Commodity relation identification method and device, computer equipment and storage medium

Abstract

The invention discloses a commodity relationship identification method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring commodity related data; performing data cleaning on the commodity related data, and obtaining an operation relationship; encoding the commodity to obtain an encoding result; generating a commodity-level industrial link according to the coding result and the operation relationship; performing network merging and network pruning on the commodity-level industrial link to obtain an industrial chain network; traversing the industry chain network to obtain upstream and downstream relationships and potential features between commodities; and outputting upstream and downstream relationships and potential features among the commodities. By implementing the method provided by the invention, a complex network and deep learning can be effectively fused, so that the recognition accuracy, efficiency and practicability are improved.

Description

Technical Field

[0001] The present invention relates to computers, and more specifically to a commodity relationship identification method, device, computer equipment and storage medium. Background Art

[0002] Commodity relationship identification plays a vital role in improving business efficiency, enhancing user experience and driving sales growth. Existing identification technologies include collaborative filtering recommendation systems, association rule mining, commodity relationship identification based on graph models and recommendation methods based on deep learning, but these technologies all have certain defects.

[0003] First, collaborative filtering is one of the most commonly used recommendation methods and is widely used in e-commerce and streaming services. However, it faces the problem of data sparsity, especially when there is less interaction data between users and products, it is difficult for the model to accurately capture user preferences. In addition, the cold start problem makes the recommendation effect of new users or new products poor due to the lack of historical data. More importantly, collaborative filtering is mainly based on direct user-product interaction, which makes it difficult to identify the deep relationship between products and cannot capture complex product relationships.

[0004] Secondly, association rule mining identifies association rules between commodities by analyzing a large amount of transaction data. It is often used in market basket analysis to help merchants formulate commodity combinations and promotion strategies. However, this method has high computational complexity on large data sets, consumes significant time and space, and may generate a large number of redundant or insignificant rules, increasing the complexity of analysis. In addition, the static characteristics of association rule mining make it difficult to adapt to dynamic data changes and lack a real-time update mechanism.

[0005] Furthermore, the commodity relationship recognition based on the graph model regards commodities and users as graph structures, and uses graph theory and network analysis to identify the relationship between commodities. Although this method can consider the structural relationship between commodities, it has high computational complexity and resource requirements when processing large-scale graph data, making it difficult to apply in real-time systems. In addition, the traditional graph model does not fully combine the feature extraction capabilities of deep learning, which limits the expressiveness and accuracy of the model. At the same time, it has poor adaptability to dynamic changes and high update costs.

[0006] Finally, deep learning-based recommendation methods use deep neural networks for representation learning, which improves the accuracy of recommendations, especially in image, text, and sequence data recommendations. However, these methods mainly focus on feature extraction, do not fully consider the complex network relationships between products, and have difficulty capturing the structured associations between products. The training of deep learning models requires a large amount of data and computing resources, which may lead to inefficiency, and its decision-making process is opaque, lacking explainability and credibility.

[0007] In summary, although the existing technology has made some progress in commodity relationship identification, it generally has the following defects: it fails to effectively combine complex networks and deep learning, which limits the expressiveness and accuracy of the model; the computational efficiency and scalability are insufficient, making it difficult to process large-scale data and meet real-time needs; the dynamic adaptability is poor and it is unable to reflect changes in commodity relationships in a timely manner; the interpretability and transparency are insufficient, making it difficult to provide credible explanations for merchants and users.

[0008] Therefore, it is necessary to design a new method to effectively integrate complex networks and deep learning to improve the accuracy, efficiency and practicality of recognition. Summary of the invention

[0009] The purpose of the present invention is to overcome the defects of the prior art and provide a commodity relationship identification method, device, computer equipment and storage medium.

[0010] To achieve the above object, the present invention adopts the following technical solution: a commodity relationship identification method, comprising: Get product-related data; Clean the data related to the commodities and obtain the business relationship; Encode the product to obtain an encoding result; Generate a commodity-level industrial chain according to the coding results and business relationships; Performing network merging and network pruning on the commodity-level industrial chain to obtain an industrial chain network; Traversing the industrial chain network to obtain upstream and downstream relationships and potential characteristics between commodities; Output the upstream and downstream relationships and potential characteristics between commodities.

[0011] A further technical solution is: cleaning the commodity-related data and obtaining the business relationship includes: The commodity-related data is deduplicated, outliers processed, and standardized, and the business relationship of the processed commodity-related data is analyzed.

[0012] A further technical solution is: the encoding of the commodity to obtain the encoding result includes: Extracting product information from the product-related data using named entity technology; A commodity code is generated according to the commodity information, and the commodity code is mapped to a unified standard format using a connection technology to obtain a coding result.

[0013] A further technical solution is: generating a commodity-level industrial chain according to the coding result and the business relationship includes: A complex network including commodities and enterprises is constructed according to the coding results and business relationships, wherein the coding results and enterprises corresponding to the commodities are used as nodes and the transaction relationships are used as edges; Use the Louvain algorithm to cut the network and identify different products; Performing modal dimensionality reduction on different commodities and enterprises to update the complex network; Evaluate the association paths between different commodities according to the complex network, identify the influencing paths and weights, and obtain path analysis results; According to the path analysis results and the coding results corresponding to different commodities, a new graph is constructed, attributes are calculated, core nodes are extracted, and cluster analysis is performed to obtain a commodity-level industrial chain.

[0014] The further technical solution is: constructing a new graph, calculating attributes, extracting core nodes and clustering analysis based on the path analysis results and the coding results corresponding to different commodities to obtain a commodity-level industrial chain, including: According to the path analysis results and the encoding results corresponding to different commodities, an edge set of a new graph is defined, and the upstream and downstream relationships between commodities are represented as directed edges; Calculate the amount attribute, transaction quantity attribute, and upstream reason attribute of each edge in the new graph and use them as the weight of the edge; Extract key commodity nodes in the new graph through the core decomposition algorithm; The cluster decomposition algorithm is applied to divide the products in the new graph into different clusters to obtain the product-level industrial chain.

