Artificial Intelligence-Based Rural Economic Data Analysis Methods

Abstract

This invention relates to the field of rural economic data analysis technology, specifically an artificial intelligence-based method for rural economic data analysis. The method includes: delineating an economic data perception domain in a geographic information system based on the administrative division vector data of the target administrative village; collecting multi-source heterogeneous economic data through a protocol parser of an agricultural IoT terminal and converting it into standardized economic behavior data; extracting entities using natural language processing technology to generate a structured rural economic behavior fact table; inputting the fact table into a pre-trained economic activity association model and constructing a rural industry relationship graph using a graph embedding algorithm; and calculating the centrality index of economic entities based on the graph topology to identify core operating entities. This method enables accurate processing and association analysis of rural economic data, clearly presenting industry relationships and providing effective technical support for rural economic analysis.

Description

Technical Field

[0001] This invention relates to the field of rural economic data analysis technology, and in particular to a rural economic data analysis method based on artificial intelligence. Background Technology

[0002] Current rural economic data analysis largely relies on manual statistics or simple data aggregation methods. This involves collecting various economic records from production entities within administrative villages, performing basic classification and organization to support rural economic analysis. Such technical solutions primarily focus on preliminary data collection and simple categorization, without in-depth processing and correlation analysis of the collected multi-source economic data.

[0003] Existing technologies struggle to effectively process diverse and heterogeneous rural economic data, failing to accurately identify core information and resulting in fragmented, unstructured data that fails to reflect the intrinsic connections between economic entities. Furthermore, existing technologies cannot establish the relationships between economic entities and their activities, clearly present the industrial chain structure of rural industries, or accurately identify core business entities that play a crucial role in industrial development, thus rendering rural economic data analysis lacking in focus and effectiveness.

[0004] Rural economic data comes from scattered sources and is diverse in type. How to accurately extract information from standardized economic behavior data to form structured data, and how to build relationships between economic entities and accurately screen core business entities, have become problems that need to be solved in rural economic data analysis. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an artificial intelligence-based method for rural economic data analysis.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a rural economic data analysis method based on artificial intelligence, comprising:

[0007] Based on the administrative division vector data of the target administrative village, the economic data perception domain to be analyzed is delineated in the geographic information system. The economic data perception domain includes the geographic coordinate range and administrative boundary of all natural villages under the jurisdiction of the administrative village.

[0008] By deploying a protocol parser on an agricultural IoT terminal, the system collects the agricultural input procurement flow, agricultural product output weighing records, and labor transfer and employment status of each production entity within the economic data perception domain. The collected multi-source heterogeneous data is then uniformly converted into standardized economic behavior data with timestamps.

[0009] Natural language processing technology is used to extract entities from the standardized economic behavior data, identify the names of economic entities, types of economic behaviors, and transaction amounts, and generate a structured rural economic behavior fact table.

[0010] The rural economic behavior fact table is input into a pre-trained economic activity association model, and a rural industrial relationship graph with economic entities as nodes and economic behavior as edges is constructed by graph embedding algorithm;

[0011] Based on the topological structure of the rural industrial relationship map, the centrality index of each economic entity in the industrial chain is calculated, core business entities with high connectivity are screened out, and they are used as the input data source for subsequent micro-profile analysis.

[0012] As a further aspect of the present invention, the step of using natural language processing technology to extract entities from the standardized economic behavior data, identifying the names of economic entities, types of economic behaviors, and transaction amounts, and generating a structured rural economic behavior fact table includes:

[0013] The standardized economic behavior data is segmented and tagged with parts of speech to remove stop words and meaningless modal particles, while retaining valid semantic segments containing numbers and proper nouns.

[0014] Load a pre-built rural economic field dictionary, which includes a crop name database, an agricultural input category database, and a directory of agricultural financial institutions, and compare and match the valid semantic fragments with the rural economic field dictionary;

[0015] For the valid semantic fragment that is successfully matched, its entity category is determined according to its context. The entity category is limited to three categories: economic entity name, economic behavior type, and transaction amount.

[0016] For fragmented records of the same economic transaction appearing in different data sources, deduplication and merging are performed based on timestamps and geographic location information, and missing transaction amount fields are filled in;

[0017] All completed entity entries and their relationships are written into the rural economic behavior fact table, where each row of the rural economic behavior fact table corresponds to an independent rural economic event.

[0018] As a further aspect of the present invention, the rural economic behavior fact table is input into a pre-trained economic activity association model, and a rural industrial relationship graph is constructed using a graph embedding algorithm, with economic entities as nodes and economic behaviors as edges, including:

[0019] The rural industry relationship map includes the connection weights of planting, breeding, processing and sales links;

[0020] Traverse all records in the rural economic behavior fact table, extract the unique name of the economic entity as a candidate node set, and assign a unique graph node identifier to each candidate node.

[0021] Based on the transaction flow in the economic behavior fact table, the buyer node and seller node of the transaction are found in the candidate node set. Each successful buy-sell pair is defined as a directed edge, with the starting point of the directed edge being the seller node and the ending point being the buyer node.

[0022] Based on the transaction amount in the economic behavior fact table, the initial weight value of the directed edge is set, and the initial weight value is decayed and corrected according to the time of the transaction to generate the final edge weight.

[0023] The candidate node set, the directed edge set, and the final edge weight set are imported into a graph database, and a graph traversal algorithm is used to identify the implicit industry clusters, which correspond to specific agricultural production chains.

[0024] The identified industry clusters are visually bound to their corresponding nodes and edges, and the rural industry relationship graph containing node attributes, edge attributes, and cluster attributes is output.

[0025] As a further aspect of the present invention, based on the topological structure of the rural industrial relationship graph, the centrality index of each economic entity in the industrial chain is calculated, core business entities with high connectivity are selected, and these are used as the input data source for subsequent micro-level profiling analysis, including:

[0026] The degree centrality operator is run on the rural industry relationship graph to count the number of directly connected neighboring nodes of each node and quantify the local influence of the node.

[0027] Run the betweenness centrality computation operator to calculate the frequency of each node appearing on the shortest path and quantify the global importance of the node to the control of information resources;

[0028] By introducing a weighted degree centrality algorithm, the weight of each incoming edge is included in the calculation, and fund hub-type nodes with a small number of connections but huge single transaction amounts are identified.

[0029] A comprehensive score threshold is set, which is obtained by weighted summation of the normalized scores of degree centrality, betweenness centrality, and weighted degree centrality;

[0030] Traverse all nodes in the rural industry relationship graph, remove edge nodes with a comprehensive score lower than the comprehensive score threshold, retain the remaining nodes as the core business entities with high connectivity, and export their unique identifiers to the micro-profile dataset.

[0031] As a further aspect of the present invention, it also includes the step of constructing a multidimensional feature vector of the business entity based on the microscopic portrait dataset:

[0032] Retrieve the business registration information and land transfer filing information of each core business entity in the micro-profile dataset, and extract two static attributes: registered capital size and contracted land area.

[0033] By retrospectively examining the historical records related to the core business entities in the rural economic behavior fact table, we can statistically analyze two dynamic attributes: the frequency of agricultural input procurement and the total sales amount of agricultural products in the past agricultural production cycle.

[0034] By accessing a third-party credit reporting interface, the credit performance attributes of the core operating entity are obtained, including whether it has overdue repayment records.

