Electronic component online sales data management and maintenance system

Abstract

The invention relates to the technical field of sales data management, and discloses an electronic component online sales data management and maintenance system, which comprises a data acquisition and processing module, a user demand matching module, a data verification module, an inventory optimization module and the like. The data acquisition and processing module acquires multi-dimensional sales data, and performs natural language processing and dynamic knowledge graph construction, processing and data association; the user demand matching module generates an initial matching result based on a collaborative filtering algorithm and semantic clustering; the data verification module verifies the data by using a Hash algorithm and a block chain network; and the inventory optimization module optimizes the inventory according to the timeliness weight factor and a reinforcement learning algorithm. In addition, the system is further provided with an abnormal transaction detection module, a multi-source data integration module and an interactive query optimization module which are respectively used for detecting abnormal transactions, integrating multi-source data and optimizing query. The system effectively solves the problem of electronic component online sales data management, and improves sales efficiency and user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of sales data management, and particularly to an online sales data management and maintenance system for electronic components. Background Art

[0002] In today's digital age, the online sales industry of electronic components has developed vigorously, and the transaction scale has been continuously expanding. However, with the growth of business, the management of online sales data of electronic components faces many challenges.

[0003] The traditional management methods of electronic component sales data are relatively scattered and lack systematic integration. Data from different sources, such as supplier data, sales platform data, etc., often have semantic differences and are difficult to uniformly process and analyze. This makes it difficult for enterprises to comprehensively and accurately grasp the sales situation of electronic components and unable to provide strong support for decision-making. For example, different suppliers describe the parameters of electronic components in different ways, resulting in confusion when summarizing data and affecting the accuracy and usability of the data.

[0004] In terms of data processing, in the face of a large amount of sales data, existing methods are difficult to process and analyze efficiently. Multi-dimensional sales data contains a large amount of unstructured information, such as user evaluations, product descriptions, etc. Traditional data processing technologies cannot quickly convert it into structured data and are difficult to mine valuable information from it. For example, when analyzing the feedback of users on a certain electronic component, it is difficult to extract key information due to the inconsistent data format, and it is impossible to timely understand the advantages and disadvantages of the product, thereby affecting the optimization and promotion of the product.

[0005] User demand matching is a key link in the online sales of electronic components, but the current matching method is not accurate enough. Existing recommendation systems often rely on simple rules or basic algorithms and cannot deeply understand the complex needs of users. For example, simply recommending based on the user's historical purchase records without considering the dynamic changes in user needs and the semantic associations between products results in a deviation between the recommended electronic components and the actual needs of users, reducing the user's purchase experience and sales conversion rate.

[0006] The security and reliability of data verification also need to be improved urgently. The sales of electronic components involve many transaction links, and the authenticity and integrity of data are crucial. However, traditional data verification methods are vulnerable to tampering and forgery threats and are difficult to ensure the credibility of data. For example, during the transaction process, the attribute information of electronic components may be maliciously tampered with, affecting the fairness and security of the transaction and bringing potential losses to enterprises and users.

[0007] Inventory management is also a difficult problem. The inventory of electronic components is affected by various factors, such as shelf life, changes in market demand, etc. Traditional inventory management methods cannot dynamically adjust inventory strategies according to these factors, easily leading to inventory backlogs or out-of-stock situations. Inventory backlogs will occupy a large amount of funds and increase enterprise costs; out-of-stock situations will affect customer satisfaction and miss sales opportunities. For example, for some electronic components with a short shelf life, if the inventory cannot be adjusted in a timely manner according to market demand, it is easy to cause expired waste.

[0008] In addition, the ability to detect abnormal transactions is insufficient, and it is impossible to detect and prevent fraud in a timely manner. During the online sales process of electronic components, there may be situations such as abnormal orders and false transactions. Existing detection methods are difficult to quickly and accurately identify these abnormal behaviors, bringing economic losses to enterprises. At the same time, there are also deficiencies in multi-source data integration and query optimization, unable to efficiently integrate external data, with low query efficiency, affecting the operation efficiency of enterprises and customer experience. To sum up, it is of great practical significance to develop an efficient online sales data management and maintenance system for electronic components. Summary of the Invention

[0009] The purpose of the present invention is to provide an online sales data management and maintenance system for electronic components to solve the problems raised in the above background technology.