[0015] A further technical solution is: the network merging and network pruning of the commodity-level industrial chain to obtain an industrial chain network includes: Standardizing the encoding results of the commodity-level industrial chain to obtain a merged network; The NLU deep learning model is used to perform network pruning on the merged network to obtain the industrial chain network.

[0016] A further technical solution is: traversing the industrial chain network to obtain the upstream and downstream relationships and potential characteristics between commodities includes: The industry chain network is traversed using depth-first search and breadth-first search to obtain upstream and downstream relationships and potential features between commodities.

[0017] The present invention also provides a commodity relationship identification device, comprising: A data acquisition unit, used to acquire commodity-related data; A cleaning unit, used for cleaning the commodity-related data and obtaining business relationships; A coding unit, used for coding the commodity to obtain a coding result; A link generation unit, used to generate a commodity-level industrial chain according to the encoding result and the business relationship; A network generation unit, used for performing network merging and network pruning on the commodity-level industrial chain to obtain an industrial chain network; A traversal unit, used to traverse the industrial chain network to obtain upstream and downstream relationships and potential features between commodities; The output unit is used to output the upstream and downstream relationships and potential features between commodities.

[0018] The present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above method when executing the computer program.

[0019] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention collects commodity-related data and performs data cleaning to ensure data quality, thereby identifying business relationships; encodes the cleaned commodity data to generate a unified encoding result to provide a basis for subsequent analysis; constructs a commodity-level industrial chain based on the encoding result and business relationship to form a preliminary network structure; merges and prunes the industrial chain network to simplify the network structure and improve analysis efficiency and accuracy; traverses the optimized industrial chain network to explore the upstream and downstream relationships and potential features between commodities, and outputs analysis results to enhance practicality, thereby achieving effective integration of complex networks and deep learning to improve recognition accuracy, efficiency and practicality.

[0021] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0023] Figure 1 A schematic diagram of an application scenario of the commodity relationship identification method provided by an embodiment of the present invention; Figure 2 A schematic diagram of a flow chart of a commodity relationship identification method provided by an embodiment of the present invention; Figure 3A traversal diagram provided for an embodiment of the present invention Figure 1 ; Figure 4 A traversal diagram provided for an embodiment of the present invention Figure 2 ; Figure 5 A partial schematic diagram of an industrial chain network provided by an embodiment of the present invention; Figure 6 A schematic block diagram of a commodity relationship identification device provided by an embodiment of the present invention; Figure 7 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0026] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0028] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of the commodity relationship identification method provided in an embodiment of the present invention. Figure 2A schematic flow chart of a commodity relationship identification method provided in an embodiment of the present invention. The commodity relationship identification method is applied in a server. The server interacts with a terminal to obtain commodity-related data, cleans the data, and extracts business relationships; encodes commodity information to generate a unified encoding result; constructs a complex network of commodity-level industrial chains based on the encoding result and the business relationship; uses the Louvain algorithm to cut the network to identify different commodities and their associated paths; constructs a new graph based on the path analysis results, calculates the attributes of the edges, and extracts the core nodes; merges and prunes the commodity-level industrial chain to form an industrial chain network; traverses the industrial chain network through depth-first and breadth-first searches; and finally outputs the upstream and downstream relationships and potential features between commodities. Effective integration of complex networks and deep learning is achieved to improve the accuracy, efficiency, and practicality of recognition.

[0029] Figure 2 FIG. 1 is a flow chart of a commodity relationship identification method provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S170.

[0030] S110, obtaining commodity-related data.

[0031] In this embodiment, commodity-related data refers to data related to various attributes, transaction and circulation information of commodities.

[0032] Basic product information: product name; product number or barcode; category or classification; brand; Price information: sales price; cost price; discount and promotion information; Transaction information: sales quantity; transaction date; transaction method (online or offline); invoice information (seller and buyer); Inventory information: current inventory; inventory warning; inbound and outbound records; Supply chain information: supplier information; logistics and distribution information; Customer feedback: user reviews and ratings; return and exchange records; This data is crucial for companies' inventory management, sales analysis, marketing and decision-making. By analyzing product-related data, companies can better understand market demand, optimize supply chains and improve customer satisfaction.

[0033] S120: Clean the commodity-related data and obtain the business relationship.

[0034] In this embodiment, the commodity-related data is deduplicated, outliers are processed, and standardized, and the business relationship of the processed commodity-related data is analyzed.

[0035] Data cleaning is the first step in data processing, and its purpose is to improve the data quality for subsequent analysis. By cleaning the raw data, its accuracy and reliability are ensured.

[0036] In an invoice, the relationship between the seller and the buyer reflects the business relationship. In each transaction, a company may be both the seller and the buyer, so complex algorithms can be used to analyze the relationship between goods. See Table 1.

[0037] Table 1. Relationship between commodities and products Sales commodity Buyer Company A Product a Company B Company B Product Company C Company A Product g Company C Specifically, duplicates are detected and removed to ensure that each piece of data is unique, reducing the impact of data entry errors and improving overall data quality.

[0038] Use statistical and machine learning methods to identify outliers in the data that may distort the analysis results. For example, negative transaction amounts are obvious anomalies that need to be removed.

[0039] Standardize the data to ensure consistency. For example, unify the date format and currency unit (such as unifying "kg" and "kilogram" as "kg"). This helps to evaluate the company's purchase and sales data under the same standard and identify potential fraudulent behavior, so that unqualified data can be deleted without affecting subsequent model calculations.

[0040] Improve data accuracy and ensure reliable analysis results. Help identify and correct data entry errors and reduce unnecessary interference. Ensure data consistency, make subsequent analysis smoother, and enhance the effectiveness of the model.

[0041] S130: Encode the commodity to obtain an encoding result.

[0042] In this embodiment, the coding result refers to a unified code generated after the commodity name is standardized.

[0043] In one embodiment, the above-mentioned step S130 may include steps S131 - S132 .