[0035] The static attributes, dynamic attributes, and credit performance attributes are vectorized and encoded to form a high-dimensional multidimensional feature vector of the business entity, where each dimension corresponds to a specific economic feature label.

[0036] The generated multidimensional feature vectors of the business entities are associated and stored with the corresponding core business entity unique identifiers to form a labeled sample set for training the classification model.

[0037] As a further aspect of the present invention, it also includes training a long short-term memory network model using the labeled sample set to establish an operational risk early warning mechanism, including:

[0038] The structure of the Long Short-Term Memory Network model is initialized, and the Long Short-Term Memory Network model includes an input layer, a hidden layer, and an output layer for receiving the multi-dimensional feature vector of the business entity;

[0039] A training subset and a validation subset are divided from the labeled sample set. The multidimensional feature vectors of the business entities in the training subset are arranged in chronological order and then input into the long short-term memory network model.

[0040] During model training, the credit performance attributes in the labeled sample set are used as supervision signals to calculate the error between the prediction results output by the long short-term memory network model and the actual credit status.

[0041] Based on the error value, the parameters of the gated recurrent unit inside the long short-term memory network model are adjusted by backpropagation until the misclassification rate of the long short-term memory network model on the validation subset reaches a stable state.

[0042] The trained Long Short-Term Memory (LSTM) network model is serialized and saved, and configured as an online service capable of receiving new multi-dimensional feature vectors of the business entity for real-time risk scoring.

[0043] As a further aspect of the present invention, it also includes performing time-series evolution analysis on the rural industry relationship map to generate a regional economic development trend report, including:

[0044] The rural economic behavior fact tables for different time segments are extracted from the distributed storage array on a quarterly basis, and historical industrial relationship sub-graphs for each time segment are constructed respectively.

[0045] A graph isomorphism comparison is performed on the historical industry relationship subgraphs of two adjacent time sections to identify newly added economic entity nodes, disappeared economic entity nodes, and edges whose edge weights have changed.

[0046] The identified graph structure changes are quantified into an industry expansion index, an industry decline index, and an industry transfer index. The industry transfer index is obtained by calculating the rate of change of the edges pointing between nodes.

[0047] The industry expansion index, industry decline index, and industry transfer index are mapped onto the geographic base map of the economic data perception domain to generate a visualized spatial distribution heat map.

[0048] By integrating the industry expansion index, industry decline index, industry transfer index, and spatial distribution heat map from all time segments and arranging them according to time series, a regional economic development trend report is automatically generated.

[0049] As a further aspect of the present invention, the step of performing graph isomorphism comparison on the historical industry relationship sub-graphs of two adjacent time segments to identify newly added economic entity nodes, disappeared economic entity nodes, and edges with changed edge weights includes:

[0050] Establish a node mapping table for the historical industry relationship sub-graph of two adjacent time sections, and find corresponding node pairs with the same or highly similar names by using a node name similarity matching algorithm;

[0051] Nodes that could not be found to have a corresponding relationship in the historical industrial relationship sub-graph of the other party's time segment are marked as the newly added economic entity node and the disappeared economic entity node, respectively.

[0052] For the successfully matched corresponding node pair, compare the absolute value of the difference between the edge weights connecting the two nodes at two time points. If the absolute value of the difference exceeds a preset fluctuation threshold, then mark the connection as an edge whose edge weight has changed.

[0053] Extract the attribute information of all the newly added economic entity nodes, the disappeared economic entity nodes, and the edges whose edge weights have changed, and summarize them into a graph structure difference record table.

[0054] The graph structure difference record table is used as the basic data source for calculating the industry expansion index, industry decline index, and industry transfer index.

[0055] As a further aspect of the present invention, the construction of the pre-trained economic activity correlation model includes:

[0056] Collect historical rural economic transaction data covering multiple regions and time periods, clean and standardize the historical rural economic transaction data, and generate a standardized training fact table.

[0057] The standardized training fact table is subjected to feature engineering to extract a comprehensive feature vector containing transaction amount features, transaction commodity type features, transaction time interval features, and historical behavior features of the transaction subject;

[0058] Based on the comprehensive feature vector, a continuous sequence of nodes is generated on the historical transaction network graph using a random walk algorithm. The node sequence reflects the potential connection paths between economic entities.

[0059] Using a negative sampling algorithm, negative sample nodes that are not directly connected to the target node are randomly selected from the historical transaction network graph;

[0060] Sequence data containing positive and negative sample node pairs are input into a graph embedding model. By optimizing the loss function, the vector representations of positive sample node pairs are made closer in the graph embedding space, while the vector representations of negative sample node pairs are made farther apart. This trains model parameters that can output vector representations of economic entities, ultimately forming the pre-trained economic activity association model.

[0061] As a further aspect of the present invention, a betweenness centrality computation operator is run to calculate the frequency of each node appearing on the shortest path, quantifying the global importance of nodes to information resource control, including:

[0062] On the rural industry relationship graph, two different nodes are arbitrarily selected as the source node and the target node, respectively, and all shortest paths between the source node and the target node are calculated.

[0063] Count the number of paths that pass through the current calculated node among all the shortest paths from the source node to the target node, and calculate the proportion of the number of paths that pass through the current calculated node to the total number of all shortest paths;

[0064] In the rural industry relationship graph, all source nodes and target nodes are paired and combined, the proportion of the number of paths passing through the currently calculated node to the total number of shortest paths is repeatedly calculated, and the calculation results of all pairing combinations are summed.

[0065] Divide the summation result by the total number of node pairings in the rural industry relationship graph to obtain the original betweenness centrality value of the currently calculated node;

[0066] The original betweenness centrality values ​​of all nodes in the rural industry relationship graph are normalized so that the betweenness centrality values ​​of all nodes fall within the range of zero to one. The normalized betweenness centrality values ​​are used to quantify the global importance of nodes in the control of information resources.

[0067] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0068] Natural language processing (NLP) technology is used to extract entities from standardized economic behavior data, identifying the names of economic entities, types of economic activities, and transaction amounts, generating a structured fact table of rural economic behavior. This technology can perform in-depth analysis of standardized economic behavior data, accurately extracting key economic information and transforming originally fragmented and unstructured economic data into logically clear and uniformly formatted structured data. This avoids the problem that unstructured data is difficult to use directly for analysis, allowing the inherent information of economic data to be fully explored, eliminating interference caused by data redundancy and information gaps, and enabling economic data to be directly used for subsequent correlation analysis.

