computer terminal 30 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0106] The transmission module 306 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 30. In one example, the transmission module 306 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 306 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0107] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 30.

[0108] It should be noted here that, in some optional embodiments, the above... Figure 3 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 3 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0109] It should be noted that, Figure 3 The computer terminal shown is used to execute Figure 1 The data processing method shown above is also applicable to this electronic device, and will not be repeated here.

[0110] This application also provides a non-volatile storage medium, which includes a stored program, wherein the program controls the device where the storage medium is located to execute the above data processing method when it runs.

[0111] A non-volatile storage medium performs the following functions: acquires business intelligence data, extracts entities and relationships from the business intelligence data, and constructs a business intelligence knowledge graph based on the entities and relationships; maps the entities and relationships in the business intelligence knowledge graph to semantic vectors, obtaining a set of business intelligence semantic vectors; aligns and aggregates the set of business intelligence semantic vectors based on the time dimension, and combines user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps, obtaining a node state sequence; generates an initial demand prediction vector based on the node state sequence, and constructs a hypergraph structure based on the initial demand prediction vector, where the hyperedges in the hypergraph structure are used to represent business behaviors; and optimizes the initial demand prediction vector based on the hypergraph structure to obtain the target demand prediction vector.

[0112] This application also provides an electronic device, including a memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the program executes the above-described data processing method during runtime.

[0113] The processor runs a program that performs the following functions: acquires business intelligence data, extracts entities and relationships from the business intelligence data, and constructs a business intelligence knowledge graph based on the entities and relationships; maps the entities and relationships in the business intelligence knowledge graph to semantic vectors, obtaining a set of business intelligence semantic vectors; aligns and aggregates the set of business intelligence semantic vectors based on the time dimension, and combines user behavior data and event data to construct the node state of each entity in the business intelligence knowledge graph at continuous time steps, obtaining a node state sequence; generates an initial demand prediction vector based on the node state sequence, and constructs a hypergraph structure based on the initial demand prediction vector, where the hyperedges in the hypergraph structure are used to represent business behaviors; and optimizes the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector.

[0114] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0115] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0116] In the above embodiments of this application, the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with relevant laws, regulations and standards, take necessary protective measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0121] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A data processing method, characterized by, The method comprises the following steps: acquiring business data, extracting entities and relationships from the business data, and constructing a business knowledge graph based on the entities and the relationships; mapping the entities and the relationships in the business knowledge graph into semantic vectors to obtain a set of business semantic vectors; aligning and aggregating the set of business semantic vectors based on a time dimension, and combining user behavior data and event data to construct a node state of each entity in the business knowledge graph at a continuous time step to obtain a node state sequence; generating an initial demand prediction vector according to the node state sequence, and constructing a hypergraph structure based on the initial demand prediction vector, wherein a hyperedge in the hypergraph structure is used to represent a business behavior; optimizing the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector.

2. The method of claim 1, wherein, The method comprises the following steps: performing natural language processing on the business data to obtain target triples, wherein the target triples include subjects, relationships, and objects; mapping the target triples to a demand semantic space through a semantic representation learning function to obtain event vectors, and calculating semantic similarity between the event vectors and a plurality of predefined demand category vectors; selecting target event vectors with semantic similarity to any one of the demand category vectors exceeding a preset threshold from the event vectors; performing linear transformation on the target event vectors through a mapping matrix to obtain demand feature representations; extracting local topological information of adjacent nodes and relationships of each node based on the topological connections of the business knowledge graph; concatenating the local topological information and the demand feature representations to obtain initial features; updating the initial features iteratively using a graph neural network model to obtain node vectors, wherein in the iterative updating process, the demand category vectors are used as prior guidance signals of graph attention mechanisms, and the semantic similarity is used to weight the importance of message passing between nodes to guide the alignment of node representations to the semantic space related to user demand; constructing the set of business semantic vectors according to the node vectors.

3. The method of claim 1, wherein: generating an initial demand prediction vector according to the node state sequence comprises: extracting spatial topological features and temporal dynamic features of the node state simultaneously using a time series graph neural network to obtain target spatio-temporal representations; processing the target spatio-temporal representations using a fully connected layer to obtain an initial demand prediction vector; optimizing the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector comprises: using a neural differential equation to drive the evolution of node states on the hypergraph structure with the initial demand prediction vector, the hypergraph structure, and incentive signals of the business behavior as inputs to optimize the initial demand prediction vector and obtain the target demand prediction vector.