[0044] S131. Extract product information from the product-related data using named entity technology.

[0045] Specifically, named entity recognition (NER) is a natural language processing technology that aims to automatically identify and classify specific information from text, such as product name, brand, quantity, and other product information. In our system, advanced NER technologies such as BERT and BiLSTM-CRF are used to improve the accuracy and efficiency of recognition.

[0046] When processing product descriptions, NER technology can accurately extract various entities. For example, in the text "1 Brand A laptop", the system can identify: Brand: Brand A; Product Name: Laptop; Unit: Taiwan; Quantity: 1; Each entity prediction comes with a confidence score, which reflects the system's confidence in the prediction. According to business needs, users can set the corresponding confidence threshold to ensure the quality of the extraction results. See Table 2 to Table 4.

[0047] Table 2

[0048] Table 3

[0049] Table 4

[0050] Automatically extract product information to reduce manual intervention; unify recognition standards to ensure consistency of product information in different texts; through confidence scores, users can filter and select high-quality extraction results as needed.

[0051] S132: Generate a product code according to the product information, and map the product code to a unified standard format using a connection technology to obtain a coding result.

[0052] In this embodiment, the main purpose of the Linking technology is to ensure the consistency and standardization of data. By mapping different naming and encoding methods to a unified standard format, the Linking technology can avoid information confusion and misunderstanding.

[0053] Through Linking technology, brand identifiers such as abbreviations can be mapped to more complete and official names, such as the full Chinese name. This mapping not only improves the readability of data, but also enhances the relevance of data.

[0054] Ensure the consistency of brand and product information in different data sources to avoid confusion caused by inconsistent names. By standardizing names, data is easier to analyze and use, especially in a big data environment. Data integration between different systems and platforms becomes smoother, which helps to build a more comprehensive product information library.

[0055] In summary, through NER and Linking technology, product-related data can be processed and managed efficiently and accurately, thereby improving the operational efficiency and decision-making quality of the enterprise.

[0056] S140: Generate a commodity-level industrial chain according to the coding result and the business relationship.

[0057] In this embodiment, the commodity-level industrial chain refers to a complex network that reflects the circulation and transaction of commodities by analyzing the relationship between commodities and enterprises. This network uses commodities and enterprises as nodes and connects them through transaction relationships, thereby showing the flow path and interdependence of commodities in the industrial chain.

[0058] Specifically, the commodity-level industrial chain can reveal the following aspects: Relationship between commodities and enterprises: Through coding results and operating relationships, relevant commodities are associated with participating enterprises to form a network structure.

[0059] Network analysis: Use algorithms (such as the Louvain algorithm) to cut and analyze the network to identify the relationship between different products and their position and importance in the overall network.

[0060] Path evaluation: Evaluate the association paths between commodities, including identifying the impact paths and calculating the weights, so as to understand the interactions and impacts of different commodities in the industrial chain.

[0061] Clustering and core node identification: Build a new graph based on the analysis results, calculate attributes, extract core nodes, and perform cluster analysis to identify key products and companies.

[0062] Through these analyses, we can better understand the role of commodities in the industrial chain, optimize supply chain management, and improve operational efficiency.

[0063] In one embodiment, the above-mentioned step S140 may include steps S141 to S145.

[0064] S141. Construct a complex network including commodities and enterprises according to the coding results and business relationships, wherein the coding results corresponding to the commodities and the enterprises are used as nodes, and the transaction relationships are used as edges.

[0065] In this embodiment, the complex network algorithm can reveal the interconnection and influence of various commodities in the industrial chain by deeply analyzing the relationship between commodities. This method uses data cleaning and relationship mining, based on complex network analysis, to identify the core influence and connection between different commodities (nodes).

[0066] S142. Use Louvain algorithm to segment the network and identify different products.

[0067] In this embodiment, the Louvain community detection algorithm is used to divide the complex industrial chain network into multiple communities. Each community represents a group of commodities with similar characteristics or functions. Through this method, the characteristics of each community are analyzed in detail to identify its role in the entire industrial chain; more effective business strategies and marketing plans are formulated based on the functions of different communities.

[0068] The multiple divided communities are actually modules, and modularity (Q) is an indicator to measure the quality of network division, which aims to maximize the connection density within the community while minimizing the connection between communities. Specifically, , where Q is the modularity and the goal is to maximize it. m is the sum of the weights of all edges in the network. Aij is the weight of the edge between node i and node j. If there is no edge between i and j, then Aij=0. ki and kj are the degrees of node i and node j respectively (the sum of the weights of all edges). δ(ci, cj) is an indicator function, if node i and node j are in the same community, then δ(ci, cj)=1, otherwise δ(ci, cj)=0.

[0069] S143. Perform modal dimensionality reduction on different commodities and enterprises to update the complex network.

[0070] In this embodiment, both commodities and enterprises are nodes in a complex network. Commodities and enterprises are two modes. In order to reduce the computational complexity, the two modes of commodities and enterprises are reduced to one mode of "commodities", and the complex industrial chain network is converted into a low-dimensional representation. This technology can greatly simplify the process of data analysis and visualization, thereby improving the efficiency of industrial chain analysis.

[0071] Specifically, an adjacency matrix is ​​created, with commodities and enterprises as the rows and columns of the matrix, and the corresponding transaction amounts or transaction frequencies are filled in. Enterprises are regarded as the aggregation of commodities traded with them. In the adjacency matrix, the transaction information of enterprises is integrated into the rows of the corresponding commodities, so that each commodity node can represent the transaction characteristics with all related enterprises.

[0072] The PCA technique is used to reduce the dimension of the aggregated adjacency matrix and extract the main components. The principal component that retains 95% of the variance is selected to simplify the feature dimension.

[0073] Use graph embedding techniques (such as Node2Vec, DeepWalk, etc.) to embed nodes in complex networks into low-dimensional space. These methods can effectively capture the similarities and relationships between nodes.