[0069] A structured table of rural economic behavior facts is input into a pre-trained economic activity association model. A rural industrial relationship graph is constructed using a graph embedding algorithm, with economic entities as nodes and economic behaviors as edges. Based on the topological structure of this graph, the centrality index of each economic entity in the industrial chain is calculated, and core operating entities with high connectivity are identified. This technology clearly presents the relationships between economic entities, outlines a complete rural industrial chain, and intuitively reflects the position and role of each economic entity within the industry. It accurately distinguishes the importance of different economic entities, avoids misjudgment and omission of core operating entities, makes the industrial structure layout clearer, and enables rural economic data analysis to focus on key links and core entities. Attached Figure Description

[0070] Figure 1This is a flowchart of the artificial intelligence-based rural economic data analysis method described in this invention;

[0071] Figure 2 A flowchart for selecting core operating entities. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0073] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0074] See Figure 1, starting from the administrative division vector data of the target administrative village, accurately delineate the economic data perception domain to be analyzed in the geographic information system software. The economic data perception domain spatially and completely covers the geographic coordinate range and administrative boundaries of all natural villages under the administrative village. Subsequently, through the protocol parser pre-deployed on the agricultural Internet of Things terminal device, collect multi-source heterogeneous data such as the agricultural input purchase flow, agricultural product output weighing records, and labor transfer employment status generated by each production entity within the economic data perception domain. The protocol parser uniformly converts the collected original data into economic behavior data with a standardized timestamp according to a predetermined specification. Next, use natural language processing technology to perform entity extraction on the standardized economic behavior data, automatically identify three types of key entities, namely the economic entity name, economic behavior type, and transaction amount, from the unstructured text records, and generate a structured rural economic behavior fact table based on these entities and their relationships. The generated rural economic behavior fact table is input into a pre-trained economic activity association model. The model uses the graph embedding algorithm to construct a rural industrial relationship map that can reflect the transaction and cooperation relationships between economic entities with the identified economic entities as nodes and economic behaviors as edges. Finally, based on the topological structure of the constructed rural industrial relationship map, calculate the centrality index of each economic entity node in the industrial chain. According to the set screening conditions, select the core operating entities with high connectivity from all nodes. These selected core operating entities and their related data will be used as the input data source for subsequent micro-portrait analysis.

[0075] In an embodiment of the present invention, the step of generating a rural economic behavior fact table based on natural language processing technology processes the following standardized economic behavior data entries: "On July 10, 2025, 08:15:23, Wang Moumou, a villager in Qinglong Village, bought two bags of urea from Huimin Agricultural Input Store at a price of two hundred yuan and paid with WeChat", and another entry is "On July 10, Wang Moumou in Qinglong Village bought two packages of urea for two hundred yuan exactly". The word segmentation process first divides the text into independent word units. The词性标注 (lexical category tagging) assigns lexical category tags to each word. The operation of removing stop words and meaningless modal particles filters out words such as "is", "of", "already", "with", etc., and the set of remaining effective semantic fragments is {"Qinglong Village", "Wang Moumou", "buy", "urea", "two bags", "two hundred yuan", "pay"}. Load the pre-constructed rural economic domain dictionary, which contains a crop name library, an agricultural input category library, and a list of agricultural-related financial institutions. Compare and match the effective semantic fragments with the rural economic domain dictionary. The fragment "urea" is successfully matched in the agricultural input category library. The fragments "Qinglong Village" and "Wang Moumou" are not directly found in the dictionary but are recognized as possible proper nouns. [[ID=(-1)]] [[ID=(-2)]]

[0076] It should be noted that the term "词性标注" in the original text seems to be a Chinese term that might need further clarification in the context. Here, I translated it as "lexical category tagging" tentatively. If it has a specific and accurate English equivalent in the relevant field, it can be adjusted accordingly.For successfully matched semantic fragments, the entity category is determined based on the context of the fragment. In the context "Wang, a villager from Qinglong Village, purchased urea from Huimin Agricultural Supplies Store," the fragments "Wang, a villager from Qinglong Village" and "Huimin Agricultural Supplies Store" are associated with the actions "from" and "purchase" in the context, and are thus classified as economic entity names. The fragment "purchase" is classified as an economic behavior type, and the fragment "two hundred yuan" is modified by the adjacent context "the price is," and is classified as a transaction amount. For fragmented records of the same economic transaction appearing in different data sources, deduplication and merging are performed based on timestamps and geographic location information. The first record has a precise timestamp "2025-07-10 08:15:23," and the second record has a vague time "July 10th." Through date normalization, "July 10th" is converted to "2025-07-10." The timestamps of the two records point to the same date, and the geographic location information "Qinglong Village" and "Qinglong Village" are confirmed to be the same area through administrative division mapping. The system determines that the two records describe the same event. The merge operation extracts the most complete information, using the precise timestamp "2025-07-10 08:15:23", to identify the economic entities "Wang Moumou, a villager of Qinglong Village" and "Huimin Agricultural Supplies Store", the economic behavior type "purchase", the traded commodity "urea", the quantity "two bags", and to fill in the missing transaction amount field, finally confirming the transaction amount as "200" yuan.

[0077] All completed entity entries and their relationships are written into the Rural Economic Behavior Fact Table. The Rural Economic Behavior Fact Table has the following structure: Record Unique Identifier, Timestamp, Buyer's Name, Seller's Name, Economic Behavior Type, Transaction Goods, Quantity of Goods, and Transaction Amount. For the example above, a single record would have the following structure: Record Unique Identifier "EVENT_20250710_001", Timestamp "2025-07-10 08:15:23", Buyer's Name "Qinglong Village Resident Wang Moumou", Seller's Name "Huimin Agricultural Supplies Store", Economic Behavior Type "Purchase", Transaction Goods "Urea", Quantity "2", and Transaction Amount "200". Each record in the Rural Economic Behavior Fact Table corresponds to an independent rural economic event. In practice, when merging multiple fragmented records, conflict resolution rules are followed. When the same field has different values ​​in different records, the value from the data source with higher credibility is preferred. If the credibility is the same, the record with the more accurate value is used. When filling in missing values ​​for numeric fields, the average value of historical transactions of the same type by the same entity is used for estimation.

[0078] In one embodiment of the present invention, the step of constructing a rural industrial relationship graph begins by traversing all records in the rural economic behavior fact table and extracting unique economic entity names from these records to form a candidate node set. Assuming the rural economic behavior fact table contains records involving "Wang Moumou," "Huimin Agricultural Supplies Store," "Li Moumou," "Jinlong Rice Mill," "Xingnong Cooperative," and "Zhao Moumou," the extracted candidate node set contains these names. The system assigns a unique graph node identifier to each economic entity in the candidate node set; for example, "Wang Moumou" is assigned the identifier "Node_001," "Huimin Agricultural Supplies Store" is assigned the identifier "Node_002," "Li Moumou" is assigned the identifier "Node_003," "Jinlong Rice Mill" is assigned the identifier "Node_004," "Xingnong Cooperative" is assigned the identifier "Node_005," and "Zhao Moumou" is assigned the identifier "Node_006."

[0079] In practice, based on the transaction flow in the rural economic behavior fact table, the system searches for buyer and seller nodes in the candidate node set. For example, if a record shows "Wang Moumou" purchasing "urea" from "Huimin Agricultural Supplies Store," then "Wang Moumou" is the buyer node and "Huimin Agricultural Supplies Store" is the seller node. The system defines this event as a directed edge from the seller node "Huimin Agricultural Supplies Store" to the buyer node "Wang Moumou." Similarly, if a record shows "Li Moumou" selling "rice" to "Jinlong Rice Mill," then "Li Moumou" is the seller node and "Jinlong Rice Mill" is the buyer node. The system defines a directed edge from "Li Moumou" to "Jinlong Rice Mill." Another record shows "Xingnong Cooperative" providing "agricultural machinery services" to "Zhao Moumou," then "Xingnong Cooperative" is the seller node and "Zhao Moumou" is the buyer node. A directed edge is defined from "Xingnong Cooperative" to "Zhao Moumou." Each successful buy-sell pairing corresponds to a directed edge.