[0010] To achieve the above purpose, the present invention provides the following technical solution: An online sales data management and maintenance system for electronic components, the system includes:

[0011] A data acquisition and processing module, used to acquire multi-dimensional sales data of electronic components, perform natural language processing on the sales data to obtain a structured data set, and construct the association relationship between entities in the structured data set through a dynamic knowledge graph;

[0012] A user demand matching module, which generates a user demand feature vector based on the collaborative filtering algorithm, performs semantic clustering on the structured data set to obtain multiple candidate electronic component categories, and generates an initial matching result according to the semantic similarity between the user demand feature vector and each candidate electronic component category;

[0013] A data verification module, which performs feature encoding on the electronic component attribute information in the initial matching result based on the hash algorithm to generate a set of unique identification codes, and performs distributed verification on the set of unique identification codes through a blockchain network to obtain a verified target electronic component data set;

[0014] An inventory optimization module, which performs timeliness marking on the inventory information in the target electronic component data set according to a preset timeliness weight factor, and dynamically adjusts the update strategy of the timeliness weight factor based on the reinforcement learning algorithm to generate optimized inventory status data.

[0015] Preferably, the method for constructing the dynamic knowledge graph includes:

[0016] Extract the entity names, attribute descriptions, and transaction records from the structured data set, and identify the implicit semantic relationships between entities through a bidirectional long short-term memory network model;

[0017] Construct an initial embedding representation of entity nodes based on the graph convolutional network algorithm, and calculate the weight value of the relationship edge according to the transaction frequency between entities and the time decay coefficient;

[0018] Adopt a dynamic time window mechanism to incrementally update the knowledge graph. When new transaction data is detected, recalculate the embedding representation of the affected entity nodes and the weight value of the relationship edge.

[0019] Preferably, the method for generating the user demand feature vector includes:

[0020] Collect the user's historical search records and order data, and convert the unstructured text into a semantic vector through a word embedding model;

[0021] Extract the key demand features in the user behavior sequence based on the attention mechanism, and calculate the decay coefficient of the demand features in combination with the timestamp information;

[0022] Use a generative adversarial network to perform data augmentation on the sparse demand vector to generate a high-dimensional continuous user demand feature vector.

[0023] Preferably, the implementation method of semantic clustering includes:

[0024] Perform dependency syntactic analysis on the electronic component description text in the structured data set, and extract the functional parameter, package type, and electrical characteristic triples;

[0025] Based on the density peak clustering algorithm, perform multi-modal feature fusion on the triples to generate the core clustering clusters of candidate electronic component categories;

[0026] Perform secondary partitioning on the core clustering clusters through the spectral clustering algorithm to eliminate the noise data points in the semantic overlapping area.

[0027] Preferably, the feature encoding rule of the hash algorithm includes:

[0028] Concatenate the model number, production batch, and supplier information of the electronic component into an original string, and generate a hash digest of a fixed length through the SHA-3 algorithm;

[0029] Based on the Bloom filter, perform deduplication processing on the hash digest and retain the unique identification code;

[0030] The Merkle tree structure is adopted to perform hierarchical verification on the set of unique identification codes, and an immutable hash proof chain is generated.

[0031] Preferably, the calculation method of the timeliness weight factor includes:

[0032] Obtain the inventory receipt time, shelf life, and market circulation rate of the electronic components, and predict the remaining effective cycle of the inventory through the exponential smoothing model;

[0033] Construct a time-sensitive function, and calculate the initial timeliness weight according to the ratio of the remaining effective cycle to the preset threshold;

[0034] Introduce the Markov decision process to simulate the inventory state transition path, and dynamically adjust the decay rate of the timeliness weight.

[0035] Preferably, the update strategy of the reinforcement learning algorithm includes:

[0036] Define the inventory state as the observation space of the agent, the timeliness weight adjustment action as the action space, and the inventory turnover rate as the reward function;

[0037] Train the policy network through the deep deterministic policy gradient algorithm to generate the optimal timeliness weight update strategy;

[0038] Adopt an experience replay pool to store historical state transition data, and preferentially sample high-variance state segments to update the policy network parameters.

