Circulation and monitoring method and system based on AI agricultural product knowledge graph

Abstract

The present invention belongs to the field of intelligent circulation of agricultural products, and in particular, relates to a circulation and monitoring method and system based on an AI agricultural product knowledge graph. The method firstly integrates public data of agricultural products with internal business data to construct a multi-dimensional unified agricultural product information format. Secondly, entity recognition and relationship extraction technology are used to construct an agricultural product knowledge graph and update it in real time. Thirdly, the circulation data of agricultural products is monitored in real time in combination with Internet of Things devices to realize dynamic updating of the knowledge graph. Fourthly, the knowledge graph is used to construct an agricultural product circulation path model, and the key indicators of the supply chain are monitored through a graph analysis algorithm to realize abnormal warning and visual display. Finally, based on the updated knowledge graph and real-time monitoring data, a deep learning recommendation algorithm is used to push matching business opportunities to users, and the recommendation model is optimized through user feedback. The present invention effectively improves the circulation efficiency of agricultural products and the accuracy of business opportunity monitoring, and provides users with more accurate business opportunity information.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent circulation of agricultural products, and specifically to a circulation and monitoring method and system based on AI agricultural product knowledge graph. Background Art

[0002] With the rapid development of modern agriculture and the deep integration of information technology, the circulation and commercial activities of agricultural products are becoming increasingly complex and data-intensive. Traditional agricultural product circulation management and business opportunity mining mainly rely on manual experience, limited data sources and simple statistical analysis, which have many limitations and cannot fully meet the needs of the current digital transformation of the agricultural industry. 1) Existing agricultural product circulation data are often scattered in multiple independent information systems, platforms or institutions, forming data islands, which makes it difficult to gather and share information comprehensively and timely. 2) Existing agricultural product circulation data analysis methods mostly stay at the surface level and fail to make full use of advanced knowledge representation and reasoning technologies such as knowledge graphs to conduct in-depth mining and systematic organization of entities such as agricultural products, production areas, circulation channels, consumers and their complex relationships. This limits the deep understanding and forward-looking prediction of agricultural product circulation patterns, market laws, and business opportunity clues. 3) At present, the monitoring of agricultural product circulation status mainly relies on regular reports, manual inspections and other means, lacking real-time and dynamic data collection and analysis capabilities. When faced with emergencies such as agricultural product price fluctuations, supply chain disruptions, and changes in market demand, it is difficult to respond quickly and take effective countermeasures.

[0003] For example, a Chinese patent with publication number CN104732435A discloses a system and method for matching supply and demand of agricultural products. It uses model prediction, production monitoring, network information capture and other methods to predict in advance the time and quantity of agricultural products in each production area to be put on the market; it predicts in advance the consumption demand for various types of agricultural products in various places from aspects such as population size, income level, and consumption habits; it uses big data technology to judge and predict the market price of agricultural products from aspects such as monitoring data and network data, and matches the supply and demand of agricultural products in advance based on factors such as the market period of agricultural products, shelf life, transportation and preservation conditions, and prices at the production and sales locations.

[0004] For example, a Chinese patent with publication number CN109376909A discloses an agricultural product monitoring and early warning model system, including a data sorting module, a statistical analysis module, a prediction module, an early warning module, a visual display module and a system management module; the data sorting module obtains big data for pre-processing, centralized storage and calculation, the statistical analysis module performs general and multi-dimensional statistics on the data to form a statistical analysis report; the prediction module models, analyzes and predicts the statistical analysis report to obtain the prediction result; the early warning module determines whether the prediction result reaches the warning threshold, and issues an early warning for the prediction result that reaches the warning threshold; the visual display module performs visual display of the prediction and early warning information on the large screen; the system management module manages each module.

[0005] The above existing technologies all have the following problems: 1) The scattered data circulation forms data islands, which makes it difficult to comprehensively and timely gather and share information; 2) The existing technologies fail to make full use of advanced knowledge representation and reasoning technologies such as knowledge graphs to deeply mine and systematically organize entities such as agricultural products, production areas, distribution channels, markets, and their complex relationships, which limits the in-depth understanding and forward-looking prediction of agricultural product circulation patterns, market laws, and business opportunities. In order to solve the above problems, the present invention provides a circulation and monitoring method and system based on AI agricultural product knowledge graphs. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a circulation and monitoring method and system based on AI agricultural product knowledge graph. The method first constructs a multi-dimensional unified agricultural product information format by integrating public agricultural product data and internal business data. Secondly, entity recognition and relationship extraction technology are used to construct an agricultural product knowledge graph and update it in real time. Thirdly, the Internet of Things devices are combined to monitor the agricultural product circulation data in real time to realize the dynamic update of the knowledge graph. Fourthly, the knowledge graph is used to construct an agricultural product circulation path model, and the key indicators of the supply chain are monitored through a graph analysis algorithm to achieve abnormal warning and visual display. Finally, based on the updated knowledge graph and real-time monitoring data, a deep learning recommendation algorithm is used to push matching business opportunities to users, and the recommendation model is optimized through user feedback. The present invention effectively improves the efficiency of agricultural product circulation and the accuracy of business opportunity monitoring, and provides users with more accurate business opportunity information.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The circulation and monitoring methods based on AI agricultural product knowledge graph include:

[0009] S1. Collect and integrate various public data and business data of agricultural products, obtain multi-dimensional agricultural product data, and perform structured preprocessing on the collected multi-dimensional agricultural product data so that the multi-dimensional agricultural product data maintains a unified format;

[0010] S2. Introduce deep learning algorithms to extract keywords and corresponding relationship attributes from the multi-dimensional agricultural product data after structured processing, use the obtained agricultural product entity keywords and corresponding attribute relationships to construct an agricultural product triple information chain, input the constructed agricultural product triple information chain into the graph database, obtain the agricultural product knowledge graph, and update the constructed knowledge graph in real time or regularly;

[0011] S3. Use IoT devices to monitor different types of agricultural product circulation data in real time, clean and process the monitored agricultural product circulation data, perform correlation analysis on the processed agricultural product circulation data and the agricultural product knowledge graph, obtain correlation analysis results, and update the obtained correlation analysis results to the corresponding agricultural product category entity node through the query interface of the graph database;

[0012] S4. Use the multi-dimensional agricultural product data and market circulation data in the agricultural product knowledge graph to build a complete circulation path graph model of agricultural products from the place of production to the consumer end, optimize the circulation path graph model through the graph analysis algorithm, find the optimal circulation path, and conduct real-time monitoring, abnormal warning and visualization of key indicators between supply chains in the circulation path model;

[0013] S5: Based on the updated knowledge graph database in S3 and the key indicators between supply chains monitored in real time in S4, the deep learning recommendation algorithm is used to push monitoring information that matches the business needs of users, and the recommendation model is continuously optimized by configuring the user feedback mechanism.