[0080] Based on the transaction amounts in the rural economic behavior fact table, initial weights are set for directed edges. The transaction amount from "Huimin Agricultural Supplies Store" to "Wang Moumou" is 200 yuan, so the initial weight of this directed edge is set to 200. The transaction amount from "Li Moumou" to "Jinlong Rice Mill" is 5,000 yuan, so the initial weight of this directed edge is set to 5,000. The transaction amount from "Xingnong Cooperative" to "Zhao Moumou" is 800 yuan, so the initial weight of this directed edge is set to 800. The initial weights are adjusted based on the time of the transactions. The system uses the current time as a benchmark; the further away the transaction is from the current time, the greater the weight decay, reflecting the greater impact of recent transactions on the strength of the current relationship. For example, if the current time is January 1, 2026, and one transaction occurred on July 10, 2025, and another transaction occurred on December 20, 2025, then the edge weight of the transaction on July 10, 2025, will receive a larger decay factor. The formula for generating the final edge weights is shown below:

[0081]

[0082] Where: symbol This represents the final edge weight generated after decay correction, with the symbol... This represents the initial weight value of the directed edge set according to the transaction amount, with the symbol... It is a natural constant, symbol It is the preset time decay coefficient, symbol This represents the time difference between the current time and the time the transaction occurred, in years. Applying this formula, the directed edge weight decay is smaller for transactions occurring on December 20, 2025, while the directed edge weight decay is larger for transactions occurring on July 10, 2025.

[0083] In practice, the determined set of candidate nodes, the set of directed edges, and the final set of edge weights calculated using the decay formula are imported into a graph database, Neo4j. Using the graph database's built-in graph traversal algorithms, such as the label propagation algorithm, implicit industry clusters are identified. During the traversal of nodes and edges, the graph traversal algorithm discovers dense connections between "Wang Moumou," "Li Moumou," "Zhao Moumou," and "Xingnong Cooperative," while "Huimin Agricultural Supplies Store" and "Jinlong Rice Mill" are also connected to these nodes. However, the connection pattern shows that "Wang Moumou," "Li Moumou," "Zhao Moumou," and "Xingnong Cooperative" tend to form a group with more frequent internal transactions. The graph traversal algorithm identifies this group as an industry cluster, which may correspond to the "rice planting and primary services" chain. The system visually binds the identified industry clusters with their corresponding nodes and edge attributes. In the output rural industry relationship graph, nodes belonging to the same industry cluster are assigned the same color, and the thickness of the edges is proportional to the final edge weight. The graph displays the nodes, edges, and cluster attributes in a graphical way.

[0084] In some embodiments, the construction process of the pre-trained economic activity correlation model collects historical rural economic transaction data covering multiple regions and time periods. This data originates from agricultural product transaction records and agricultural input procurement records from multiple provinces over the past five years. The historical rural economic transaction data undergoes cleaning and standardization. Cleaning includes removing duplicate records and correcting incorrectly formatted amounts and dates. Standardization involves mapping entity names from different sources to a unified code, generating a standardized training fact table. Feature engineering is then applied to the standardized training fact table to extract a comprehensive feature vector containing transaction amount features, transaction commodity type features, transaction time interval features, and historical behavior features of the transaction entity. The transaction amount feature is the log-normalized monetary value; the transaction commodity type feature is the one-hot code of the commodity category; the transaction time interval feature is the number of days between the current transaction and the previous transaction; and the historical behavior features of the transaction entity are the total number of transactions and the average transaction amount of the entity over the past three months.

[0085] Based on comprehensive feature vectors, a random walk algorithm is used to generate a continuous sequence of nodes on a historical transaction network graph. The random walk algorithm starts from any node in the historical transaction network graph, and at each step jumps to a neighboring node of the current node with a certain probability, or to any random node in the graph with a small probability, generating a node sequence of a preset length. The generated node sequence reflects the potential connection paths formed between economic entities through multiple transactions. For example, after "Wang Moumou" transacts with "Huimin Agricultural Supplies Store," "Huimin Agricultural Supplies Store" then transacts with "Jinlong Rice Mill," forming a potential supply path.

[0086] In some embodiments, a negative sampling algorithm is used to randomly select negative sample nodes that are not directly connected to the target node in the historical transaction network graph. For the positive sample node pair "Node_001, Node_002", the negative sampling algorithm randomly selects a node in the graph that is neither a neighbor of "Node_001" nor directly connected to "Node_001" by an edge, such as "Node_006", as the negative sample node, forming the negative sample node pair "Node_001, Node_006". The sequence data containing the positive and negative sample node pairs is input into a graph embedding model, which is a Node2Vec model. By optimizing the loss function, the design of the loss function makes the vector representation distance of the positive sample node pair in the graph embedding space closer, and the vector representation distance of the negative sample node pair farther. During model training, the graph embedding model continuously adjusts its parameters to ensure that the vectors of "Node_001" and "Node_002" have a high cosine similarity in the vector space, while the vectors of "Node_001" and "Node_006" have a low cosine similarity. After multiple rounds of iterative training, the model parameters converge, ultimately forming a pre-trained economic activity association model capable of outputting high-quality vector representations for any input economic entity.

[0087] In one embodiment of the present invention, see [reference] Figure 2 The steps for selecting core business entities based on the rural industry relationship graph first involve running the degree centrality calculation operator to count the number of directly connected neighboring nodes for each node in the rural industry relationship graph. The node "Xingnong Cooperative" has directed edges connected to nodes "Wang Moumou," "Li Moumou," "Zhao Moumou," and "Huimin Agricultural Supplies Store," so the number of neighboring nodes for "Xingnong Cooperative" is 4, and its degree centrality is 4. Similarly, the node "Jinlong Rice Mill" has directed edges connected to nodes "Li Moumou" and "Huimin Agricultural Supplies Store," so its number of neighboring nodes is 2, and its degree centrality is 2. The degree centrality calculation quantifies the local influence of a node; nodes with a larger number of neighboring nodes have more extensive direct connections in the graph.

[0088] Run the betweenness centrality operator to calculate the frequency with which each node appears on the shortest path between any two distinct nodes in the rural industrial relationship graph. In the rural industrial relationship graph, arbitrarily select two distinct nodes as the source node and the target node, and calculate all shortest paths between the source node and the target node. Count the number of paths passing through the currently calculated node among all shortest paths from the source node to the target node, and calculate the proportion of the number of paths passing through the currently calculated node to the total number of shortest paths. In the rural industrial relationship graph, traverse all source node and target node pairings, repeatedly calculate the proportion of the number of paths passing through the currently calculated node to the total number of shortest paths, and sum the calculation results for all pairings. For the node "Xingnong Cooperative", we need to calculate the path ratio of the node "Xingnong Cooperative" in the following 10 node pairings: "Wang Moumou" and "Li Moumou", "Wang Moumou" and "Zhao Moumou", "Wang Moumou" and "Huimin Agricultural Supplies Store", "Wang Moumou" and "Jinlong Rice Mill", "Li Moumou" and "Zhao Moumou", "Li Moumou" and "Huimin Agricultural Supplies Store", "Li Moumou" and "Jinlong Rice Mill", "Zhao Moumou" and "Huimin Agricultural Supplies Store", "Zhao Moumou" and "Jinlong Rice Mill", and "Huimin Agricultural Supplies Store" and "Jinlong Rice Mill". Then, we sum these 10 ratios. Dividing the sum by the total number of node pairings in the rural industrial relationship graph yields the original betweenness centrality value of the currently calculated node. Assuming the rural industrial relationship graph contains 6 nodes, the total number of node pairings is 15. The node "Xingnong Cooperative" appears on the shortest path of some of the 15 node pairs, with a cumulative proportion of 3.5. Therefore, the original betweenness centrality value of node "Xingnong Cooperative" is 3.5 divided by 15, approximately equal to 0.233. The original betweenness centrality value of node "Zhao Moumou" may be 0.05. Normalization is performed on the original betweenness centrality values ​​of all nodes in the rural industrial relationship graph. The normalization method uses the maximum-minimum value normalization method, ensuring that the betweenness centrality values ​​of all nodes fall within the range of 0 to 1. The original betweenness centrality value of node "Xingnong Cooperative" (0.233) is the maximum value in the graph; after normalization, the betweenness centrality value of node "Xingnong Cooperative" is 1. The original betweenness centrality value of node "Zhao Moumou" (0.05) is approximately 0.215 after normalization. The normalized betweenness centrality value is used to quantify the global importance of a node in controlling information resources. Nodes with high betweenness centrality values ​​are located on multiple critical paths in the graph and have stronger control over resource flow.

