CN-120374171-B - Market data analysis method and system based on digital platform

Abstract

The invention discloses a market data analysis method and a system based on a digital platform, which relate to the field of market data digital analysis, and the invention constructs a unified behavior event sequence based on logistics data, sales volume data and task completion degree data, the formation mechanism of the sales risk node can be mined in a multi-dimensional and hierarchical manner by combining key technical means such as a disturbance causal graph, a time hypergraph convolution network, a back-distance diffusion kernel function, vietoris-Rips Complex topological modeling and the like. The explanatory traceability analysis of sales risk is realized through a graph backtracking algorithm, and the accuracy of anomaly identification and the prospective of risk response are remarkably improved.

Inventors

• CHEN TINGTING

Assignees

• 西藏拉萨啤酒有限公司

Dates

Publication Date

20260512

Application Date

20250417

Claims (6)

1. 1. The market data analysis method based on the digital platform is characterized by comprising the following steps of: S1, three types of core data including logistics data, sales volume data and task completion degree data are extracted through digital platform integration, the three types of core data are abstractly converted into behavior events capable of being linked based on unified behavior modeling rules, and an event sequence with a consistent structure is formed to serve as a unified analysis inlet, wherein the event sequence comprises logistics nodes, sales volume nodes and task completion degree nodes; S2, based on the co-timeliness, the sequence dependency relationship and the abnormal propagation chain among the behavior events, taking the sales volume node as a response target, taking the delivery delay and/or the incomplete task as a source node, and constructing a disturbance causal graph, wherein the nodes of the disturbance causal graph represent disturbance behavior events, and the edges represent potential causal paths among the events; S3, based on a disturbance causal graph, constructing an event hypergraph structure by utilizing a sliding time window mechanism, wherein each hyperedge is connected with a plurality of disturbance behavior events affecting the same sales node in the same time period, extracting a higher-order structural relation between the disturbance behavior events through a time hypergraph convolution network, and outputting a disturbance response vector corresponding to the node; S4, carrying out weighted aggregation on the adjacent disturbance behavior event state vector of each sales node through an anti-distance diffusion kernel function, calculating disturbance energy values of the sales nodes, defining a dynamic significance threshold, dynamically updating the threshold according to the latest disturbance energy value, and carrying out real-time adjustment on the updated threshold according to the historical disturbance energy change trend in a time window, wherein when the disturbance energy value of a certain sales node exceeds the dynamic significance threshold, the node is subjected to significant disturbance, potential risks exist, the node is marked as a sales risk node, the disturbance energy value represents the total influence degree of all disturbance behavior events received by the sales node in a given time window and is used for measuring the overall disturbance intensity received by the sales node, and the anti-distance diffusion kernel function simulates the influence diffusion intensity of a certain node on other nodes in a graph structure or space distribution through the characteristic that the disturbance influence is weakened along with the increase of the space or structure distance of the events based on the distance attenuation principle; S5, constructing a topological structure for the sales risk nodes, calculating persistent holes and Gao Weituan blocks in the topological structure, and identifying an abnormal structure region by combining a Vietoris-Rips Complex method; S6, carrying out connection analysis on sales volume risk nodes of the abnormal structure area and disturbance behavior event nodes related to the sales volume risk nodes through a graph backtracking algorithm, identifying a key path affecting the sales volume nodes, tracking a causal path between the sales volume risk nodes and disturbance event nodes behind the sales volume risk nodes, generating an interpretable risk traceability map, and outputting a dynamic early warning signal; The step S3 specifically comprises the following substeps: s301, dividing a time sequence into a plurality of long-time windows, wherein the window length is set according to the service requirement and the day, hour or minute, extracting all disturbance events influencing the same sales node in each time window, regarding the events as a set with cooperative disturbance characteristics, and constructing an overtravel; s302, taking nodes and edges of an event hypergraph as input of a time hypergraph convolution network, and extracting a time sequence relation between the nodes through graph convolution operation of the time hypergraph convolution network; s303, integrating the interference information of the state vector of each event node on the whole time sequence through multi-layer convolution, and outputting the final state vector as a disturbance response vector; The step S302 specifically includes the following substeps: s3021, defining an incidence relation matrix of nodes and supersides in an event supergraph, initializing feature vectors of each disturbance behavior event node, and aggregating node features in each superside by adopting weighted average to obtain a superside vector representation; s3022, transmitting the superside vector back to the disturbance