[0074] After the dimensionality reduction is completed, a new commodity modal network is formed: Based on the reduced-dimensional data, a new low-dimensional network is constructed, where nodes only retain commodities and their relationships with other commodities.

[0075] S144. Evaluate the association paths between different commodities according to the complex network, identify the influencing paths and weights, and obtain path analysis results.

[0076] In this embodiment, path analysis is used to evaluate the association between products and reveal the influencing paths and their weights. : By focusing on products, the analysis dimensions are reduced and efficiency is improved. The simplified network structure is easy to intuitively understand and display.

[0077] Specifically, the complex network includes nodes of two modes: enterprise and commodity. Enterprise nodes and commodity nodes appear alternately.

[0078] The edges of a complex network have attributes such as transaction time, transaction quantity, and transaction amount. For example, if enterprise A sells commodity Q to enterprise B, the amount attribute of the edge is q, and if enterprise A buys commodity P from enterprise C, the amount attribute of the edge is p.

[0079] The goal of this embodiment is to use the algorithm to reduce the dimension. When the transaction time of commodity P is earlier than the transaction time of commodity Q, delete the enterprise mode (enterprise A / B / C nodes) and only keep the commodity mode (commodity P / Q nodes). After the dimension reduction, the relationship is a directed relationship of P=>Q, and the amount attribute of the edge between P and Q is the smaller value of p and q.

[0080] S145. According to the path analysis results and the coding results corresponding to different commodities, a new graph is constructed, attributes are calculated, core nodes are extracted, and cluster analysis is performed to obtain a commodity-level industrial chain.

[0081] In one embodiment, the above-mentioned step S145 may include steps S1451 to S1454.

[0082] S1451. Define an edge set of a new graph according to the path analysis result and the encoding results corresponding to different commodities, and represent the upstream and downstream relationships between commodities as directed edges.

[0083] In this embodiment, when the transaction time of commodity P is earlier than that of commodity Q, a directed edge P→Q is established, and the weight of the edge depends on the minimum transaction amount. This method effectively reveals the upstream and downstream relationship between commodities.

[0084] Specifically, define the edge set (E') of the new graph: ; This formula defines the set of edges in the new graph (G'). For each pair of goods (P) and (Q), if there is a company (A) such that (A) buys (P) first and then sells (Q), then there is an edge from (P) to (Q).

[0085] S1452. Calculate the amount attribute, transaction quantity attribute, and upstream reason attribute of each edge in the new graph and use them as the weight of the edge.

[0086] In this embodiment, the amount attribute (m'(e')) of the new edge is defined as follows: ; This formula defines the amount attribute of each edge ((P, Q)) in the new graph (G'). For each edge ((P, Q)), its amount attribute (m'(P, Q)) is the smaller value of (m(A, P)) and (m(A, Q)).

[0087] Define the transaction quantity attribute (n'(e')) of the new graph edge: ; This formula defines the transaction quantity attribute of each edge ((P, Q)) in the new graph (G'). For each edge ((P, Q)), its transaction quantity attribute (n'(P, Q)) is the smaller value of (n(A, P)) and (n(A, Q)).

[0088] Define the upstream cause attribute (r'(e')) of the new graph edge: ; This formula defines the upstream cause property of each edge ((P, Q)) in the new graph (G'). For each edge ((P, Q)), its transaction quantity property (r'(P, Q)) is the smaller value of (r(A, P)) and (r(A, Q)).

[0089] The upstream reason attribute is used to describe that product P is the upstream of product Q (for example, "gas compressor-air compressor" is the upstream of "refrigeration and air conditioning equipment-air conditioning" because refrigeration and air conditioning equipment requires the use of air compressors). Its function is to assist manual labeling in the process of "active learning (manually judging samples that are not easily recognized by machines)" to improve labeling efficiency.

[0090] In this way, we can obtain an upstream and downstream relationship network between commodities, in which the weight of the edge is based on the transaction amount between enterprises, the number of traded commodities, and the reason why commodity P is the upstream of commodity Q.

[0091] S1453. Extract key commodity nodes in the new graph through the core decomposition algorithm.

[0092] In this embodiment, the most important core nodes (commodities) in the network are identified to facilitate subsequent analysis and decision-making.

[0093] S1454. Apply a cluster decomposition algorithm to divide the products in the new graph into different clusters to obtain a product-level industrial chain.

[0094] In this embodiment, commodities are divided into different clusters to help understand commodity characteristics and market classification.

[0095] Through core node analysis, we focus on the most influential products and optimize resource allocation.

[0096] Through cluster analysis, customized marketing strategies can better meet the needs of different groups.

[0097] After complex network algorithms and path analysis, a commodity-level industrial chain is formed. This link not only reveals the position of the commodity in the industrial chain, but also provides the correlation between commodities. Specific advantages include: Through commodity-level analysis, we can better understand the operation model of the entire industrial chain.

[0098] Provide enterprises with more scientific decision-making basis and enhance market competitiveness.

[0099] Through complex network algorithms, we can deeply understand the relationships between commodities in the industrial chain, optimize resource allocation, formulate effective market strategies, and ultimately improve the efficiency and benefits of the entire industrial chain.

[0100] S150: Perform network merging and network pruning on the commodity-level industrial chain to obtain an industrial chain network.

[0101] In this embodiment, the industrial chain network refers to the network formed after the commodity-level industrial chain is merged and pruned. Figure 5 shown.

[0102] Specifically, network merging, through algorithmic processing, merges redundant product networks, which can simplify the model structure and improve processing efficiency. This merging helps eliminate duplicate information, making product classification clearer, thereby improving the accuracy of subsequent analysis.

[0103] Network pruning: Pruning unnecessary paths in the model and filtering out invalid data. This process can reduce the complexity of the model, reduce the consumption of computing resources, and ensure higher prediction accuracy and faster response time.

[0104] In one embodiment, the above-mentioned step S150 may include steps S151-S152.

[0105] S151. Standardize the encoding results of the commodity-level industrial chain to obtain a merged network.