[0089] A weighted degree centrality algorithm is introduced. When calculating the weighted degree centrality, the weight of each incoming edge pointing to the currently calculated node is included in the calculation. The node "Jinlong Rice Mill" has two incoming edges: one from node "Li Moumou" with a weight of 4500, and the other from node "Huimin Agricultural Supplies Store" with a weight of 300. Therefore, the weighted in-degree of node "Jinlong Rice Mill" is 4800. The node "Xingnong Cooperative" has three incoming edges from nodes "Wang Moumou," "Li Moumou," and "Zhao Moumou," with weights of 200, 150, and 80 respectively. Therefore, the weighted in-degree of node "Xingnong Cooperative" is 430. The weighted degree centrality algorithm identifies nodes that are financial hubs, even though they have few connections. The node "Jinlong Rice Mill" has only 2 neighboring nodes, but its weighted in-degree is 4800, which is much higher than the weighted in-degree of "Xingnong Cooperative" (430). This indicates that the node "Jinlong Rice Mill" is an important financial hub.

[0090] A comprehensive score threshold is set, which is obtained by weighted summation of the normalized scores of degree centrality, betweenness centrality, and weighted degree centrality. The degree centrality, betweenness centrality, and weighted degree centrality are each normalized to their minimum and maximum values, ensuring each score is between 0 and 1. The weights for degree centrality, betweenness centrality, and weighted degree centrality are set to 0.3, 0.3, and 0.4, respectively. For the node "Xingnong Cooperative," its normalized degree centrality score is 1, its normalized betweenness centrality score is 1, and its normalized weighted degree centrality score is assumed to be 0.1. Therefore, the comprehensive score for the node "Xingnong Cooperative" is calculated as 1 multiplied by 0.3 plus 1 multiplied by 0.3 plus 0.1 multiplied by 0.4, which equals 0.64. For the node "Jinlong Rice Mill", its normalized degree centrality score is assumed to be 0.4, its normalized betweenness centrality score is assumed to be 0.6, and its normalized weighted degree centrality score is 1. Therefore, the comprehensive score for the node "Jinlong Rice Mill" is calculated as 0.4 multiplied by 0.3 plus 0.6 multiplied by 0.3 plus 1 multiplied by 0.4, which equals 0.7. The comprehensive score threshold is set to 0.5.

[0091] Traverse all nodes in the rural industry relationship graph and calculate the comprehensive score for each node. The comprehensive score for node "Xingnong Cooperative" is 0.64, for node "Jinlong Rice Mill" it is 0.7, for node "Wang Moumou" it is assumed to be 0.35, for node "Li Moumou" it is assumed to be 0.45, for node "Zhao Moumou" it is assumed to be 0.3, and for node "Huimin Agricultural Supplies Store" it is assumed to be 0.55. Remove marginal nodes with comprehensive scores below the threshold of 0.5. The nodes with comprehensive scores below 0.5 are nodes "Wang Moumou", "Li Moumou", and "Zhao Moumou". Retain the remaining nodes as core operating entities with high connectivity. The retained nodes are nodes "Xingnong Cooperative", "Jinlong Rice Mill", and "Huimin Agricultural Supplies Store". Export the unique identifiers "Node_005", "Node_004", and "Node_002" of nodes "Xingnong Cooperative", "Jinlong Rice Mill", and "Huimin Agricultural Supplies Store" to the micro-profile dataset. During the betweenness centrality calculation, the formula for calculating all shortest paths between the source node and the target node is shown below, where the path weights are calculated based on the final edge weights:

[0092]

[0093] Where: symbol Indicates from the source node To the target node The combined metric of all possible shortest paths between them; this value is used to compare the lengths of different paths in the weighted graph, with the sign... Indicates from node To the node A specific path, symbol Indicates from node To the node The set of all paths, symbols Representing a path A directed edge on, symbol Represents a directed edge The final edge weight, sign Indicates path All directed edges Perform a series of multiplications. It has the smallest... The path with the value is considered the shortest path.

[0094] In one embodiment of the present invention, the step of constructing a multi-dimensional feature vector for the operating entity begins by retrieving the business registration information and land transfer filing information of each core operating entity from the micro-profile dataset. The micro-profile dataset contains unique identifiers of the core operating entities selected from the rural industrial relationship map, such as "Node_005", "Node_004", and "Node_002". For the core operating entity "Xingnong Cooperative", its unique identifier is "Node_005". Based on the unique identifier "Node_005", the system retrieves the business registration information of "Xingnong Cooperative" from the business registration database, extracts the registered capital scale from the business registration information of "Xingnong Cooperative", which is "1,000,000" yuan, and retrieves the land transfer filing information of "Xingnong Cooperative" from the land transfer filing system, extracts the contracted land area from the land transfer filing information of "Xingnong Cooperative", which is "500" mu. The registered capital scale and the contracted land area are two static attributes. For the core operating entity "Jinlong Rice Mill", its unique identifier is "Node_004", its registered capital extracted from the business registration information is "500,000" yuan, and its contracted land area extracted from the land transfer filing information is "50" mu. For the core operating entity "Huimin Agricultural Supplies Store", its unique identifier is "Node_002", its registered capital extracted from the business registration information is "200,000" yuan, and the land transfer filing information shows that "Huimin Agricultural Supplies Store" has no contracted land, and the contracted land area is "0" mu.

[0095] By reviewing historical records related to core business entities in the Rural Economic Behavior Facts Table, taking "Xingnong Cooperative" as an example, the system queries all records where "Xingnong Cooperative" is either the seller or buyer, limiting the query time to a complete past agricultural production cycle, such as from "2025-01-01" to "2025-12-31". The system then calculates the frequency of agricultural input purchases by "Xingnong Cooperative" during this cycle. The results show that "Xingnong Cooperative," as the buyer, purchased seeds, fertilizers, and pesticides a total of "12" times, resulting in a purchase frequency of "12". Finally, the system calculates the total sales amount of agricultural products by "Xingnong Cooperative" during this cycle. The results show that "Xingnong Cooperative," as the seller, sold rice and vegetables for a total of "150,000" yuan, resulting in a total agricultural product sales amount of "150,000" yuan. The frequency of agricultural input purchases and the total sales amount of agricultural products are two dynamic attributes. For the core operating entity, "Jinlong Rice Mill," the frequency of agricultural input procurement was "3" times during the past agricultural production cycle, and the total sales amount of agricultural products was "800,000" yuan. For the core operating entity, "Huimin Agricultural Input Store," the frequency of agricultural input procurement was "25" times, and the total sales amount of agricultural products was "0" yuan, because "Huimin Agricultural Input Store" is the main entity selling agricultural inputs and does not sell agricultural products itself.