[0039] Preferably, the system further includes an abnormal transaction detection module, and its implementation method includes:

[0040] Monitor the user transaction behavior data stream in real time, and detect abnormal patterns of order amount, frequency, and geographical location through the isolation forest algorithm;

[0041] Build a distributed abnormal scoring model based on the federated learning framework, and aggregate the local gradients of multiple data nodes to update the global model parameters;

[0042] When the abnormal score exceeds the dynamic threshold, trigger the transaction traceability mechanism, and associate and query the historical transaction hash records in the blockchain network.

[0043] Preferably, the system further includes a multi-source data integration module, and its implementation method includes:

[0044] Connect to the third-party supplier database, and eliminate the semantic ambiguity of different data sources through ontology alignment technology;

[0045] Based on the transfer learning algorithm, map the external data features to the embedding space of the local knowledge graph to generate a unified semantic representation;

[0046] Use differential privacy technology to add noise to the integrated sensitive data and generate a fused dataset that meets the privacy protection requirements.

[0047] Preferably, the system further includes an interactive query optimization module, and its implementation method includes:

[0048] Parse the user's natural language query statement, and extract the query intent and constraint conditions through semantic role labeling;

[0049] Construct a multi-objective optimization function to balance query response time, data integrity, and computing resource consumption;

[0050] Based on the quantum particle swarm optimization algorithm, search for the optimal query execution plan and dynamically generate the index call sequence of the distributed database.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] At the data processing and analysis level, the data acquisition and processing module can obtain multi-dimensional sales data, transform it into a structured data set with the help of natural language processing technology, and construct the association relationship between entities through a dynamic knowledge graph. This process can not only efficiently process a large amount of complex data, but also deeply explore the data value. For example, by analyzing the entity relationship in the electronic component sales data, the association rules between different components in sales can be found, providing a basis for the enterprise to formulate a combined sales strategy, helping the enterprise better grasp the market trend, optimize the product layout, and improve the market competitiveness.

[0053] The user demand matching module generates a user demand feature vector based on the collaborative filtering algorithm, and gives an initial matching result by combining semantic clustering and similarity calculation. Compared with the traditional recommendation method, this module can understand the user's needs more accurately, improve the accuracy and pertinence of the recommendation. Taking the procurement of electronic components as an example, it can accurately match the components that meet the user's needs according to the user's historical search and order data, reduce the user's screening time, improve the purchase efficiency, thereby increasing the user's satisfaction and loyalty, and bringing more potential sales opportunities to the enterprise.

[0054] The data verification module uses the hash algorithm and blockchain technology to perform feature encoding and distributed verification on the electronic component attribute information. This ensures the uniqueness and immutability of the data, and guarantees the security of the transaction and the credibility of the data. In the complex transaction environment of electronic component sales, it effectively prevents the data from being maliciously tampered with, reduces the transaction risk, protects the legitimate rights and interests of enterprises and users, and provides a reliable data basis for the online sales of electronic components.

[0055] The inventory optimization module marks inventory information according to the preset timeliness weight factor and dynamically adjusts the strategy through the reinforcement learning algorithm. This module can fully consider factors such as the inventory receipt time, shelf life, and market circulation rate of electronic components to achieve dynamic optimization management of inventory. By accurately predicting the remaining effective cycle of inventory and reasonably adjusting the inventory status, enterprises can avoid inventory backlogs or shortages, reduce capital occupation, lower operating costs, ensure the timeliness of product supply, and improve customer satisfaction. The newly added abnormal transaction detection module in the system monitors transaction behaviors in real time through the isolation forest algorithm and the federated learning framework, detects abnormal transactions in a timely manner, and triggers the traceability mechanism. This module effectively prevents fraud and safeguards the enterprise's capital security. During the frequent trading process of online sales of electronic components, it can quickly identify abnormal orders, avoid economic losses to the enterprise, and maintain the normal order of market transactions. The multi-source data integration module realizes the efficient integration and privacy protection of multi-source data with the help of ontology alignment technology, transfer learning algorithm, and differential privacy technology. Enterprises can access third-party supplier databases, eliminate semantic ambiguities, enrich data resources, and ensure the security of sensitive data at the same time. This helps enterprises obtain more comprehensive market information, provides richer data support for decision-making, and enhances the enterprise's market insight and decision-making scientificity.