[0014] Specifically, the specific steps of constructing the agricultural product knowledge graph in S2 include:

[0015] S201, save the structured preprocessed multi-dimensional agricultural product data in text form, and use the BIO annotation method and the relationship attribute annotation method to perform entity annotation and entity-attribute annotation on the saved multi-dimensional agricultural product data;

[0016] S202: Construct a joint extraction model of fusion similarity entity relations, and convert the entity text sequence labeled BIO and entity text sequences annotated with entity-attributes , and simultaneously input into the Bert encoder layer 1 and Bert encoder layer 2 in the fusion similarity entity relationship joint extraction model, and output the corresponding character node feature vector and the relationship node feature vector ,in Indicates N tagged entity words, Indicates M Entity-attribute annotated with relation type, N Represents the number of entity words in the input text sequence, M Represents the number of entity-attributes in the input text sequence;

[0017] S203: Setting relationship similarity threshold , the acquired relationship node feature vector Input to the similarity measurement layer to calculate the relationship node features Character Node and Similarity ,like Greater than or equal to the similarity threshold , then keep the current node features ,like Less than the similarity threshold , then count the character nodes Character node feature vector The similarity of each character node feature in and take the character node The similarity is the largest and exceeds the similarity threshold Character node feature replacement , get new relationship node features ;in Represents the feature vector of the relationship node The k-th relationship node feature in and Represents relationship node features The i-th and j-th character node features in ;

[0018] Specifically, the specific steps of constructing the agricultural product knowledge graph in S2 also include:

[0019] S204, character node feature vector And the replaced relationship node feature vector Input to the heterogeneous graph attention layer to obtain the character node feature vector of the fused relationship feature ,Will Input to the entity recognition layer and output entity words;

[0020] S205, will and After cascading, it is input into the relation extraction layer, and the extracted entity-attribute is output. The extracted entity-attribute is input into the attribute matching layer, and entities with the same attributes are connected through attribute relations to obtain triple information chains;

[0021] S206. Input the obtained triple information chain into the graph database to obtain the agricultural product knowledge graph library.

[0022] Specifically, the specific steps for S3 to conduct association analysis between the processed circulation data and the agricultural product knowledge graph include:

[0023] S301, inputting the multi-dimensional agricultural product data stored in the agricultural product knowledge graph library and the pre-processed agricultural product circulation data into the association analysis algorithm to calculate the degree of association between different agricultural products;

[0024] S302, constructing triples using the calculated association degrees and corresponding agricultural products of different categories, and updating the constructed triples into the agricultural product knowledge graph through the agricultural product category attributes;

[0025] Specifically, the circulation path model in S4 configures a graph path optimization strategy to obtain the optimal circulation path. The specific steps include:

[0026] S401, using the agricultural product circulation data stored in the knowledge graph after dynamic update of S3 and combining with graph theory algorithm, construct an agricultural product circulation path graph model, where the nodes in the circulation path graph model are entities in the agricultural product circulation supply chain, and the nodes u To Node v The marginal variables between them are measured by setting the circulation speed, circulation efficiency, circulation cost, and supply-demand matching indicators;

[0027] S402: Based on the constructed agricultural product circulation path graph model, a graph analysis algorithm is used to construct a graph path optimization function, so that the obtained agricultural products can achieve the minimum circulation cost and the maximum benefit among all nodes in the entire circulation path, wherein the graph path optimization function is specifically:

[0028]

[0029] in, represents the total cost of the nth agricultural product in the circulation process at the current time t, Indicates the nodes in the overall circulation path of the nth agricultural product u and nodes v A side composed of E Represents the set of edges composed of all nodes u and v in the overall circulation path; Indicates that the nth agricultural product is at the node u and nodes v The unit cost of circulation between Indicates that the nth agricultural product is at the node u and nodes v The flow between Indicates n Planting agricultural products at nodes u and nodes v The time factor of circulation between Indicates n Agricultural products in the current t Market saturation at the pre-determined target endpoint; Indicates that all types of agricultural products are at the node u and nodes v The average circulation time between the two; Indicates nThe optimal target time for the circulation of agricultural products between nodes u and v, Indicates n The market saturation of the agricultural product at the end of the circulation path at the current moment, Indicates n The market supply of the agricultural product at the end of the circulation path at the current moment, Indicates the constraint value of the corresponding constraint function, which is set by market specific data; Indicates n Planting agricultural products at nodes u and nodes v The maximum value that can be stored; Indicates n Planting agricultural products at nodes u and nodes v Unit budget costs for circulation between;

[0030] S403. Solve the constructed graph path optimization function through the particle swarm algorithm optimized by the BP neural network to solve the optimal path node for each agricultural product to reach the target destination and the corresponding optimal total cost value for circulation, and sort the solved agricultural product types in descending order according to circulation cost and market saturation to obtain the corresponding agricultural product market matching ranking table, and save the obtained optimal path for each agricultural product to reach the target destination, the corresponding optimal total cost value for circulation and the market saturation through the corresponding agricultural product category attributes to the corresponding agricultural product attribute node in the agricultural product knowledge graph library.