[0096] By accessing a third-party credit reporting interface, the system obtains the credit performance attributes of core operating entities, including whether they have overdue payment records. The system queries the credit reports of core operating entities from the Credit Reference Center of the People's Bank of China via an application programming interface. For example, the credit report of "Xingnong Cooperative" shows one overdue payment record within the past three years, thus its credit performance attribute is marked as "1," indicating a negative credit record. The credit report of "Jinlong Rice Mill" shows no overdue payment records, thus its credit performance attribute is marked as "0," indicating good credit. The credit report of "Huimin Agricultural Supplies Store" shows two overdue payment records, thus its credit performance attribute is marked as "1."

[0097] Static attributes, dynamic attributes, and credit performance attributes are vectorized and encoded to form a high-dimensional multi-dimensional feature vector for the business entity, where each dimension corresponds to a specific economic characteristic label. The static attributes of registered capital and contracted land area are mapped to the first and second dimensions of the vector, respectively; the dynamic attributes of agricultural input purchase frequency and total agricultural product sales are mapped to the third and fourth dimensions, respectively; and the credit performance attribute is mapped to the fifth dimension. The specific vectorization encoding rules are as follows: registered capital is scaled by dividing by 10 (in ten thousand yuan); contracted land area is directly expressed in mu (in mu); agricultural input purchase frequency is directly expressed as the number of times; total agricultural product sales are scaled by dividing by 10 (in ten thousand yuan); and the credit performance attribute is directly marked with "0" or "1". See Table 1 for an example of the encoded multi-dimensional feature vector for the business entity.

[0098] Table 1: Characteristic Vector Coding Table for Business Entities

[0099]

[0100] In the table, vector dimension 1 is the value of registered capital divided by "10", vector dimension 2 is the original value of contracted land area, vector dimension 3 is the original value of agricultural input procurement frequency, vector dimension 4 is the value of total agricultural product sales divided by "10", and vector dimension 5 is the original label of credit performance attribute. The generated multi-dimensional feature vectors of the operating entity are associated and stored with the corresponding core operating entity unique identifier. The storage structure uses key-value pairs, where the key is the unique identifier and the value is an array of feature vectors, forming a labeled sample set for training the classification model. The label in the labeled sample set is the credit performance attribute.

[0101] In practice, a labeled sample set is used to train the Long Short-Term Memory (LSTM) network model. The LTM network model is initialized with an input layer, hidden layers, and an output layer. The number of neurons in the input layer is set to 5, consistent with the number of dimensions of the multi-dimensional feature vectors of the business entities, and is used to receive these feature vectors. The hidden layer consists of two stacked LTM network units, each containing 128 hidden units. The output layer is a fully connected layer connected to a sigmoid activation function, outputting a scalar value between 0 and 1, representing the predicted probability of business risk. A training subset and a validation subset are divided from the labeled sample set. The labeled sample set contains feature vectors and labels of 1000 core business entities, randomly divided in an 8:2 ratio, with 800 samples used as the training subset and 200 samples as the validation subset. The multidimensional feature vectors of the business entities in the training subset are arranged in chronological order and then input into the long short-term memory network model. The chronological order is determined based on the timestamps of the core business entities being entered into the system, ensuring that the model learns the sequential patterns of feature changes over time.

[0102] During model training, credit performance attributes from the labeled sample set are used as supervision signals to calculate the error between the prediction output of the Long Short-Term Memory (LSTM) network model and the actual credit status. For "Xingnong Cooperative," its actual credit status label is "1," and the model outputs a predicted probability value, such as "0.7." The error value is the difference between the model's predicted probability "0.7" and the true label "1." The error calculation uses the binary cross-entropy loss function, and the formula is shown below:

[0103]

[0104] Where: symbol Represents the binary cross-entropy loss value, symbol Represents the total number of samples in a training batch, symbol Indicates the first The true credit performance attribute label for each sample takes the value of "0" or "1", and the symbol is... This indicates that the Long Short-Term Memory network model is effective for the first... The output value of the predicted risk probability for each sample, with the sign... This represents the natural logarithm operation. Backpropagation is performed on the gated recurrent unit parameters within the Long Short-Term Memory (LSTM) network model based on the error value. The gated recurrent unit parameters include the weight matrices and bias vectors of the input gate, update gate, and reset gate. The backpropagation process calculates the gradient of the loss function with respect to each gated recurrent unit parameter and updates the parameters using gradient descent. The learning rate for parameter updates is set to 0.001. The training process iterates until the misclassification rate of the LSM network model on the validation subset reaches a stable state. The misclassification rate stabilizes around 0.15, with fluctuations less than 0.01 for more than 10 consecutive training epochs; at this point, the model is considered converged. The trained LSM network model is serialized and saved as a hierarchical data format file, and configured to receive new multi-dimensional feature vectors of business entities for real-time risk scoring as an online service. The online service receives a feature vector array containing 5 dimensions of values ​​through a web service interface, calls the loaded model for forward propagation calculation, and returns a risk score.

[0105] In one embodiment of the present invention, a time-series evolution analysis of the rural industry relationship graph is performed. Rural economic behavior fact tables for different time segments are extracted from a distributed storage array, which employs the Hadoop Distributed File System. The rural economic behavior fact table for the first quarter of 2025 is extracted from the Hadoop Distributed File System, containing economic event records from January 1, 2025 to March 31, 2025. Based on the rural economic behavior fact table for the first quarter of 2025, a historical industry relationship sub-graph for the first quarter of 2025 is constructed using the same method as for constructing the rural industry relationship graph. The historical industry relationship sub-graph for the first quarter of 2025 contains four nodes: "A", "B", "C", and "D", with edges between nodes representing transaction relationships. The rural economic behavior fact table for the second quarter of 2025 is extracted from the Hadoop Distributed File System, containing economic event records from April 1, 2025 to June 30, 2025. Based on the rural economic behavior fact table for the second quarter of 2025, a historical industrial relationship sub-graph for the second quarter of 2025 was constructed. The historical industrial relationship sub-graph for the second quarter of 2025 contains four nodes: “A”, “B”, “C”, and “E”. Node “D” disappeared, and node “E” was added.

[0106] A graph isomorphism comparison was performed on the historical industry relationship sub-graphs of two adjacent time segments, comparing the historical industry relationship sub-graphs of "Q1 2025" and "Q2 2025". A node mapping table was established for the historical industry relationship sub-graphs of the two adjacent time segments. The node mapping table is a two-dimensional table, with rows representing the node list of the historical industry relationship sub-graph of "Q1 2025" and columns representing the node list of the historical industry relationship sub-graph of "Q2 2025". An edit distance algorithm was used to calculate the edit distance between the node names in the historical industry relationship sub-graphs of "Q1 2025" and "Q2 2025". The edit distance between nodes "A" and "B" is 0, the edit distance between nodes "B" and "C" is 0, the edit distance between node "D" and all nodes in the historical industry relationship subgraph for "Q2 2025" is greater than 2, and the edit distance between node "E" and all nodes in the historical industry relationship subgraph for "Q1 2025" is greater than 2. Find the corresponding node pairs with the same name or similar height. The corresponding node pairs are "AA", "BB", and "CC".