[0106] Specifically, synonyms are processed for the coding results within the commodity-level industrial chain to ensure that different names refer to the same commodity. This is crucial for commodity management and data analysis. For example, potatoes and potatoes are different names, but refer to the same commodity. Through standardization, companies can establish effective connections in complex networks and optimize inventory management and procurement decisions.

[0107] By using the format of Harbin Institute of Technology Synonym Dictionary and combining it with actual invoice data, we can extract synonyms for commodities. This method improves data consistency and ensures that the model can accurately identify products with different names when processing commodities.

[0108] For example, if one company sells potatoes and another company buys potatoes, these two companies can be connected in a complex network.

[0109] In constructing the synonym dictionary, we refer to the format of the Harbin Institute of Technology Synonym Dictionary and combine it with the product data in the real invoice to mine the synonyms of the products.

[0110] An example is as follows: Bh06A46 = sweet potato, sweet potato, sweet potato, yam, sweet potato, taro, cassava; Bh06A47=potato, yam, egg, potato; Bh06A48=yam, sweet potato, yam, and potato; S152. Use the NLU deep learning model to perform network pruning on the merged network to obtain the industrial chain network.

[0111] NLU (natural language understanding) uses the self-attention mechanism and feedforward neural network to effectively understand the upstream and downstream relationships of the industrial chain in the text. For example, identifying the relationship between "gas compressor" and "refrigeration and air-conditioning equipment" can provide data support for industrial chain management, as shown in Table 5.

[0112] Table 5. Data effects Gas compressor-air compressor Refrigeration and air conditioning equipment - air conditioning yes Refrigeration and air conditioning equipment requires the use of air compressors Cotton-Cotton Textile products-cotton cloth yes The production of cotton cloth requires the use of cotton NLU (Natural Language Understanding) is a subfield of Natural Language Processing (NLP) that focuses on understanding and interpreting human language. The NLU network architecture consists of multiple self-attention layers and feed-forward neural network layers. The self-attention mechanism allows the model to take into account all other elements in the sequence when processing each element in the sequence.

[0113] In addition, the NLG network architecture can be used for network pruning. NLG (Natural Language Generation): Based on the GPT model, it is trained through the parallel network WfqParallelNet. The model can generate explanations and reasons to make data relationships more intuitive. For example, it can generate a text description of "refrigeration and air conditioning equipment requires an air compressor" to help users understand the connection between products.

[0114] Specifically, the NLG network architecture uses GPT as the model base, builds a parallel network to train weights, and then merges it into the original network architecture after successful training.

[0115] GPT (Generative Pre-trained Transformer) is a Transformer-based NLG model. The NLG model uses self-attention mechanism, multi-head attention, feedforward neural network, residual connection and layer normalization to generate text. It is first pre-trained on a large amount of text, and then can be fine-tuned for specific tasks. During the pre-training stage, GPT predicts the next word in the text, which enables it to learn to generate coherent text.

[0116] The parallel network is named WfqParallelNet. The core of the network is that the low-rank matrix is ​​parallel to the pre-trained model. During back propagation, only the weights of the low-rank matrix are updated. During model inference, the weight matrix is ​​merged into the weights of the pre-trained model: ; Among them, W0 is the original weight matrix. is the change of weight matrix. B is a dxr matrix. A is a rxk matrix.

[0117] Specifically, the independence and distributed operation of each commodity industry chain make the management of the industry chain more flexible. The generation of sub-networks can reflect the industrial relations of specific commodities and improve the visualization and analysis capabilities of data.

[0118] Application of NEO4J graph database: Inserting the output commodity industry chain sub-network into the NEO4J database can achieve efficient management and query of complex data. Finally, it will be integrated into a larger industry chain structure, which is helpful for global analysis and decision support.

[0119] Through network merging and pruning, the model computing efficiency is improved and resources are saved; the combination of NLU and NLG makes complex data relationships easier to understand and enhances the basis for business decisions; standardizing product names and building synonym dictionaries improves data integration and accuracy; the independent sub-network design of the industrial chain provides a more flexible management method that can quickly adapt to market changes.

[0120] In summary, the implementation of this model and its methods will greatly improve the efficiency and accuracy of commodity management and industrial chain analysis, and provide strong data support for corporate decision-making.

[0121] S160, traversing the industrial chain network to obtain upstream and downstream relationships and potential characteristics between commodities.

[0122] In this embodiment, the upstream and downstream relationships and potential characteristics between commodities refer to the interdependence and connection between different commodities in the supply chain or industrial chain, as well as the characteristics or attributes they may possess.

[0123] Upstream and downstream relationships: Upstream goods: goods that provide raw materials, components or services in the production or supply process. For example, steel is an upstream commodity in automobile manufacturing.

[0124] Downstream goods: refers to goods that are produced or processed using upstream goods and are ultimately sold to consumers. For example, cars are downstream goods of steel.

[0125] Potential features: Functional characteristics: the purpose or function of a product, such as the heat resistance and corrosion resistance of a certain material.

[0126] Market demand: consumer demand for goods, popular trends, etc.

[0127] Price volatility: Potential characteristics related to price changes of other commodities that may affect supply chain decisions.

[0128] Quality standards: standards and requirements that goods must meet during the production and sales process.

[0129] By analyzing these upstream and downstream relationships and potential characteristics, companies can better understand market dynamics, optimize inventory management, formulate procurement strategies, etc.

[0130] Specifically, the industry chain network is traversed using depth-first search and breadth-first search to obtain the upstream and downstream relationships and potential features between commodities, such as Figure 3 and Figure 4 As shown,.

[0131] In practical applications, two search algorithms, depth-first search (DFS) and breadth-first search (BFS), are mainly used to traverse the industrial chain network of commodities. The combination of the two can effectively reveal the upstream and downstream relationships between commodities and other potential structural characteristics.

[0132] Depth-first search (DFS): DFS starts from the starting node, goes as deep as possible to the deepest point of each branch, and then backtracks to the nearest fork point to continue exploring other branches. This method is suitable for situations where complex paths or relationships need to be explored.