behavior event nodes participating in connection in the superside, weighting and aggregating the adjacent superside characteristics through the structure convolution operation, and updating the state vector of the nodes; s3023, performing cross-time window fusion on the node state vectors by introducing a time sequence gating mechanism, and extracting a time sequence relation between disturbance behavior event nodes; The step S5 specifically comprises the following substeps: S501, integrating sales risk nodes, and constructing a sales risk sub-graph, wherein edges of the sales risk sub-graph represent association relations among sales risk nodes, and each sales risk node in the sales risk sub-graph represents a node with abnormal sales response; S502, analyzing topological characteristics of sales risk nodes and neighborhood thereof in the sales risk subgraph through topological data analysis; s503, calculating the durability of the nodes and the edges through the connection relation between the nodes, and identifying the holes under different scales; s504, converting sales volume risk nodes and adjacent nodes around the sales volume risk nodes into a high-dimensional network structure according to distances and relations by a Vietoris-Rips Complex method, and identifying and obtaining Gao Weituan blocks, wherein Gao Weituan blocks represent synergy and potential risks among sales volume risk events; s505, carrying out topological structure analysis on Gao Weituan blocks, and identifying and obtaining an abnormal cluster structure, wherein the abnormal cluster structure represents a potential abnormal structure area.
2. 2. The market data analysis method based on a digitizing platform according to claim 1, wherein the step S2 comprises the following sub-steps: S201, calculating the correlation between each event node based on various behavior events in the event sequence, and calculating the weight value of the edge according to the strength of the correlation; S202, calibrating time sequence marks of event nodes according to time sequence relations among events to generate a directed disturbance causal graph, wherein each side represents a potential causal relation, and the weight of the side reflects the strength of causal influence; S203, based on the constructed disturbance causality graph, pruning operation of the graph is carried out, edges without obvious causality are removed, the graph structure is optimized, and the efficiency and the accuracy of subsequent analysis are improved; S204, for the ordering of the nodes, generating an ordering sequence of the disturbance causal graph by using a topological ordering method according to the time sequence and the influence, and determining the priority of important nodes and paths in subsequent analysis.
3. 3. The method for analyzing market data based on a digital platform according to claim 1, wherein in the step S6, the specific flow of the graph backtracking algorithm is to start backtracking from sales risk nodes in an abnormal structure area, and backtrack the disturbance causal graph through depth-first search.
4. 4. The method for analyzing market data based on a digital platform according to claim 1, wherein in the step S4, the disturbance energy value of the sales node is calculated specifically by squaring euclidean norms.
5. 5. A market data analysis system based on a digital platform, the system being implemented based on the market data analysis method based on a digital platform according to any one of claims 1 to 4, and comprising a market digital office module and a data analysis module, wherein: The market digital office module comprises: the background data statistics unit is used for collecting three types of core data, including logistics data, sales volume data and task completion degree data; the task scheduling unit is used for distributing tasks to related staff, tracking execution conditions and feeding back task completion degree data to the background data statistics unit in real time; the data analysis module comprises: the disturbance causal graph construction unit is used for constructing a disturbance causal graph through a causal inference and graph mining method based on the time sequence of the behavior events corresponding to various core data; The sales risk identification unit is used for calculating the disturbance energy value of each sales node through a reverse distance diffusion kernel function according to a disturbance response vector obtained by the time hypergraph convolution network, identifying the sales node with obviously increased disturbance through comparison with a historical disturbance energy mean value and a dynamic significance threshold value, and marking the sales node as a sales risk node; the topological anomaly detection unit is used for carrying out a topological modeling method Vietoris-Rips Complex on the disturbance causal graph, detecting anomaly groups in topology and outputting possible risk areas; and the risk path tracing and early warning unit is used for carrying out connection analysis on the sales volume risk nodes and the disturbance behavior event nodes related to the sales volume risk nodes through a graph tracing algorithm, identifying a key path affecting the sales volume nodes, tracing a causal path from attendance checking abnormality and task delay to logistics bottleneck, generating an interpretable risk tracing map and outputting a dynamic early warning signal.
6. 6. The digital platform based market data analysis system of claim 5, wherein the market digital office module further comprises: the attendance management unit is used for recording the working time and attendance state of staff; And the early warning visualization unit is used for receiving the dynamic early warning signals and presenting the dynamic early warning signals through the visualization terminal.