[0107] Nodes for which no corresponding relationship can be found in the historical industry relationship sub-graph of the other party's time segment are marked as newly added economic entity nodes and disappeared economic entity nodes, respectively. Node "E" cannot find a corresponding node in the historical industry relationship sub-graph of "Q1 2025", so node "E" is marked as a newly added economic entity node in "Q2 2025" relative to "Q1 2025". Node "D" cannot find a corresponding node in the historical industry relationship sub-graph of "Q2 2025", so node "D" is marked as a disappeared economic entity node in "Q2 2025" relative to "Q1 2025".

[0108] For successfully matched pairs of nodes, the absolute difference between the edge weights connecting the two nodes is compared across two timeframes. The edge weight of the matching node pair "AB" in the historical industry relationship subgraph for the first quarter of 2025 is "100", and its edge weight in the historical industry relationship subgraph for the second quarter of 2025 is "150", with an absolute difference of "50". The preset fluctuation threshold is "20". If the absolute difference of "50" exceeds the preset fluctuation threshold "20", then the edge connecting node "A" and node "B" is marked as an edge with a changed edge weight. The edge weight of the matching node pair "BC" in the first quarter of 2025 is "80", and its edge weight in the second quarter of 2025 is "85", with an absolute difference of "5". Since the absolute difference of "5" does not exceed the preset fluctuation threshold "20", this edge is not marked as changed. The corresponding node pair "AC" has an edge in "Q1 2025" with a weight of "60", but no edge in "Q2 2025". This situation is considered as the edge weight changing from "60" to "0", and the absolute value of the difference is "60", which exceeds the preset fluctuation threshold of "20". This edge is marked as an edge whose edge weight has changed.

[0109] Extract the attribute information of all newly added economic entity nodes, disappeared economic entity nodes, and edges with changed edge weights, and summarize them into a graph structure difference record table. The graph structure difference record table contains the following record rows: Record "New Node", node identifier "E", quarter "Q2 2025". Record "Disappeared Node", node identifier "D", quarter "Q1 2025". Record "Edge Change", node pair "AB", weight change "from 100 to 150", absolute change value "50". Record "Edge Change", node pair "AC", weight change "from 60 to 0", absolute change value "60". The graph structure difference record table serves as the basic data source for subsequent calculations of the industry expansion index, industry decline index, and industry transfer index.

[0110] The identified graph structure changes are quantified into an industry expansion index, an industry decline index, and an industry transfer index. The industry expansion index is obtained by calculating the ratio of the number of newly added economic entity nodes to the total number of nodes in the previous period. The formula is: "Industry Expansion Index = (Number of New Nodes) / (Total Number of Nodes in the Previous Period)". The total number of nodes in the historical industry relationship sub-graph for "Q1 2025" is "4", and the number of newly added nodes in the historical industry relationship sub-graph for "Q2 2025" is "1", so the industry expansion index is calculated as "1 / 4 = 0.25". The industry decline index is obtained by calculating the ratio of the number of disappeared economic entity nodes to the total number of nodes in the previous period. The formula is: "Industry Decline Index = (Number of Disappeared Nodes) / (Total Number of Nodes in the Previous Period)". The total number of nodes in the historical industry relationship sub-graph for "Q1 2025" is "4", and the number of disappeared nodes in the historical industry relationship sub-graph for "Q2 2025" is "1", so the industry decline index is calculated as "1 / 4 = 0.25". The industrial transfer index is obtained by calculating the rate of change of the edges pointing between nodes. It is an indicator that measures the degree of redistribution of trading relationships among different economic entities.

[0111] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A rural economic data analysis method based on artificial intelligence, characterized in that: The method includes: Based on the administrative division vector data of the target administrative village, the economic data perception domain to be analyzed is delineated in the geographic information system. The economic data perception domain includes the geographic coordinate range and administrative boundary of all natural villages under the jurisdiction of the administrative village. By deploying a protocol parser on an agricultural IoT terminal, the system collects the agricultural input procurement flow, agricultural product output weighing records, and labor transfer and employment status of each production entity within the economic data perception domain. The collected multi-source heterogeneous data is then uniformly converted into standardized economic behavior data with timestamps. Natural language processing technology is used to extract entities from the standardized economic behavior data, identify the names of economic entities, types of economic behaviors, and transaction amounts, and generate a structured rural economic behavior fact table. The rural economic behavior fact table is input into a pre-trained economic activity association model, and a rural industrial relationship graph with economic entities as nodes and economic behavior as edges is constructed by graph embedding algorithm; Based on the topological structure of the rural industrial relationship map, the centrality index of each economic entity in the industrial chain is calculated, core business entities with high connectivity are screened out, and they are used as the input data source for subsequent micro-profile analysis.

2. The method for analyzing rural economic data based on artificial intelligence according to claim 1, characterized in that, The process involves using natural language processing technology to extract entities from the standardized economic behavior data, identifying the names of economic entities, types of economic behaviors, and transaction amounts, and generating a structured rural economic behavior fact table, including: The standardized economic behavior data is segmented and tagged with parts of speech to remove stop words and meaningless modal particles, while retaining valid semantic segments containing numbers and proper nouns. Load a pre-built rural economic field dictionary, which includes a crop name database, an agricultural input category database, and a directory of agricultural financial institutions, and compare and match the valid semantic fragments with the rural economic field dictionary; For the valid semantic fragment that is successfully matched, its entity category is determined according to its context. The entity category is limited to three categories: economic entity name, economic behavior type, and transaction amount. For fragmented records of the same economic transaction appearing in different data sources, deduplication and merging are performed based on timestamps and geographic location information, and missing transaction amount fields are filled in; All completed entity entries and their relationships are written into the rural economic behavior fact table, where each row of the rural economic behavior fact table corresponds to an independent rural economic event.

3. The method for analyzing rural economic data based on artificial intelligence according to claim 2, characterized in that, The aforementioned rural economic behavior fact table is input into a pre-trained economic activity association model. A rural industrial relationship graph is constructed using a graph embedding algorithm, with economic entities as nodes and economic behaviors as edges. This includes: The rural industry relationship map includes the connection weights of planting, breeding, processing and sales links; Traverse all records in the rural economic behavior fact table, extract the unique name of the economic entity as a candidate node set, and assign a unique graph node identifier to each candidate node. Based on the transaction flow in the economic behavior fact table, the buyer node and seller node of the transaction are found in the candidate node set. Each successful buy-sell pair is defined as a directed edge, with the starting point of the directed edge being the seller node and the ending point being the buyer node. Based on the transaction amount in the economic behavior fact table, the initial weight value of the directed edge is set, and the initial weight value is decayed and corrected according to the time of the transaction to generate the final edge weight. The candidate node set, the directed edge set, and the final edge weight set are imported into a graph database, and a graph traversal algorithm is used to identify the implicit industry clusters, which correspond to specific agricultural production chains. The identified industry clusters are visually bound to their corresponding nodes and edges, and the rural industry relationship graph containing node attributes, edge attributes, and cluster attributes is output.