[0031] Specifically, the specific steps of pushing monitoring information matching the business needs of the user in S5 include:

[0032] S501, collecting the agricultural product type business currently engaged in or expected to be engaged in by the target point user, the user's geographical location information and the corresponding budget information, and pre-processing the collected information;

[0033] S502, constructing a matching recommendation model based on the Bert algorithm, inputting the agricultural product category business, geographic location information, and budget data of the pre-processed target point user and the agricultural product market matching ranking table data set saved in the knowledge graph in S403 into the constructed matching recommendation model for training, and obtaining a trained matching recommendation model;

[0034] S503, deploying the obtained matching recommendation model to the query interface of the agricultural product knowledge graph library, when user A inputs his address and the agricultural product business he is engaged in, the corresponding agricultural product market matching ranking table is obtained through the matching algorithm; the agricultural product market matching ranking table includes but is not limited to the name of the agricultural product, the origin of the agricultural product, the target location, the optimal circulation cost from the origin to the target location, the market saturation of the agricultural product corresponding to the user's area, and the related agricultural products of the user's corresponding agricultural product;

[0035] Specifically, the specific steps of pushing monitoring information matching the business needs of the user in S5 also include:

[0036] S504, setting a matching index threshold, calculating a matching index based on the obtained optimal circulation cost of agricultural products, the market saturation of the corresponding user area, the business matching degree of the corresponding agricultural products and the business engaged in by the user, and filtering the obtained agricultural product market matching degree ranking table, when the matching index corresponding to the nth agricultural product is greater than the matching index threshold, recommending the corresponding nth agricultural product to the user, when the number of agricultural products greater than the matching index threshold exceeds one, arranging the recommended agricultural products in descending order according to the matching index, forming an agricultural product user matching degree ranking table, and recommending the agricultural product user matching degree ranking table to the user;

[0037] S505, when the user does not search for the corresponding agricultural product at the current moment, the agricultural product user matching ranking table or the single matching agricultural product formed in S504 is pushed to the corresponding user through a fixed or mobile terminal based on the user's historical search records combined with an automatic push algorithm;

[0038] S506. Set a push threshold for the correlation degree of associated products. Based on the names of the agricultural products pushed manually or automatically in S504 and S505 and the correlation degrees of different agricultural products calculated in S302, when the correlation degree between the associated agricultural products and the pushed agricultural products is higher than the push threshold for the correlation degree, the associated agricultural products and the agricultural products pushed in S504 and S505 are pushed to the corresponding users simultaneously through fixed or mobile terminals.

[0039] The circulation and monitoring system based on AI agricultural product knowledge graph includes: data collection and processing module, knowledge graph module;

[0040] The data acquisition and processing module includes a data collection unit and an IoT data unit; the knowledge graph module includes a knowledge graph generation unit;

[0041] The data collection unit is used to integrate various types of agricultural products' public data sources and business data to obtain multi-dimensional agricultural products data. The Internet of Things data unit is used to monitor and collect different categories of agricultural products' circulation data in real time through Internet of Things devices. The knowledge graph generation unit is used to build and update the agricultural products' knowledge graph library based on multi-dimensional agricultural products data and different categories of agricultural products' circulation data through the configured graph database and natural language algorithm.

[0042] Specifically, the circulation and monitoring system further includes: a circulation process monitoring module and a monitoring recommendation module; the circulation process monitoring module includes a circulation path modeling unit and a circulation state monitoring unit; the monitoring recommendation module includes a matching and mining unit and a personalized recommendation unit;

[0043] The circulation path modeling unit is used to use different categories of agricultural product circulation data in the knowledge graph to build a complete circulation path model of agricultural products from the place of production to the consumer end; the circulation status monitoring unit is used to apply graph analysis algorithms to monitor the circulation speed, efficiency, and cost indicators of agricultural products, and set thresholds or early warning rules to make real-time corrections and adjustments to key indicators between supply chains in the circulation path model; the agricultural product matching unit is used to calculate the agricultural product market matching ranking table and the agricultural product user matching ranking table based on the constructed knowledge graph and the user's local market saturation; the personalized recommendation unit is used to push the corresponding searched agricultural products and related products to users based on the calculated agricultural product user matching ranking table and the degree of correlation between different agricultural products, and continuously optimize the recommendation model.

[0044] A computer-readable storage medium having computer instructions stored thereon, which, when executed, executes a circulation and monitoring method based on an AI agricultural product knowledge graph.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] In view of the deficiencies of the prior art, the present invention realizes the comprehensive aggregation and sharing of data resources by integrating multi-dimensional agricultural product data and internal business data, and constructs an agricultural product knowledge graph by introducing entity recognition and relationship extraction technology. It not only deeply mines the complex relationships between entities such as agricultural products, production areas, distribution channels, and markets, but also makes these relationships more intuitive and easy to understand through the visualization of the knowledge graph. In addition, the Internet of Things devices are used to monitor the agricultural product circulation data in real time, and the association analysis is performed with the knowledge graph, which realizes the dynamic update of the knowledge graph, and enables the knowledge graph to reflect market dynamics in real time, thereby improving the response speed to market changes.

[0047] The present invention uses the monitored agricultural product circulation data to construct an agricultural product circulation path graph model and applies the graph path optimization strategy, thereby accurately determining the optimal circulation path of agricultural products from the place of production to the consumer end, minimizing the circulation cost and maximizing the benefit. At the same time, the present invention uses a deep learning recommendation algorithm in combination with user needs and agricultural product market data to calculate the user matching degree, accurately pushes business opportunity information that is highly matched with their business needs to users, and improves the efficiency of discovering business opportunities. In addition, the present invention further enriches the user's choices and improves the user experience by pushing related products. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a circulation and monitoring method based on AI agricultural product knowledge graph according to Embodiment 1 of the present invention;

[0049] Figure 2 This is a flowchart of agricultural product triple information extraction and knowledge graph construction in Example 1 of the present invention;

[0050] Figure 3 This is a module architecture diagram of the circulation and monitoring system based on the AI ​​agricultural product knowledge graph according to Example 2 of the present invention. DETAILED DESCRIPTION

[0051] Example 1

[0052] See also Figure 1 , an embodiment provided by the present invention: a circulation and monitoring method based on AI agricultural product knowledge graph, comprising the following steps:

[0053] S1. Collect and integrate various public data and business data of agricultural products, obtain multi-dimensional agricultural product data, and perform structured preprocessing on the collected multi-dimensional agricultural product data so that the multi-dimensional agricultural product data maintain a unified format; wherein, the multi-dimensional agricultural product data includes but is not limited to external macro data and internal micro data of agricultural products; external macro data includes but is not limited to: agricultural product name, origin, variety, price, sales volume, and distribution channel; internal micro data includes but is not limited to: agricultural product growth or production cycle, quality grade, nutritional value, consumer preference, user location, and user location market saturation; wherein structured preprocessing is achieved by textualizing the collected data;

[0054] S2 introduces a deep learning algorithm to extract keywords and corresponding relationship attributes from the multi-dimensional agricultural product data after structured processing, and constructs an agricultural product triple information chain using the obtained agricultural product entity keywords and corresponding attribute relationships. The constructed agricultural product triple information chain is input into the graph database to obtain the agricultural product knowledge graph, and the constructed knowledge graph is updated in real time or regularly.