4. The method of claim 3, wherein, After obtaining the target demand prediction vector, the method further comprises: calculating a prediction deviation vector between the target demand prediction vector and the initial demand prediction vector; weighting and calibrating the prediction deviation vector based on the optimized hyperedge weight obtained after evolution of the neural differential equation, to obtain a calibrated deviation vector; adding the calibrated deviation vector to the target demand prediction vector to obtain an optimized target demand prediction vector.

5. The method of claim 3, wherein, The method further comprises: dimension reduction and visual rendering of the hypergraph structure, the initial demand prediction vector, and the node state evolution path in the process of optimizing the initial demand prediction vector, to obtain a multi-dimensional interactive interface; receiving an adjustment instruction sent by a user through the interactive interface, wherein the adjustment instruction comprises adjustment of the weight of a specified hyperedge or the evolution parameter of the neural differential equation; in response to the adjustment instruction, taking the initial demand prediction vector as a fixed initial value, re-executing the evolution process of the neural differential equation based on the adjusted hyperedge weight or evolution parameter, generating an updated target demand prediction vector, and updating the visual display in the interactive interface.

6. The method of claim 5, wherein, After updating the visual display in the interactive interface, the method further comprises: constructing a training data set from an interaction log, taking the adjustment instruction in a single interaction record as a label for supervised learning, and taking the initial hypergraph structure representation, initial demand prediction vector, and user historical behavior embedding corresponding to the adjustment instruction after concatenation and standardization processing as input features of a deep learning model; training the deep learning model with the training target of minimizing the loss between the predicted instruction generated for the deep learning model and the real instruction, wherein the trained deep learning model is used to predict the adjustment instruction of the user, and the prediction result is sent to the user as an adjustment strategy.

7. The method of claim 1, wherein, The method further comprises: determining the top N key entities in the ranking of the predicted values of each entity in the initial demand prediction vector, and obtaining the semantic vector associated with the key entities, wherein N is a positive integer; constructing the top N key entities, the semantic vector associated with the key entities, and historical business behavior data into prompt information, and inputting the prompt information into a generative model; using the generative model to reason the prompt information, and constraining the generative model to output generated content in a predefined structured data format, wherein the generated content at least includes a list of entity combinations with synergistic or causal potential in the predicted scenario, and a natural language explanation and confidence generated for each entity combination; locating the corresponding entities in the business knowledge graph according to the entity identifiers in the generated content, and selecting a target entity combination from the list of entity combinations whose confidence satisfies a preset condition; based on the target entity combination, dynamically generating or optimizing the hyperedge in the hypergraph structure, and storing the natural language explanation corresponding to the target entity combination as an attribute of the hyperedge corresponding to the target entity combination.

8. A data processing apparatus, characterized by, includes: An acquisition module is configured to acquire business data, extract entities and relationships from the business data, and construct a business knowledge graph based on the entities and the relationships; A mapping module is configured to map the entities and the relationships in the business knowledge graph into semantic vectors to obtain a set of business semantic vectors; A construction module is configured to align and aggregate the set of business semantic vectors based on a time dimension, and construct node states of each of the entities in the business knowledge graph at consecutive time steps in combination with user behavior data and event data to obtain a node state sequence; A generation module is configured to generate an initial demand prediction vector according to the node state sequence, and construct a hypergraph structure based on the initial demand prediction vector, wherein a hyperedge in the hypergraph structure is used to represent a commercial behavior; An optimization module is configured to optimize the initial demand prediction vector based on the hypergraph structure to obtain a target demand prediction vector.

9. A non-volatile storage medium, comprising: The non-volatile storage medium includes a stored program, wherein the program controls the device in which the non-volatile storage medium is located to perform the data processing method of any one of claims 1 to 7 when the program is running.

10. An electronic device, comprising: Comprise: A memory and a processor, the processor is used to run the program stored in the memory, wherein the program executes the data processing method of any one of claims 1 to 7 when the program is running.

11. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to realize the data processing method of any one of claims 1 to 7.

Images