[0133] It can quickly find a feasible path and is suitable for deep structural feature analysis. It is more efficient for some specific structures (such as tree structures).

[0134] Breadth-first search (BFS): BFS starts from the starting node, visits all adjacent nodes first, and then expands outward layer by layer until all nodes are visited. It is suitable for scenarios where the shortest path or extensive relationships need to be found.

[0135] It can comprehensively cover all nodes of the network and is suitable for discovering extensive connections between products.

[0136] It is suitable for hierarchical analysis and can more intuitively display the relationship between products.

[0137] By combining DFS and BFS, we can simultaneously obtain in-depth details and extensive connections during the traversal process. This comprehensive analysis enables us to better understand the complexity of the industrial chain.

[0138] Combining the advantages of the two algorithms can achieve efficient traversal in complex networks. For example, DFS can help quickly find the direct upstream and downstream of a specific product, while BFS can ensure that no potential indirect relationships are missed.

[0139] By traversing the entire industrial chain network, the implicit relationships and structural characteristics between commodities can be revealed. This helps to discover new opportunities, such as potential partners or changes in market demand.

[0140] In-depth analysis of upstream and downstream relationships can provide data support for supply chain management, inventory control and market strategy formulation, helping companies optimize the decision-making process.

[0141] The traversal results can be used to construct visual graphics to clearly show the relationship between products, making it easier for management and related personnel to understand and analyze.

[0142] By combining the DFS and BFS algorithms, we can efficiently traverse the industrial chain network and deeply explore the upstream and downstream relationships and other potential characteristics between commodities. This not only provides companies with comprehensive market insights, but also helps optimize operations and decision support.

[0143] S170. Upstream and downstream relationships and potential characteristics between output commodities.

[0144] Output the upstream and downstream relationships and potential features between commodities to the terminal for display.

[0145] The method of this embodiment has the following advantages: Combination of complex network and deep learning commodity relationship identification method: innovative integration of complex network community detection and dimensionality reduction technology with deep learning model to achieve commodity relationship identification, breaking through the limitations of traditional methods that rely solely on network analysis or deep learning.

[0146] Industrial chain analysis based on the prediction of the main products of enterprises: By combining statistical analysis with deep learning, the main products of enterprises can be accurately predicted, providing high-quality basic data for industrial chain analysis and improving the accuracy of upstream and downstream relationship identification.

[0147] Product name standardization and coding method: Apply named entity recognition (NER) and linking technology to automatically identify and standardize product names, solve the problem of inconsistent product naming and coding, and improve data processing efficiency and accuracy.

[0148] Network pruning and low-rank matrix parallel network model: WfqParallelNet parallel network architecture is proposed, which combines low-rank matrix and pre-trained model parallel training to improve training efficiency and prediction performance without increasing model complexity.

[0149] Mining upstream and downstream relationships of commodities based on DFS: Using depth-first search and breadth-first search algorithms, we can deeply explore the upstream and downstream relationships between commodities and the potential network structure characteristics, providing new technical means for industrial chain analysis.

[0150] Generate explainable product relationship identification results: Combining natural language understanding (NLU) and natural language generation (NLG) models, it can not only identify product relationships, but also explain the reasons for the formation of relationships, enhance the interpretability of the results, and facilitate user understanding and application.

[0151] The above-mentioned commodity relationship identification method collects commodity-related data and performs data cleaning to ensure data quality, thereby identifying business relationships; encodes the cleaned commodity data to generate a unified encoding result to provide a basis for subsequent analysis; constructs a commodity-level industrial chain based on the encoding results and business relationships to form a preliminary network structure; merges and prunes the industrial chain network to simplify the network structure and improve analysis efficiency and accuracy; traverses the optimized industrial chain network to explore the upstream and downstream relationships and potential features between commodities, and outputs analysis results to enhance practicality, thereby achieving effective integration of complex networks and deep learning to improve recognition accuracy, efficiency and practicality.

[0152] Figure 6 is a schematic block diagram of a commodity relationship identification device 300 provided in an embodiment of the present invention. Figure 6 As shown, corresponding to the above commodity relationship identification method, the present invention also provides a commodity relationship identification device 300. The commodity relationship identification device 300 includes a unit for executing the above commodity relationship identification method, and the device can be configured in a server. Figure 6 The commodity relationship identification device 300 includes a data acquisition unit 301, a cleaning unit 302, an encoding unit 303, a link generation unit 304, a network generation unit 305, a traversal unit 306 and an output unit 307.

[0153] The data acquisition unit 301 is used to acquire commodity-related data; the cleaning unit 302 is used to clean the commodity-related data and acquire the business relationship; the encoding unit 303 is used to encode the commodity to obtain the encoding result; the link generation unit 304 is used to generate a commodity-level industrial chain according to the encoding result and the business relationship; the network generation unit 305 is used to merge and prune the commodity-level industrial chain to obtain the industrial chain network; the traversal unit 306 is used to traverse the industrial chain network to obtain the upstream and downstream relationships and potential characteristics between commodities; the output unit 307 is used to output the upstream and downstream relationships and potential characteristics between commodities.

[0154] In one embodiment, the cleaning unit 302 is used to remove duplication, process outliers, and standardize the commodity-related data, and analyze the business relationship of the processed commodity-related data.

[0155] In one embodiment, the encoding unit 303 includes: An extraction subunit, used to extract product information from the product-related data using named entity technology; The commodity coding subunit is used to generate a commodity code according to the commodity information, and map the commodity code to a unified standard format using a connection technology to obtain a coding result.

[0156] In one embodiment, the link generation unit 304 includes: A network construction subunit, used to construct a complex network including commodities and enterprises according to the coding results and business relationships, wherein the coding results corresponding to the commodities and the enterprises are used as nodes and the transaction relationships are used as edges; The cutting subunit is used to cut the network using the Louvain algorithm to identify different commodities; A dimension reduction subunit, used for performing modal dimension reduction on different commodities and enterprises to update the complex network; An analysis subunit, for evaluating association paths between different commodities according to the complex network, identifying influencing paths and weights, so as to obtain path analysis results; The processing subunit is used to construct a new graph, calculate attributes, extract core nodes and perform cluster analysis according to the path analysis results and the encoding results corresponding to different commodities, so as to obtain a commodity-level industrial chain.