[0055] For further information, see Figure 2 In this step, we construct a knowledge graph of agricultural products and configure a deep entity and relationship extraction strategy based on the Bert pre-training model. The specific steps include:

[0056] S201, save the structured preprocessed multi-dimensional agricultural product data in text form, and use the BIO annotation method and the relationship attribute annotation method to perform entity annotation and entity-attribute annotation on the saved multi-dimensional agricultural product data;

[0057] S202: Construct a joint extraction model of fusion similarity entity relations, and convert the entity text sequence labeled BIO and entity text sequences annotated with entity-attributes , and simultaneously input into the Bert encoder layer 1 and Bert encoder layer 2 in the fusion similarity entity relationship joint extraction model, and output the corresponding character node feature vector and the relationship node feature vector ,in Indicates N tagged entity words, Indicates M Entity-attribute annotated with relation type, N Represents the number of entity words in the input text sequence, M Indicates the number of entity-attributes in the input text sequence; further, the Bert encoder layer 1 and the Bert encoder layer 2 in this embodiment have the same structure, and the Chinese pre-trained BERT model used encodes the multi-dimensional agricultural product text data characters into vectors in the embedding layer, and the Chinese pre-trained BERT model is fine-tuned and trained using the collected agricultural product historical text data set before use, and then the fine-tuned Chinese pre-trained BERT model is implanted into the Bert encoder layer 1 and the Bert encoder layer 2;

[0058] S203: Setting relationship similarity threshold , the acquired relationship node feature vector Input to the similarity measurement layer to calculate the relationship node features Character Node and Similarity ,like Greater than or equal to the similarity threshold , then keep the current node features ,like Less than the similarity threshold , then count the character nodes Character node feature vector The similarity of each character node feature in and take the character node The similarity is the largest and exceeds the similarity threshold Character node feature replacement , get new relationship node features ;in Represents the feature vector of the relationship node The k-th relationship node feature in and Represents relationship node features The i-th and j-th character node features in the ; Further, the similarity in this embodiment is calculated using the cosine similarity formula, wherein the cosine similarity is a method for measuring the similarity of the directions of two vectors. The larger the value, the more similar the directions of the two vectors; the smaller the value, the less similar the directions of the two vectors. This similarity calculation can reduce the phenomenon of small number of entity recognition and wrong recognition in the entity recognition results; In this embodiment, the similarity threshold is set to 0.95;

[0059] S204, character node feature vector And the replaced relationship node feature vector Input to the heterogeneous graph attention layer to obtain the character node feature vector of the fused relationship feature ,Will Input to the entity recognition layer, output entity words; In this embodiment, the heterogeneous graph attention layer adopts a collaborative attention network to regard character nodes as neighbor nodes of relationship nodes, and updates the representation of characters and relationship nodes through the heterogeneous graph attention network message passing mechanism; Specifically, the node update representation is combined by a bilinear function softmax Function implementation, which calculates the attention scores of character nodes and relationship nodes 1; similarly, for each relationship node, its attention score to all character nodes is calculated 2; so that each node can "pay attention" to the node that is most relevant to it;

[0060] S205, will and After cascading, it is input into the relationship extraction layer, and the extracted entity-attribute is output. The extracted entity-attribute is input into the attribute matching layer, and entities with the same attributes are connected through attribute relations to obtain an entity 1-attribute-entity 2 triple information chain; in this embodiment, three fully connected layers are used in the relationship extraction layer to output the entity-attribute; the constructed triple information chain includes but is not limited to corn-growth cycle-summer, apple-origin-Shandong Province, grape-quality grade-excellent grade, grape-distribution channel-e-commerce platform, citrus-nutritional value-rich in vitamin C, rice-variety-japonica rice, potato-growth cycle-short-cycle crop;

[0061] S206. Input the obtained triple information chain into the graph database to obtain the agricultural product knowledge graph library.

[0062] S3. Use IoT devices to monitor different types of agricultural product circulation data in real time, clean and process the monitored agricultural product circulation data, perform correlation analysis on the processed agricultural product circulation data and the agricultural product knowledge graph, obtain the correlation analysis results, and update the obtained correlation analysis results to the corresponding agricultural product category entity nodes through the query interface of the graph database, and dynamically update the agricultural product knowledge graph;

[0063] Market circulation data include production, storage, circulation, transaction records and market saturation data; agricultural product category data include but are not limited to: food crops, cash crops, vegetable crops, livestock products and fruit crops; food crops include rice, wheat, corn, barley, oats, rye, sorghum and millet; cash crops include oil crops (such as soybeans and rapeseed), fiber crops (such as cotton), sugar crops (such as sugarcane and sugar beets), beverage crops (such as tea and coffee) and medicinal crops; vegetable crops include cabbage, tomato, cucumber and potato; livestock products include beef, pork, poultry, fresh milk, cheese and eggs; fruit crops include apples, pears, citrus, grapes and bananas;

[0064] Furthermore, the specific steps of S3 to conduct correlation analysis between the processed circulation data and the agricultural product knowledge graph include:

[0065] S301, input the multi-dimensional agricultural product data stored in the agricultural product knowledge graph library and the pre-processed agricultural product circulation data into the association analysis algorithm to calculate the degree of association between different agricultural products; the association analysis algorithm used in this embodiment includes but is not limited to the Apriori algorithm or the FP-growth algorithm;

[0066] S302, constructing triples using the calculated association degrees and corresponding different categories of agricultural products, and updating the constructed triples to the agricultural product knowledge graph through the agricultural product category attributes; further, the triples of this step are specifically apple-association degree-grape, apple-association degree-orange, ginger-association degree-garlic sprout;

[0067] S4. Use the multi-dimensional agricultural product data and market circulation data in the agricultural product knowledge graph to build a complete circulation path graph model of agricultural products from the place of production to the consumer end, optimize the circulation path graph model through the graph analysis algorithm, find the optimal circulation path, and conduct real-time monitoring, abnormal warning and visualization of key indicators between supply chains in the circulation path model;