[0056] The interactive query optimization module parses natural language query statements, constructs a multi-objective optimization function, and uses the quantum particle swarm algorithm to search for the optimal query execution plan. This module significantly improves query efficiency, reduces query response time, and ensures data integrity and reasonable utilization of computing resources. When users query information related to electronic components, they can quickly obtain accurate results, which improves the user experience and the efficiency of internal data retrieval and analysis within the enterprise, promoting the improvement of enterprise operation efficiency. Description of the Drawings

[0057] Figure 1 is the working principle diagram of the online sales data management and maintenance system for electronic components described in the present invention;

[0058] Figure 2 is the flowchart for generating the user demand feature vector;

[0059] Figure 3 is the working flowchart of the abnormal transaction detection module. Detailed Embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0061] Please refer to Figures 1 - 3 , the present invention provides a technical solution: an online sales data management and maintenance system for electronic components, the system includes:

[0062] Data acquisition and processing module: This module is responsible for acquiring multi-dimensional sales data of electronic components. These data come from a wide range of sources, including transaction records on online sales platforms, user evaluations, market research data, etc. The acquired data usually contains a large amount of unstructured information, which needs to be processed by natural language processing. Using means such as lexical analysis and syntactic analysis in natural language processing technology, the unstructured sales data is transformed into a structured data set for subsequent processing. The association relationships between entities in the structured data set are constructed through dynamic knowledge graphs. For example, information such as the model number, manufacturer, sales time, and sales quantity of electronic components is extracted from the sales data. Taking these information as entities, by analyzing the connections between them, a knowledge graph reflecting the sales situation of electronic components is constructed, intuitively showing the relationships between entities.

[0063] User demand matching module: Generate user demand feature vectors based on the collaborative filtering algorithm. The collaborative filtering algorithm analyzes the historical behavior data of users to find other users with similar behaviors to the target user, thereby predicting the needs of the target user. Semantic clustering is performed on the structured data set to obtain multiple candidate electronic component categories. Semantic clustering classifies according to the semantic features of electronic components, grouping electronic components with similar functions and characteristics into one category. The initial matching result is generated according to the semantic similarity between the user demand feature vector and each candidate electronic component category. By calculating the similarity between vectors, the electronic component category that best matches the user's needs is selected to provide accurate recommendations for users.

[0064] Data verification module: Feature encode the electronic component attribute information in the initial matching result based on the hash algorithm to generate a set of unique identification codes. The hash algorithm transforms the attribute information of electronic components into hash values of a fixed length, ensuring that each electronic component has a unique identifier. The set of unique identification codes is verified distributively through the blockchain network to obtain the verified target electronic component data set. The decentralized and tamper-proof characteristics of blockchain technology ensure the authenticity and reliability of electronic component data, preventing data from being tampered with or forged.

[0065] Inventory Optimization Module: Mark the timeliness of inventory information in the target electronic component data set according to the preset timeliness weight factor. The timeliness weight factor takes into account factors such as the inventory receipt time, shelf life, and market circulation rate of electronic components, reasonably marks the inventory information, and distinguishes the timeliness of different electronic components. Dynamically adjust the update strategy of the timeliness weight factor based on the reinforcement learning algorithm to generate optimized inventory status data. The reinforcement learning algorithm continuously tries and learns, and automatically adjusts the update strategy of the timeliness weight factor according to indicators such as inventory turnover rate to achieve optimized inventory management and avoid inventory backlogs or shortages.