Description

Market data analysis method and system based on digital platform Technical Field The invention relates to the field of market data digital analysis, in particular to a market data analysis method and system based on a digital platform. Background Existing market data analysis methods generally rely on traditional data mining techniques, and mainly predict and make decisions through static historical data models, typically including single data source analysis based on regression analysis, time series analysis, rule engines, and the like. These methods focus on market dynamics monitoring and risk prediction from a single dimension of logistics, sales, or mission completion, typically by linear modeling or empirical rules to identify potential problems and to make simple decision support. The core of most schemes is that trend prediction is carried out through batch data processing or a simplified statistical model, and complex interaction and dynamic evolution between market behavior data are ignored. However, the shortcomings of conventional schemes are mainly manifested in the inability to efficiently handle complex associations and causal relationships between multi-source, time-varying data, especially when faced with complex market environments, where the ability to identify and predict is compromised. The prior art often relies on a static analysis method, ignores dynamic changes of market data and time sequence relations among events, cannot adapt to sudden changes in real time, and is difficult to provide deep root cause analysis. Moreover, these methods fail to accurately identify propagation chains and causal paths between different disturbance factors, making it difficult to discover potential sources of risk and to perform effective interventions. In view of these shortcomings, a more flexible and accurate data analysis scheme is needed that fully exploits and exploits the deep relationships between multi-dimensional data. Disclosure of Invention The invention aims to solve the problems in the prior art and provide a market data analysis method and system based on a digital platform, and aims to solve the core problems in the existing market data analysis, such as technical problems of unknown reasons of logistics and task data cutting, unknown sales fluctuation, risk early warning hysteresis and the like. By constructing a linkage event sequence based on a unified behavior modeling rule, integrating a disturbance causal graph and a time hypergraph convolution network to mine deep causal relations among multisource behavior events, precisely identifying and tracing sales risk paths by combining topology modeling and graph backtracking algorithms, and constructing a complete risk perception and dynamic early warning mechanism, the prospective identification and interpretable backtracking of abnormal market fluctuation are realized, and market response capability and operation decision efficiency of enterprises under complex disturbance environments are greatly improved. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: The market data analysis method based on the digital platform comprises the following steps: s1, three types of core data including logistics data, sales volume data and task completion degree data are extracted through digital platform integration, the three types of core data are abstractly converted into behavior events capable of being linked based on unified behavior modeling rules, and an event sequence with a consistent structure is formed to serve as a unified analysis inlet, wherein the event sequence comprises logistics nodes, sales volume nodes and task completion degree nodes; S2, based on the co-timeliness, the sequence dependency relationship and the abnormal propagation chain among the behavior events, taking the sales node as a response target, constructing a disturbance causal graph from the delivery delay and the task not to be a source node, wherein the nodes of the disturbance causal graph represent disturbance behavior events, and the edges represent potential causal paths among the events; S3, based on a disturbance causal graph, constructing an event hypergraph structure by utilizing a sliding time window mechanism, wherein each hyperedge is connected with a plurality of disturbance behavior events affecting the same sales node in the same time period, extracting a higher-order structural relation between the disturbance behavior events through a time hypergraph convolution network, and outputting disturbance response vectors corresponding to the nodes; s4, carrying out weighted aggregation on the adjacent disturbance behavior event state vector of each sales node through a back-distance diffusion kernel function, and calculating the disturbance energy value of the sales node; s5, constructing a topological structure for the sales risk nodes, calculating persistent holes and Gao Weituan blocks in the topologica