[0068] Furthermore, the circulation path model in S4 configures a graph path optimization strategy to obtain the optimal circulation path. The specific steps include:

[0069] S401, using the agricultural product data stored in the knowledge graph after dynamic update of S3 and combining with graph theory technology, construct an agricultural product circulation path graph model, where the nodes in the circulation path graph model are entities in the agricultural product circulation supply chain, and the nodes u To Node v The edge variables between them are measured by setting the circulation speed, circulation efficiency, circulation cost, and supply-demand matching indicators; the nodes in this embodiment include but are not limited to origin, distributors, logistics service providers, and trading platforms; each node is accompanied by geographical coordinates, entity type, credit rating, processing capacity and other attributes; the circulation speed represents the time efficiency of agricultural products from one node to the next node per unit time; the circulation efficiency represents the loss and delay of agricultural products in circulation; the circulation cost represents the accumulated cost between all nodes in the entire circulation path, including but not limited to logistics costs, storage costs, and transaction costs; the supply-demand matching represents the satisfaction of the market demand for specific agricultural products;

[0070] S402: Based on the constructed agricultural product circulation path graph model, a graph analysis algorithm is used to construct a graph path optimization function, so that the obtained agricultural products can achieve the minimum circulation cost and the maximum benefit among all nodes in the entire circulation path, wherein the graph path optimization function is specifically:

[0071]

[0072] in, represents the total cost of the nth agricultural product in the circulation process at the current time t, Indicates the nodes in the overall circulation path of the nth agricultural product u and nodes v A side composed of E Represents the set of edges composed of all nodes u and v in the overall circulation path; Indicates that the nth agricultural product is at the node u and nodes v The unit cost of circulation between Indicates that the nth agricultural product is at the node u and nodes v The flow between Indicates n Planting agricultural products at nodes u and nodes v The time factor of circulation between Indicates n Agricultural products in the current t Market saturation at the pre-determined target endpoint; Indicates that all types of agricultural products are at the node u and nodes v The average circulation time between the two; Indicates n The optimal target time for the circulation of agricultural products between nodes u and v, Indicates n The market saturation of the agricultural product at the end of the circulation path at the current moment, Indicates n The market supply of the agricultural product at the end of the circulation path at the current moment, Indicates the constraint value of the corresponding constraint function, which is set by market specific data; Indicates n Planting agricultural products at nodes u and nodes v The maximum value that can be stored; Indicates n Planting agricultural products at nodes u and nodes v Unit budget costs for circulation between;

[0073] S403, using the particle swarm algorithm optimized by the BP neural network, solving the constructed graph path optimization function, solving the optimal path node for each agricultural product to reach the target destination and the corresponding optimal total circulation cost value, and sorting the solved agricultural product types in descending order according to the circulation cost and market saturation, obtaining the corresponding agricultural product market matching ranking table, and connecting and saving the obtained optimal path for each agricultural product to reach the target destination, the corresponding optimal total circulation cost value and the market saturation through the agricultural product category attribute through the knowledge graph interface; further, the specific steps of solving the constructed graph path optimization function by using the constructed particle swarm algorithm include:

[0074] 1) Randomly initialize the particle swarm, each particle represents a configuration of the agricultural product circulation path. Each particle consists of multiple dimensions, corresponding to the node sequence on the agricultural product circulation path and the circulation speed, circulation efficiency and other decision variables at each pair of nodes (u, v). At the same time, each particle is given an initial position (i.e., a circulation path solution) and speed;

[0075] 2) Set the key parameters of the algorithm, including the individual best position of the particle (pBest), the group best position (gBest), the inertia weight (w), the cognitive coefficient (c1), and the social coefficient (c2);

[0076] 3) Update the speed of each particle according to the current particle position, the individual optimal position, the group optimal position, and the above parameters, and move each particle to a new position based on the updated speed;

[0077] 4) Calculate the objective function value corresponding to the new position of each particle (i.e., the objective function related to circulation cost and market saturation), and compare it with the individual best position and the group best position. If it is better, update it;

[0078] 5) Check whether the preset number of iterations or algorithm convergence conditions are reached. If the termination conditions are met, stop the iteration; otherwise, repeat steps 2)-5) to continuously iterate the set key parameters until the algorithm convergence conditions are met or the preset number of iterations is reached;

[0079] 6) After the iteration is completed, the optimal path node, the corresponding optimal total circulation cost value and market saturation of each agricultural product are determined according to the group's best position (gBest), and sorted as required.

[0080] S5. Based on the updated knowledge graph database in S3 and the key indicators between supply chains monitored in real time in S4, the deep learning recommendation algorithm is used to push monitoring information matching the business needs of the users, and the matching monitoring information is used to continuously optimize the recommendation model through the configured user feedback mechanism; the monitoring information in this embodiment is the monitored business opportunity information;

[0081] Furthermore, the specific steps of pushing monitoring information matching the business needs of the user in S5 include:

[0082] S501, collecting the agricultural product type business currently engaged in or expected to be engaged in by the target point user, the user's geographical location information and the corresponding budget information, and pre-processing the collected information;

[0083] S502, constructing a matching recommendation model based on the Bert algorithm, inputting the pre-processed agricultural product types and businesses, geographic location information, and budget data of the target point users and the agricultural product market matching ranking table data set saved in the knowledge graph in S403 into the constructed matching recommendation model for training, and obtaining the trained matching recommendation model; further, the matching recommendation model constructed by the Bert algorithm in this embodiment includes embedding layer 1 and embedding layer 2, feature extraction layer 1 and feature extraction layer 2, interaction layer, pooling layer 1 and pooling layer 2, and prediction layer, embedding layer 1 and embedding layer 2 have the same structure, feature extraction layer 1 and feature extraction layer 2 have the same structure, and pooling layer 1 and pooling layer 2 have the same structure, wherein, in this embodiment, word2vec is used to construct embedding layer 1 With the embedding layer 2, the input text sequence is sequence-encoded, and the position information tensor is added to the encoded sequence tensor; this embodiment uses the self-encoding layer in the transformer model to construct the feature extraction layer 1 and the feature extraction layer 2, extracts high-dimensional information from the input sequence tensor with added position information, uses the cross-attention network to construct the interaction layer, and uses the cross-attention network to cross-calculate the high-dimensional text information extracted by the feature extraction layer 1 and the feature extraction layer 2 to complete the sharing of high-dimensional text information; the pooling layer 1 and the pooling layer 2 use the global average pooling layer with a kernel size of 2 to input the dimensional text feature vector and perform dimensionality reduction operations to reduce the redundancy of the input text; the prediction layer includes two forward networks and a softmax activation function;