[0066] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0067] Embodiment 1:

[0068] This embodiment details the construction method of the dynamic knowledge graph, which aims to more accurately reflect the complex relationships between entities in the electronic component sales data and provide strong support for subsequent data analysis and decision-making. The specific methods include:

[0069] Entity and Relationship Recognition: Extract entity names, attribute descriptions, and transaction records from the structured data set. Taking the sales data of a certain electronic component as an example, entity names may include specific models of electronic components, such as "Resistor R100", "Capacitor C200", etc.; attribute descriptions cover various parameters of electronic components, such as the resistance value of a resistor and the capacitance value of a capacitor; transaction records include information such as transaction time, transaction quantity, and transaction parties. Identify the implicit semantic relationships between entities through the Bidirectional Long Short-Term Memory Network model (Bi-LSTM). Bi-LSTM can process time series data and learn the long-term dependencies in the data. In the electronic component sales data, it can analyze information such as transaction records and attribute descriptions, and discover potential relationships between electronic components, such as finding that a certain model of resistor often appears in the same order as a specific model of capacitor, thus identifying the association relationship between them.

[0070] Initial Embedding Representation and Weight Calculation: Based on the Graph Convolutional Network algorithm (GCN), the initial embedding representation of entity nodes is constructed. GCN can perform convolutional operations on graph-structured data, fuse and transform the feature information of entity nodes, and generate low-dimensional vector representations for subsequent calculation and analysis. In this system, the attribute information, transaction records, etc. of electronic components are used as node features, and the initial embedding representation of entity nodes is obtained through GCN. The weight value of the relationship edge is calculated according to the transaction frequency between entities and the time decay coefficient. The higher the transaction frequency, the closer the relationship between the two entities, and the greater the weight value. At the same time, considering the time decay coefficient, the transactions closer to the current time contribute more to the weight value. For example, if electronic components A and B have been traded 10 times in the past month and 5 times half a year ago, when calculating the weight value, the recent 10 transactions have a greater impact on the weight value.

[0071] Incremental Update of Knowledge Graph: The dynamic time window mechanism is adopted to incrementally update the knowledge graph. When new transaction data is detected, the embedding representation of the affected entity nodes and the weight value of the relationship edge are recalculated. For example, when there is a new order involving the transaction of electronic components C and D, within the dynamic time window, the new transaction data is included in the calculation scope, and the embedding representation of entity nodes C and D and the weight value of the relationship edge between them are recalculated. This can timely reflect the changes in electronic component sales data and ensure the timeliness and accuracy of the knowledge graph.

[0072] Example 2:

[0073] This example is used to describe the method for generating the user demand feature vector, which aims to more accurately capture user demands and provide a basis for precise matching of electronic components. The specific method includes:

[0074] Collect the user's historical search records and order data, which contain a large amount of unstructured text information, such as keywords entered by the user in the search box, order remarks, etc. Through word embedding models, such as Word2Vec or GloVe, the unstructured text is converted into semantic vectors. Taking the user's search for "high-performance CPU" as an example, the word embedding model will convert "high-performance" and "CPU" into corresponding semantic vectors respectively, and these vectors can reflect the semantic features of the words and the similarity between them.

[0075] Extract key requirement features from the user behavior sequence based on the attention mechanism. The attention mechanism can automatically assign weights according to the importance of user behavior data to highlight key information. In the user's search records and order data, certain keywords or operations may better reflect the user's core requirements, and these key requirement features can be extracted through the attention mechanism. Calculate the decay coefficient of the requirement features in combination with the timestamp information. The user's requirements may change over time, and recent requirements can better reflect the current actual requirements. For example, if the user searched for "old-fashioned electron tubes" a month ago and recently searched for "new semiconductor chips", the decay coefficient of the requirement features related to "new semiconductor chips" is smaller and the weight is higher when generating the requirement feature vector.

[0076] Use a generative adversarial network (GAN) to perform data augmentation on the sparse demand vector. In practical applications, the user's historical behavior data may be less, resulting in a sparse demand vector, which is difficult to accurately reflect the user's needs. A GAN consists of a generator and a discriminator. The generator can generate new samples based on existing data, and the discriminator is used to determine whether the generated samples are real. By performing data augmentation on the sparse demand vector through a GAN, a high-dimensional continuous user demand feature vector is generated, improving the accuracy and richness of the vector and more comprehensively describing the user's needs.