[0157] In one embodiment, the processing subunit includes: A definition module, used to define an edge set of a new graph according to the path analysis result and the encoding results corresponding to different commodities, and to represent the upstream and downstream relationships between commodities as directed edges; The calculation module is used to calculate the amount attribute, transaction quantity attribute, and upstream reason attribute of each edge in the new graph and use them as the weight of the edge; Decomposition module, used to extract key commodity nodes in the new graph through the core decomposition algorithm; The clustering module is used to apply the cluster decomposition algorithm to divide the products in the new graph into different clusters to obtain the product-level industrial chain.

[0158] In one embodiment, the network generation unit 305 includes: A merging subunit, used for standardizing the encoding results of the commodity-level industrial chain to obtain a merged network; The pruning subunit is used to prune the merged network using the NLU deep learning model to obtain the industrial chain network.

[0159] In one embodiment, the traversal unit 306 is used to traverse the industrial chain network using depth-first search and breadth-first search to obtain upstream and downstream relationships and potential features between commodities.

[0160] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned product relationship identification device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0161] The above-mentioned commodity relationship identification device 300 can be implemented in the form of a computer program. The computer program can be implemented in the form of a computer program such as Figure 7 Runs on the computer device shown.

[0162] See also Figure 7 , Figure 7 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0163] See also Figure 7 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0164] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 can execute a commodity relationship identification method.

[0165] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0166] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a commodity relationship identification method.

[0167] The network interface 505 is used to communicate with other devices over the network. Figure 7 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0168] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps: Acquire commodity-related data; perform data cleaning on the commodity-related data and obtain business relationships; encode the commodities to obtain encoding results; generate commodity-level industrial chains according to the encoding results and business relationships; perform network merging and network pruning on the commodity-level industrial chains to obtain industrial chain networks; traverse the industrial chain networks to obtain upstream and downstream relationships and potential features between commodities; and output upstream and downstream relationships and potential features between commodities.

[0169] In one embodiment, when the processor 502 implements the step of cleaning the commodity-related data and acquiring the business relationship, the processor 502 specifically implements the following steps: The commodity-related data is deduplicated, outliers processed, and standardized, and the business relationship of the processed commodity-related data is analyzed.

[0170] In one embodiment, when the processor 502 implements the step of encoding the commodity to obtain the encoding result, the processor 502 specifically implements the following steps: The named entity technology is used to extract the commodity information from the commodity-related data; a commodity code is generated according to the commodity information, and the commodity code is mapped to a unified standard format using a connection technology to obtain a coding result.

[0171] In one embodiment, when the processor 502 implements the step of generating a commodity-level industrial chain according to the encoding result and the business relationship, the processor 502 specifically implements the following steps: A complex network including commodities and enterprises is constructed according to the coding results and business relationships, wherein the coding results corresponding to the commodities and the enterprises are used as nodes, and the transaction relationships are used as edges; the network is cut using the Louvain algorithm to identify different commodities; modal dimension reduction is performed on different commodities and enterprises to update the complex network; the association paths between different commodities are evaluated according to the complex network, and the influencing paths and weights are identified to obtain path analysis results; new graph construction, attribute calculation, core node extraction and cluster analysis are performed according to the path analysis results and the coding results corresponding to different commodities to obtain commodity-level industrial chain paths.

[0172] In one embodiment, when the processor 502 implements the step of constructing a new graph, calculating attributes, extracting core nodes, and performing cluster analysis according to the path analysis results and the encoding results corresponding to different commodities to obtain a commodity-level industrial chain, the processor 502 specifically implements the following steps: According to the path analysis results and the encoding results corresponding to different commodities, the edge set of the new graph is defined, and the upstream and downstream relationships between commodities are represented as directed edges; the amount attribute, transaction quantity attribute and upstream reason attribute of each edge in the new graph are calculated and used as the weight of the edge; the key commodity nodes in the new graph are extracted through the core decomposition algorithm; the cluster disassembly algorithm is applied to divide the commodities in the new graph into different clusters to obtain the commodity-level industrial chain.

[0173] In one embodiment, when the processor 502 implements the step of merging and pruning the commodity-level industrial chain to obtain an industrial chain network, the processor 502 specifically implements the following steps: The encoding results of the commodity-level industrial chain are standardized to obtain a merged network; the merged network is pruned using an NLU deep learning model to obtain an industrial chain network.

[0174] In one embodiment, when the processor 502 implements the step of traversing the industrial chain network to obtain the upstream and downstream relationships and potential features between commodities, the processor 502 specifically implements the following steps: The industry chain network is traversed using depth-first search and breadth-first search to obtain upstream and downstream relationships and potential features between commodities.

[0175] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0176] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0177] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps: Acquire commodity-related data; perform data cleaning on the commodity-related data and obtain business relationships; encode the commodities to obtain encoding results; generate commodity-level industrial chains according to the encoding results and business relationships; perform network merging and network pruning on the commodity-level industrial chains to obtain industrial chain networks; traverse the industrial chain networks to obtain upstream and downstream relationships and potential features between commodities; and output upstream and downstream relationships and potential features between commodities.

[0178] In one embodiment, when the processor executes the computer program to implement the step of cleaning the commodity-related data and obtaining the business relationship, the processor specifically implements the following steps: The commodity-related data is deduplicated, outliers processed, and standardized, and the business relationship of the processed commodity-related data is analyzed.

[0179] In one embodiment, when the processor executes the computer program to implement the step of encoding the commodity to obtain the encoding result, the processor specifically implements the following steps: The named entity technology is used to extract the commodity information from the commodity-related data; a commodity code is generated according to the commodity information, and the commodity code is mapped to a unified standard format using a connection technology to obtain a coding result.