[0084] S503, deploying the obtained matching recommendation model to the query interface of the agricultural product knowledge graph library, when user A inputs his address and the agricultural product business he is engaged in, the corresponding agricultural product market matching ranking table is obtained through the matching algorithm; the agricultural product market matching ranking table includes but is not limited to the name of the agricultural product, the origin of the agricultural product, the target location, the optimal circulation cost from the origin to the target location, the market saturation of the agricultural product corresponding to the user's area, and the related agricultural products of the user's corresponding agricultural product;

[0085] S504, setting a matching index threshold, and calculating a matching index based on the obtained optimal circulation cost of agricultural products, the market saturation of the corresponding user area, the business matching degree of the corresponding agricultural products and the user's business to filter the obtained agricultural product market matching degree ranking table; when the matching index corresponding to the nth agricultural product is greater than the matching index threshold, recommend the corresponding nth agricultural product to the user; when the number of agricultural products greater than the matching index threshold exceeds one, arrange the recommended agricultural products in descending order according to the matching index to form an agricultural product user matching degree ranking table, and recommend the agricultural product user matching degree ranking table to the user; further, the agricultural product user matching degree ranking table includes but is not limited to product name, agricultural product origin, target destination, optimal circulation cost from origin to target destination, market saturation of agricultural products corresponding to the user area, and business matching degree of the corresponding agricultural products and the user's business.

[0086] S505, when the user does not search for the corresponding agricultural product at the current moment, the agricultural product user matching ranking table or the single matching agricultural product formed in S504 is pushed to the corresponding user through a fixed or mobile terminal based on the user's historical search records combined with an automatic push algorithm;

[0087] S506. Set a push threshold for the correlation degree of associated products. Based on the names of the agricultural products pushed manually or automatically in S504 and S505 and the correlation degrees of different agricultural products calculated in S302, when the correlation degree between the associated agricultural products and the pushed agricultural products is higher than the push threshold for the correlation degree, the associated agricultural products and the agricultural products pushed in S504 and S505 are pushed to the corresponding users simultaneously through fixed or mobile terminals.

[0088] Example 2

[0089] See also Figure 3 , another embodiment provided by the present invention: a circulation and monitoring system based on AI agricultural product knowledge graph, comprising: a data collection and processing module, a knowledge graph module, a circulation process monitoring module, a monitoring recommendation module and a user interaction and feedback module;

[0090] The data collection and processing module is used to collect and preprocess various agricultural products' public data, business data and agricultural products' circulation data; the knowledge graph module is used to extract agricultural products' text data and build and manage agricultural products' knowledge graphs; the circulation process monitoring module is used to construct circulation paths through knowledge graphs and graph analysis technologies, and to monitor and visualize the constructed circulation paths in real time; the monitoring and recommendation module is used to conduct agricultural products market association matching analysis based on knowledge graphs and recommendation algorithms, and to provide personalized business opportunity recommendations based on user characteristics; the user interaction and feedback module is used to provide a friendly user interface to facilitate users to interact with the system and obtain the required information. At the same time, it collects user feedback and suggestions for system optimization and improvement.

[0091] The data acquisition and processing module includes a data collection unit, an IoT data unit, and a data preprocessing unit; the knowledge graph module includes a knowledge graph generation unit and a data update unit; the circulation process monitoring module includes a circulation path modeling unit, a circulation status monitoring unit, and a visualization unit; the monitoring recommendation module includes a matching and mining unit and a personalized recommendation unit; the user interaction and feedback module includes a user interaction unit and a user feedback unit;

[0092] The data collection unit is used to integrate various types of agricultural products' public data sources and business data to obtain multi-dimensional agricultural products data; the Internet of Things data unit is used to monitor and collect different categories of agricultural products' circulation data in real time through Internet of Things devices; the data preprocessing unit is used to perform real-time preprocessing on the data collected by the data collection unit and the Internet of Things data unit; the knowledge graph generation unit is used to build and update the agricultural products' knowledge graph library based on the multi-dimensional agricultural products data and different categories of agricultural products' circulation data through the configured graph database and natural language algorithm; the data update unit is used to use the real-time acquired multi-dimensional agricultural products data to update the knowledge graph regularly or in real time to maintain the timeliness and accuracy of the knowledge graph; the circulation path modeling unit is used to use the different categories of agricultural products' circulation data in the knowledge graph to build a complete circulation path model of agricultural products from the place of production to the consumer end; the circulation status monitoring unit is used to apply graph analysis to the circulation status of agricultural products. The analysis algorithm monitors the circulation speed, efficiency and cost indicators of agricultural products, and sets thresholds or early warning rules to make real-time corrections and adjustments to key indicators between supply chains in the circulation path model; the visualization unit is used to display the dynamic changes of the agricultural product circulation process in the form of charts, maps, etc., and provide interactive analysis functions; the agricultural product matching unit is used to calculate the agricultural product market matching ranking table and the agricultural product user matching ranking table based on the constructed knowledge graph and the user's local market saturation; the personalized recommendation unit is used to push the corresponding searched agricultural products and related products to the user based on the calculated agricultural product user matching ranking table and the degree of association between different agricultural products, and continuously optimize the recommendation model; the user interaction unit is used to provide an interactive interface between the user and the system, and support users to query, analyze, and configure system parameters; the user feedback unit is used to collect user feedback and suggestions on the system, and optimize and improve the system.