[0077] Example 3:

[0078] This example details the implementation method of semantic clustering, whose function is to reasonably classify electronic components and improve the matching efficiency and accuracy. The specific method includes:

[0079] Extract triples through text analysis: Perform dependency syntactic analysis on the text descriptions of electronic components in the structured data set. Dependency syntactic analysis can analyze the dependency relationships between words in a sentence and determine the syntactic roles of each word. Through this analysis, triples of functional parameters, package types, and electrical characteristics are extracted. For example, for the description of an electronic component "This chip has high-speed data processing capabilities, uses BGA packaging, and has a working voltage of 3.3V", a triple consisting of the functional parameter "high-speed data processing capabilities", the package type "BGA", and the electrical characteristic "working voltage 3.3V" can be extracted.

[0080] Generate core clustering clusters through multimodal feature fusion: Perform multimodal feature fusion on the triples based on the density peak clustering algorithm. The density peak clustering algorithm can automatically identify clustering centers according to the density and distance information of data points. In this example, multimodal features such as functional parameters, package types, and electrical characteristics are combined, and core clustering clusters of candidate electronic component categories are generated through the density peak clustering algorithm. This can gather electronic components with similar features together to form a preliminary classification.

[0081] Secondary partitioning to eliminate noisy data points: The core clustering clusters are secondarily partitioned through the spectral clustering algorithm. The spectral clustering algorithm is a graph-theory-based clustering algorithm that transforms the clustering problem into a graph partitioning problem by constructing a similarity graph of data points. When secondarily partitioning the core clustering clusters, the spectral clustering algorithm can eliminate the noisy data points within the semantic overlapping regions, making the clustering results more accurate and clear. For example, in the classification of certain electronic components, there may be some data points with fuzzy features and difficult to accurately classify. Through the secondary partitioning of the spectral clustering algorithm, these noisy data points can be removed, improving the quality of clustering.

[0082] Example 4:

[0083] This example elaborates on the feature encoding rules and verification process of the hash algorithm, whose function is to ensure the uniqueness and security of electronic component data and prevent data from being tampered with. The specific methods include:

[0084] Concatenate the model number, production batch, and supplier information of the electronic component into an original string. For example, for an electronic component with a model number of "IC001", a production batch of "20230101", and a supplier of "ABC Company", the concatenated original string is "IC00120230101ABC Company". Generate a fixed-length hash digest through the SHA-3 algorithm. The SHA-3 algorithm is a secure hash algorithm with good collision resistance and irreversibility. Input the original string into the SHA-3 algorithm to obtain a fixed-length hash value as the hash digest of the electronic component.

[0085] Deduplicate the hash digest based on the Bloom filter. The Bloom filter is a probabilistic data structure that can quickly determine whether an element is in a set. In this example, the generated hash digest is filtered through the Bloom filter to remove duplicate hash values and retain the unique identification code. This can ensure that each electronic component has a unique identifier and avoid problems caused by data duplication.

[0086] Adopt the Merkle tree structure to hierarchically verify the set of unique identification codes. The Merkle tree is a binary tree structure that divides data into multiple leaf nodes and finally generates the hash value of the root node by calculating hash values layer by layer. In this system, the unique identification codes of electronic components are used as leaf nodes to construct a Merkle tree. Through the hierarchical verification of the tree, an immutable hash proof chain is generated. When it is necessary to verify whether the data of a certain electronic component has been tampered with, only the corresponding hash proof chain needs to be verified, improving the efficiency and security of data verification.