4. The method for analyzing rural economic data based on artificial intelligence according to claim 3, characterized in that, Based on the topological structure of the aforementioned rural industrial relationship map, the centrality index of each economic entity in the industrial chain is calculated, and core business entities with high connectivity are screened out. These core business entities are then used as the input data source for subsequent micro-level profiling analysis, including: The degree centrality operator is run on the rural industry relationship graph to count the number of directly connected neighboring nodes of each node and quantify the local influence of the node. Run the betweenness centrality computation operator to calculate the frequency of each node appearing on the shortest path and quantify the global importance of the node to the control of information resources; By introducing a weighted degree centrality algorithm, the weight of each incoming edge is included in the calculation, and fund hub-type nodes with a small number of connections but huge single transaction amounts are identified. A comprehensive score threshold is set, which is obtained by weighted summation of the normalized scores of degree centrality, betweenness centrality, and weighted degree centrality; Traverse all nodes in the rural industry relationship graph, remove edge nodes with a comprehensive score lower than the comprehensive score threshold, retain the remaining nodes as the core business entities with high connectivity, and export their unique identifiers to the micro-profile dataset.

5. The method for analyzing rural economic data based on artificial intelligence according to claim 4, characterized in that, It also includes the step of constructing a multidimensional feature vector of the business entity based on the micro-profile dataset: Retrieve the business registration information and land transfer filing information of each core business entity in the micro-profile dataset, and extract two static attributes: registered capital size and contracted land area. By retrospectively examining the historical records related to the core business entities in the rural economic behavior fact table, we can statistically analyze two dynamic attributes: the frequency of agricultural input procurement and the total sales amount of agricultural products in the past agricultural production cycle. By accessing a third-party credit reporting interface, the credit performance attributes of the core operating entity are obtained, including whether it has overdue repayment records. The static attributes, dynamic attributes, and credit performance attributes are vectorized and encoded to form a high-dimensional multidimensional feature vector of the business entity, where each dimension corresponds to a specific economic feature label. The generated multidimensional feature vectors of the business entities are associated and stored with the corresponding core business entity unique identifiers to form a labeled sample set for training the classification model.

6. The method for analyzing rural economic data based on artificial intelligence according to claim 5, characterized in that, It also includes using the labeled sample set to train a long short-term memory network model to establish an operational risk early warning mechanism, including: The structure of the Long Short-Term Memory Network model is initialized, and the Long Short-Term Memory Network model includes an input layer, a hidden layer, and an output layer for receiving the multi-dimensional feature vector of the business entity; A training subset and a validation subset are divided from the labeled sample set. The multidimensional feature vectors of the business entities in the training subset are arranged in chronological order and then input into the long short-term memory network model. During model training, the credit performance attributes in the labeled sample set are used as supervision signals to calculate the error between the prediction results output by the long short-term memory network model and the actual credit status. Based on the error value, the parameters of the gated recurrent unit inside the long short-term memory network model are adjusted by backpropagation until the misclassification rate of the long short-term memory network model on the validation subset reaches a stable state. The trained Long Short-Term Memory (LSTM) network model is serialized and saved, and configured as an online service capable of receiving new multi-dimensional feature vectors of the business entity for real-time risk scoring.

7. The method for analyzing rural economic data based on artificial intelligence according to claim 6, characterized in that, It also includes a time-series evolution analysis of the aforementioned rural industry relationship map to generate a regional economic development trend report, including: The rural economic behavior fact tables for different time segments are extracted from the distributed storage array on a quarterly basis, and historical industrial relationship sub-graphs for each time segment are constructed respectively. A graph isomorphism comparison is performed on the historical industry relationship subgraphs of two adjacent time sections to identify newly added economic entity nodes, disappeared economic entity nodes, and edges whose edge weights have changed. The identified graph structure changes are quantified into an industry expansion index, an industry decline index, and an industry transfer index. The industry transfer index is obtained by calculating the rate of change of the edges pointing between nodes. The industry expansion index, industry decline index, and industry transfer index are mapped onto the geographic base map of the economic data perception domain to generate a visualized spatial distribution heat map. By integrating the industry expansion index, industry decline index, industry transfer index, and spatial distribution heat map from all time segments and arranging them according to time series, a regional economic development trend report is automatically generated.

8. The method for analyzing rural economic data based on artificial intelligence according to claim 7, characterized in that, The process of performing graph isomorphism comparison on the historical industry relationship sub-graphs of two adjacent time segments to identify newly added economic entity nodes, disappeared economic entity nodes, and edges with changed edge weights includes: Establish a node mapping table for the historical industry relationship sub-graph of two adjacent time sections, and find corresponding node pairs with the same or highly similar names by using a node name similarity matching algorithm; Nodes that could not be found to have a corresponding relationship in the historical industrial relationship sub-graph of the other party's time segment are marked as the newly added economic entity node and the disappeared economic entity node, respectively. For the successfully matched corresponding node pair, compare the absolute value of the difference between the edge weights connecting the two nodes at two time points. If the absolute value of the difference exceeds a preset fluctuation threshold, then mark the connection as an edge whose edge weight has changed. Extract the attribute information of all the newly added economic entity nodes, the disappeared economic entity nodes, and the edges whose edge weights have changed, and summarize them into a graph structure difference record table. The graph structure difference record table is used as the basic data source for calculating the industry expansion index, industry decline index, and industry transfer index.

9. The method for analyzing rural economic data based on artificial intelligence according to claim 8, characterized in that, The construction of the pre-trained economic activity correlation model includes: Collect historical rural economic transaction data covering multiple regions and time periods, clean and standardize the historical rural economic transaction data, and generate a standardized training fact table. The standardized training fact table is subjected to feature engineering to extract a comprehensive feature vector containing transaction amount features, transaction commodity type features, transaction time interval features, and historical behavior features of the transaction subject; Based on the comprehensive feature vector, a continuous sequence of nodes is generated on the historical transaction network graph using a random walk algorithm. The node sequence reflects the potential connection paths between economic entities. Using a negative sampling algorithm, negative sample nodes that are not directly connected to the target node are randomly selected from the historical transaction network graph; Sequence data containing positive and negative sample node pairs are input into a graph embedding model. By optimizing the loss function, the vector representations of positive sample node pairs are made closer in the graph embedding space, while the vector representations of negative sample node pairs are made farther apart. This trains model parameters that can output vector representations of economic entities, ultimately forming the pre-trained economic activity association model.

10. The method for analyzing rural economic data based on artificial intelligence according to claim 9, characterized in that, Run the betweenness centrality operator to calculate the frequency of each node appearing on the shortest path, quantifying the global importance of nodes to information resource control, including: On the rural industry relationship graph, two different nodes are arbitrarily selected as the source node and the target node, respectively, and all shortest paths between the source node and the target node are calculated. Count the number of paths that pass through the current calculated node among all the shortest paths from the source node to the target node, and calculate the proportion of the number of paths that pass through the current calculated node to the total number of all shortest paths; In the rural industry relationship graph, all source nodes and target nodes are paired and combined, the proportion of the number of paths passing through the currently calculated node to the total number of shortest paths is repeatedly calculated, and the calculation results of all pairing combinations are summed. Divide the summation result by the total number of node pairings in the rural industry relationship graph to obtain the original betweenness centrality value of the currently calculated node; The original betweenness centrality values ​​of all nodes in the rural industry relationship graph are normalized so that the betweenness centrality values ​​of all nodes fall within the range of zero to one. The normalized betweenness centrality values ​​are used to quantify the global importance of nodes in the control of information resources.

Images