[0093] Example 3

[0094] Another embodiment provided by the present invention: The specific workflow of the circulation and monitoring system based on the AI ​​agricultural product knowledge graph is as follows:

[0095] First, the data collection unit in the data acquisition and processing module integrates the public data of agricultural products and internal business data to obtain multi-dimensional agricultural product data. At the same time, the Internet of Things data unit collects agricultural product circulation data at various links in real time through the Internet of Things devices. The data preprocessing unit cleans, integrates and formats the data collected by the data collection unit and the Internet of Things data unit to provide a high-quality data basis for subsequent analysis. Second, the knowledge graph generation unit in the knowledge graph module uses the pre-processed multi-dimensional agricultural product data and agricultural product circulation data, combined with the graph database and natural language processing algorithm to build and update the agricultural product knowledge graph library. At the same time, the data update and maintenance unit regularly or in real time updates the knowledge graph to ensure the timeliness and accuracy of the data; third, in the circulation process monitoring module, the circulation path model of agricultural products is constructed according to the agricultural product circulation data in the knowledge graph through the circulation path modeling unit, and the graph analysis algorithm in the circulation status monitoring unit is used to analyze and solve the constructed circulation path model, and the solved circulation indicators are monitored in real time, and the circulation indicators are analyzed according to the set thresholds or warning rules. Real-time adjustments are made, and the monitoring results are displayed in the form of charts or maps through the visualization unit, which is convenient for users to conduct interactive analysis; fourthly, the agricultural product matching unit in the monitoring and recommendation module calculates the agricultural product market matching degree according to the agricultural product knowledge graph library and the user's local market conditions, and generates a ranking table; the personalized recommendation unit pushes personalized agricultural product business opportunities to users based on the matching results and user characteristics; fifthly, the user interaction unit in the user interaction and feedback module provides a friendly user interface to support users to query, analyze and configure system parameters, and the user feedback unit collects users' feedback and suggestions, and continuously optimizes and improves the models in the knowledge graph module, flow process monitoring module and monitoring recommendation module.

[0096] Example 4

[0097] A computer-readable storage medium having computer instructions stored thereon, which, when executed, executes a circulation and monitoring method based on an AI agricultural product knowledge graph.

[0098] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the scope of protection of the purpose of the present invention and the claims, and all of these are within the protection of the present invention.

Claims

1. A circulation and monitoring method based on AI agricultural product knowledge graph, characterized in that the steps include: S1. Collect and integrate various public data and business data of agricultural products, obtain multi-dimensional agricultural product data, and perform structured preprocessing on the collected multi-dimensional agricultural product data so that the multi-dimensional agricultural product data maintains a unified format; S2. Introduce deep learning algorithms to extract keywords and corresponding relationship attributes from the multi-dimensional agricultural product data after structured processing, use the obtained agricultural product entity keywords and corresponding attribute relationships to construct an agricultural product triple information chain, input the constructed agricultural product triple information chain into the graph database, obtain the agricultural product knowledge graph, and update the constructed knowledge graph in real time or regularly; S3. Use IoT devices to monitor different types of agricultural product circulation data in real time, clean and process the monitored agricultural product circulation data, perform correlation analysis on the processed agricultural product circulation data and the agricultural product knowledge graph, obtain correlation analysis results, and update the obtained correlation analysis results to the corresponding agricultural product category entity node through the query interface of the graph database; S4. Use the multi-dimensional agricultural product data and market circulation data in the agricultural product knowledge graph to build a complete circulation path graph model of agricultural products from the place of production to the consumer end, optimize the circulation path graph model through the graph analysis algorithm, find the optimal circulation path, and conduct real-time monitoring, abnormal warning and visualization of key indicators between supply chains in the circulation path model; S5: Based on the updated knowledge graph database in S3 and the key indicators of the supply chain monitored in real time in S4, the deep learning recommendation algorithm is used to push monitoring information that matches the business needs of users, and the matching monitoring information is used to continuously optimize the recommendation model through the configured user feedback mechanism; The specific steps of constructing the agricultural product knowledge graph in S2 include: S201, save the structured preprocessed multi-dimensional agricultural product data in text form, and use the BIO annotation method and the relationship attribute annotation method to perform entity annotation and entity-attribute annotation on the saved multi-dimensional agricultural product data; S202: Construct a joint extraction model of fusion similarity entity relations, and convert the entity text sequence labeled BIO and entity text sequences annotated with entity-attributes , and simultaneously input into the Bert encoder layer 1 and Bert encoder layer 2 in the fusion similarity entity relationship joint extraction model, and output the corresponding character node feature vector and the relationship node feature vector ,in represents the Nth labeled entity word, represents the entity-attribute annotated with the Mth relation type, N represents the number of entity words in the input text sequence, and M represents the number of entity-attributes in the input text sequence; S203: Setting relationship similarity threshold , the acquired relationship node feature vector Input to the similarity measurement layer to calculate the relationship node features Character Node and Similarity ,like Greater than or equal to the similarity threshold , then keep the current node features ,like Less than the similarity threshold , then count the character nodes Character node feature vector The similarity of each character node feature in and take the character node The similarity is the largest and exceeds the similarity threshold Character node feature replacement , get new relationship node features ;in Represents the feature vector of the relationship node The k-th relationship node feature in and Represents relationship node features The i-th and j-th character node features in ; The specific steps of constructing the agricultural product knowledge graph in S2 also include: S204, character node feature vector And the replaced relationship node feature vector Input to the heterogeneous graph attention layer to obtain the character node feature vector of the fused relationship feature ,Will Input to the entity recognition layer and output entity words; S205, will and After cascading, it is input into the relation extraction layer, and the extracted entity-attribute is output. The extracted entity-attribute is input into the attribute matching layer, and entities with the same attributes are connected through attribute relations to obtain triple information chains; S206, inputting the obtained triple information chain into the graph database to obtain the agricultural product knowledge graph library; The specific steps of S3 performing association analysis on the processed circulation data and the agricultural product knowledge graph include: S301, inputting the multi-dimensional agricultural product data stored in the agricultural product knowledge graph library and the pre-processed agricultural product circulation data into the association analysis algorithm to calculate the degree of association between different agricultural products; S302, constructing a triplet using the calculated association degree and the corresponding two different categories of agricultural products, and updating the constructed triplet into the agricultural product knowledge graph through the agricultural product category attribute; The circulation path model in S4 is configured with a graph path optimization strategy for obtaining the optimal circulation path, and the specific steps include: S401, using the agricultural product circulation data stored in the knowledge graph after dynamic update in S3 combined with graph theory algorithms, construct an agricultural product circulation path graph model, where the nodes in the circulation path graph model are entities in the agricultural product circulation supply chain, and the edge variables between node u and node v are measured by setting circulation speed, circulation efficiency, circulation cost, and supply-demand matching indicators; S402: Based on the constructed agricultural product circulation path graph model, a graph analysis algorithm is used to construct a graph path optimization function, so that the obtained agricultural products can achieve the minimum circulation cost and the maximum benefit among all nodes in the entire circulation path, wherein the graph path optimization function is specifically: ； in, represents the total cost of the nth agricultural product in the circulation process at the current time t, represents an edge consisting of node u and node v in the overall circulation path of the nth agricultural product, and E represents the set of edges consisting of all nodes u and node v in the overall circulation path; represents the unit cost of the nth agricultural product flowing between node u and node v, represents the flow of the nth agricultural product between node u and node v, represents the time impact factor of the circulation of the nth agricultural product between node u and node v, It represents the market saturation of the nth agricultural product in the current t market at the preset target end point; It represents the average circulation time of all kinds of agricultural products between node u and node v; represents the optimal target time for the nth agricultural product to flow between node u and node v, represents the market saturation of the nth agricultural product at the end of the circulation path at the current moment, represents the market supply of the nth agricultural product at the end of the circulation path at the current moment, Represents the constraint value of the corresponding constraint function; It represents the maximum value of the nth agricultural product that can be stored at nodes u and v; represents the unit budget cost of the nth agricultural product flowing between node u and node v; S403. Solve the constructed graph path optimization function through the particle swarm algorithm optimized by the BP neural network to solve the optimal path node for each agricultural product to reach the target destination and the corresponding optimal total cost value for circulation, and sort the solved agricultural product types in descending order according to circulation cost and market saturation to obtain the corresponding agricultural product market matching ranking table, and save the obtained optimal path for each agricultural product to reach the target destination, the corresponding optimal total cost value for circulation and the market saturation through the corresponding agricultural product category attributes to the corresponding agricultural product attribute node in the agricultural product knowledge graph library.