[0087] Example 5:

[0088] This embodiment details the calculation method of the timeliness weight factor and the update strategy of the reinforcement learning algorithm, whose function is to optimize inventory management and improve inventory turnover rate. The specific methods include:

[0089] Obtain the inventory receipt time, shelf life, and market circulation rate of electronic components. For example, the receipt time of a certain electronic component is January 1, 2023, the shelf life is 12 months, and the market circulation rate is obtained by statistical analysis of sales data over a period of time. Predict the remaining effective cycle of the inventory through the exponential smoothing model. The exponential smoothing model can predict future data based on historical data. In this embodiment, it predicts the remaining effective cycle of the electronic component according to information such as the inventory receipt time, shelf life, and market circulation rate. Construct a time-sensitive function and calculate the initial timeliness weight according to the ratio of the remaining effective cycle to the preset threshold. For example, the preset threshold is 6 months. If the remaining effective cycle of a certain electronic component is 3 months, the initial timeliness weight calculated according to the time-sensitive function is relatively low, indicating that the timeliness of this electronic component is strong and it needs to be processed preferentially. Introduce the Markov decision process to simulate the inventory state transition path and dynamically adjust the decay rate of the timeliness weight. The Markov decision process is a mathematical model used to describe stochastic dynamic systems. In this embodiment, it dynamically adjusts the decay rate of the timeliness weight according to changes in inventory, such as increases or decreases in inventory quantity and changes in sales conditions, so that the timeliness weight is more in line with the actual situation.

[0090] Define the inventory state as the observation space of the agent, the timeliness weight adjustment action as the action space, and the inventory turnover rate as the reward function. The agent observes the inventory state, selects an appropriate timeliness weight adjustment action, and obtains a reward according to the inventory turnover rate. Train the policy network through the Deep Deterministic Policy Gradient algorithm (DDPG) to generate an optimal timeliness weight update strategy. DDPG is a reinforcement learning algorithm based on policy gradients that combines the powerful representation ability of deep neural networks and can learn the optimal policy. In this system, by training the policy network with DDPG, the agent can select the optimal timeliness weight adjustment action according to the inventory state to achieve optimized management of the inventory. Use an experience replay pool to store historical state transition data and preferentially sample high-variance state segments for policy network parameter updates. The experience replay pool can break the temporal correlation of data and improve learning efficiency. Preferentially sampling high-variance state segments for policy network parameter updates allows the agent to learn effective policies faster and accelerate convergence to the optimal solution.

[0091] In addition, the system of the present invention further includes an abnormal transaction detection module, a multi-source data integration module, an interactive query optimization module, etc. The abnormal transaction detection module monitors the data flow of user transaction behaviors in real time, and uses the isolation forest algorithm and the federated learning framework to detect abnormal transactions to ensure transaction security; the multi-source data integration module interfaces with the third-party vendor database, and uses ontology alignment technology, transfer learning algorithms, and differential privacy technology to achieve the integration and privacy protection of multi-source data; the interactive query optimization module parses the user's natural language query statements, constructs a multi-objective optimization function, and generates an optimal query execution plan based on the quantum particle swarm algorithm to improve the query efficiency and quality. These modules cooperate with the above-mentioned core modules and the technical methods in the embodiments to jointly constitute a complete online sales data management and maintenance system for electronic components, providing comprehensive and efficient support for the online sales business of electronic components.

[0092] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0093] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An online sales data management and maintenance system for electronic components, characterized in that, Including: A data acquisition and processing module, which is used to acquire multi-dimensional sales data of electronic components, perform natural language processing on the sales data to obtain a structured data set, and construct the association relationship between entities in the structured data set through a dynamic knowledge graph; A user demand matching module, which generates a user demand feature vector based on a collaborative filtering algorithm, performs semantic clustering on the structured data set to obtain multiple candidate electronic component categories, and generates an initial matching result according to the semantic similarity between the user demand feature vector and each candidate electronic component category; A data verification module, which performs feature encoding on the electronic component attribute information in the initial matching result based on a hash algorithm to generate a set of unique identification codes, and performs distributed verification on the set of unique identification codes through a blockchain network to obtain a verified target electronic component data set; An inventory optimization module, which performs timeliness marking on the inventory information in the target electronic component data set according to a preset timeliness weight factor, and dynamically adjusts the update strategy of the timeliness weight factor based on a reinforcement learning algorithm to generate optimized inventory status data.