[0180] In one embodiment, when the processor executes the computer program to implement the step of generating a commodity-level industrial chain according to the encoding result and the business relationship, the processor specifically implements the following steps: A complex network including commodities and enterprises is constructed according to the coding results and business relationships, wherein the coding results corresponding to the commodities and the enterprises are used as nodes, and the transaction relationships are used as edges; the network is cut using the Louvain algorithm to identify different commodities; modal dimension reduction is performed on different commodities and enterprises to update the complex network; the association paths between different commodities are evaluated according to the complex network, and the influencing paths and weights are identified to obtain path analysis results; new graph construction, attribute calculation, core node extraction and cluster analysis are performed according to the path analysis results and the coding results corresponding to different commodities to obtain commodity-level industrial chain paths.

[0181] In one embodiment, when the processor executes the computer program to implement the steps of constructing a new graph, calculating attributes, extracting core nodes, and clustering analysis according to the path analysis results and the coding results corresponding to different commodities to obtain a commodity-level industrial chain, the processor specifically implements the following steps: According to the path analysis results and the encoding results corresponding to different commodities, the edge set of the new graph is defined, and the upstream and downstream relationships between commodities are represented as directed edges; the amount attribute, transaction quantity attribute and upstream reason attribute of each edge in the new graph are calculated and used as the weight of the edge; the key commodity nodes in the new graph are extracted through the core decomposition algorithm; the cluster disassembly algorithm is applied to divide the commodities in the new graph into different clusters to obtain the commodity-level industrial chain.

[0182] In one embodiment, when the processor executes the computer program to implement the step of merging and pruning the commodity-level industrial chain to obtain an industrial chain network, the processor specifically implements the following steps: The encoding results of the commodity-level industrial chain are standardized to obtain a merged network; the merged network is pruned using an NLU deep learning model to obtain an industrial chain network.

[0183] In one embodiment, when the processor executes the computer program to implement the step of traversing the industrial chain network to obtain the upstream and downstream relationships and potential features between commodities, the processor specifically implements the following steps: The industry chain network is traversed using depth-first search and breadth-first search to obtain upstream and downstream relationships and potential features between commodities.

[0184] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0185] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0186] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0187] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0188] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0189] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A commodity relationship identification method, characterized in that: include: Get product-related data; Clean the data related to the commodities and obtain the business relationship; Encode the product to obtain an encoding result; Generate a commodity-level industrial chain according to the coding results and business relationships; Performing network merging and network pruning on the commodity-level industrial chain to obtain an industrial chain network; Traversing the industrial chain network to obtain upstream and downstream relationships and potential characteristics between commodities; Output the upstream and downstream relationships and potential characteristics between commodities.

2. The commodity relationship identification method according to claim 1, characterized in that: The data cleaning of the commodity-related data and obtaining the business relationship includes: The commodity-related data is deduplicated, outliers processed, and standardized, and the business relationship of the processed commodity-related data is analyzed.

3. The commodity relationship identification method according to claim 1, characterized in that: The step of encoding the commodity to obtain an encoding result includes: Extracting product information from the product-related data using named entity technology; A commodity code is generated according to the commodity information, and the commodity code is mapped to a unified standard format using a connection technology to obtain a coding result.

4. The commodity relationship identification method according to claim 1, characterized in that: Generating a commodity-level industrial chain according to the coding result and the business relationship includes: A complex network including commodities and enterprises is constructed according to the coding results and business relationships, wherein the coding results and enterprises corresponding to the commodities are used as nodes and the transaction relationships are used as edges; Use the Louvain algorithm to cut the network and identify different products; Performing modal dimensionality reduction on different commodities and enterprises to update the complex network; Evaluate the association paths between different commodities according to the complex network, identify the influencing paths and weights, and obtain path analysis results; According to the path analysis results and the coding results corresponding to different commodities, a new graph is constructed, attributes are calculated, core nodes are extracted, and cluster analysis is performed to obtain a commodity-level industrial chain.

5. The commodity relationship identification method according to claim 4, characterized in that: The new graph is constructed, attributes are calculated, core nodes are extracted, and cluster analysis is performed based on the path analysis results and the coding results corresponding to different commodities to obtain a commodity-level industrial chain, including: According to the path analysis results and the encoding results corresponding to different commodities, an edge set of a new graph is defined, and the upstream and downstream relationships between commodities are represented as directed edges; Calculate the amount attribute, transaction quantity attribute, and upstream reason attribute of each edge in the new graph and use them as the weight of the edge; Extract key commodity nodes in the new graph through the core decomposition algorithm; The cluster decomposition algorithm is applied to divide the products in the new graph into different clusters to obtain the product-level industrial chain.

6. The commodity relationship identification method according to claim 1, characterized in that: The network merging and network pruning of the commodity-level industrial chain to obtain an industrial chain network includes: Standardizing the encoding results of the commodity-level industrial chain to obtain a merged network; The NLU deep learning model is used to perform network pruning on the merged network to obtain the industrial chain network.

7. The commodity relationship identification method according to claim 1, characterized in that: The traversal of the industrial chain network to obtain the upstream and downstream relationships and potential characteristics between commodities includes: The industry chain network is traversed using depth-first search and breadth-first search to obtain upstream and downstream relationships and potential features between commodities.

8. A commodity relationship identification device, characterized in that: include: A data acquisition unit, used to acquire commodity-related data; A cleaning unit, used for cleaning the commodity-related data and obtaining business relationships; A coding unit, used for coding the commodity to obtain a coding result; A link generation unit, used to generate a commodity-level industrial chain according to the encoding result and the business relationship; A network generation unit, used for performing network merging and network pruning on the commodity-level industrial chain to obtain an industrial chain network; A traversal unit, used to traverse the industrial chain network to obtain upstream and downstream relationships and potential features between commodities; The output unit is used to output the upstream and downstream relationships and potential features between commodities.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images