2. The circulation and monitoring method based on AI agricultural product knowledge graph according to claim 1 is characterized in that: The specific steps of pushing monitoring information matching the business needs of the user in S5 include: S501, collecting the agricultural product type business currently engaged in or expected to be engaged in by the target point user, the user's geographical location information and the corresponding budget information, and pre-processing the collected information; S502, constructing a matching recommendation model based on the Bert algorithm, inputting the agricultural product category business, geographic location information, and budget data of the pre-processed target point user and the agricultural product market matching ranking table data set saved in the knowledge graph in S403 into the constructed matching recommendation model for training, and obtaining a trained matching recommendation model; S503. Deploy the obtained matching recommendation model to the query interface of the agricultural product knowledge graph library. When user A inputs his address and the agricultural product business he is engaged in, the corresponding agricultural product market matching ranking table is obtained through the matching algorithm; the agricultural product market matching ranking table includes but is not limited to the name of the agricultural product, the origin of the agricultural product, the target location, the optimal circulation cost from the origin to the target location, and the market saturation of the agricultural product corresponding to the user's area.

3. The circulation and monitoring method based on AI agricultural product knowledge graph according to claim 2 is characterized in that: The specific step of pushing monitoring information matching the business needs of the user in S5 also includes: S504, setting a matching index threshold, calculating a matching index based on the obtained optimal circulation cost of agricultural products, the market saturation of the corresponding user area, and the business matching degree between the corresponding agricultural products and the business engaged in by the user, and filtering the obtained agricultural product market matching degree ranking table, when the matching index corresponding to the nth agricultural product is greater than the matching index threshold, recommending the corresponding nth agricultural product to the user, when the number of agricultural products greater than the matching index threshold exceeds one, arranging the recommended agricultural products in descending order according to the matching index, forming an agricultural product user matching degree ranking table, and recommending the agricultural product user matching degree ranking table to the user; S505, when the user does not search for the corresponding agricultural product at the current moment, the agricultural product user matching ranking table or the single matching agricultural product formed in S504 is pushed to the corresponding user through a fixed or mobile terminal based on the user's historical search records combined with an automatic push algorithm; S506. Set a push threshold for the correlation degree of associated products. Based on the names of the agricultural products pushed manually or automatically in S504 and S505 and the correlation degrees of different agricultural products calculated in S302, when the correlation degree between the associated agricultural products and the pushed agricultural products is higher than the push threshold for the correlation degree, the associated agricultural products and the agricultural products pushed in S504 and S505 are pushed to the corresponding users simultaneously through fixed or mobile terminals.

4. A circulation and monitoring system based on AI agricultural product knowledge graph, which is used to implement the circulation and monitoring method based on AI agricultural product knowledge graph as described in any one of claims 1 to 3, characterized in that: include: Data collection and processing module, knowledge graph module; The data acquisition and processing module includes a data collection unit and an Internet of Things data unit; The knowledge graph module includes a knowledge graph generation unit; The data collection unit is used to integrate various agricultural product public data sources and business data to obtain multi-dimensional agricultural product data; The Internet of Things data unit is used to monitor and collect different categories of agricultural product circulation data in real time through Internet of Things devices; the knowledge graph generation unit is used to build and update the agricultural product knowledge graph library based on multi-dimensional agricultural product data and different categories of agricultural product circulation data through the configured graph database and natural language algorithm.

5. The circulation and monitoring system based on AI agricultural product knowledge graph according to claim 4 is characterized in that: The circulation and monitoring system further comprises: a circulation process monitoring module and a monitoring recommendation module; the circulation process monitoring module comprises a circulation path modeling unit and a circulation state monitoring unit; the monitoring recommendation module comprises a matching and mining unit and a personalized recommendation unit; The circulation path modeling unit is used to use different categories of agricultural product circulation data in the knowledge graph to construct a complete circulation path model of agricultural products from the place of production to the consumer end; the circulation status monitoring unit is used to apply the graph analysis algorithm to monitor the circulation speed, efficiency, and cost indicators of agricultural products, and set thresholds or early warning rules to make real-time corrections and adjustments to key indicators between supply chains in the circulation path model; the agricultural product matching unit is used to calculate the agricultural product market matching ranking table and the agricultural product user matching ranking table based on the constructed knowledge graph and the local market saturation of the user; the personalized recommendation unit is used to push the corresponding searched agricultural products and related products to the user based on the calculated agricultural product user matching ranking table and the degree of correlation between different agricultural products, and continuously optimize the recommendation model.

6. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, the circulation and monitoring method based on the AI ​​agricultural product knowledge graph described in any one of claims 1 to 3 is executed.

Images