2. The online sales data management and maintenance system for an electronic component according to claim 1, characterized in that, The method for constructing the dynamic knowledge graph includes: Extracting the entity names, attribute descriptions, and transaction records in the structured data set, and identifying the implicit semantic relationships between entities through a bidirectional long short-term memory network model; Constructing an initial embedding representation of entity nodes based on a graph convolutional network algorithm, and calculating the weight value of the relationship edge according to the transaction frequency between entities and the time decay coefficient; Adopting a dynamic time window mechanism to incrementally update the knowledge graph. When new transaction data is detected, recalculate the embedding representation of the affected entity nodes and the weight value of the relationship edge.

3. An online sales data management and maintenance system for electronic components according to claim 1, characterized in that, The method for generating the user demand feature vector includes: Collecting the user's historical search records and order data, and converting the unstructured text into a semantic vector through a word embedding model; Extracting key demand features in the user behavior sequence based on an attention mechanism, and calculating the decay coefficient of the demand features in combination with timestamp information; Using a generative adversarial network to perform data augmentation on the sparse demand vector to generate a high-dimensional continuous user demand feature vector.

4. An online sales data management and maintenance system for electronic components according to claim 1, characterized in that, The implementation method of the semantic clustering includes: Performing dependency syntactic analysis on the electronic component description text in the structured data set, and extracting the functional parameter, package type, and electrical characteristic triples; Performing multi-modal feature fusion on the triples based on a density peak clustering algorithm to generate the core clustering clusters of candidate electronic component categories; Performing secondary partitioning on the core clustering clusters through a spectral clustering algorithm to eliminate the noise data points in the semantic overlapping area.

5. An online sales data management and maintenance system for an electronic component as claimed in claim 1, characterized in that, The feature encoding rules of the hash algorithm include: Concatenating the model number, production batch, and supplier information of the electronic component into an original string, and generating a hash digest of a fixed length through the SHA-3 algorithm; Performing duplicate removal processing on the hash digest based on a Bloom filter, and retaining the unique identification code; Using a Merkle tree structure to perform hierarchical verification on the set of unique identification codes to generate an immutable hash proof chain.

6. The online sales data management and maintenance system for an electronic component according to claim 1, characterized in that, The calculation method of the timeliness weight factor includes: Obtain the inventory receipt time, shelf life, and market circulation rate of electronic components, and predict the remaining effective cycle of the inventory through the exponential smoothing model; Construct a time-sensitive function and calculate the initial timeliness weight according to the ratio of the remaining effective cycle to the preset threshold; Introduce the Markov decision process to simulate the inventory state transition path and dynamically adjust the decay rate of the timeliness weight.

7. An online sales data management and maintenance system for electronic components according to claim 1, characterized in that, The update strategy of the reinforcement learning algorithm includes: Define the inventory state as the observation space of the agent, the timeliness weight adjustment action as the action space, and the inventory turnover rate as the reward function; Train the policy network through the deep deterministic policy gradient algorithm to generate the optimal timeliness weight update strategy; Use an experience replay pool to store historical state transition data and preferentially sample high-variance state segments for policy network parameter update.

8. An online sales data management and maintenance system for electronic components according to claim 1, characterized in that The system also includes an abnormal transaction detection module, and its implementation method includes: Real-time monitor the user transaction behavior data stream, and detect abnormal patterns of order amount, frequency, and geographical location through the isolation forest algorithm; Construct a distributed abnormal scoring model based on the federated learning framework, and aggregate the local gradients of multiple data nodes to update the global model parameters; When the abnormal score exceeds the dynamic threshold, trigger the transaction traceability mechanism and associate and query the historical transaction hash records in the blockchain network.

9. An online sales data management and maintenance system for electronic components as claimed in claim 1, wherein The system also includes a multi-source data integration module, and its implementation method includes: Connect to the third-party supplier database and eliminate semantic ambiguities of different data sources through ontology alignment technology; Based on the transfer learning algorithm, map the external data features to the embedding space of the local knowledge graph to generate a unified semantic representation; Use differential privacy technology to add noise to the integrated sensitive data to generate a fusion data set that meets the privacy protection requirements.

10. A data management and maintenance system for online sales of electronic components according to claim 1, characterized in that, The system also includes an interactive query optimization module, and its implementation method includes: Parse the user's natural language query statement and extract the query intent and constraint conditions through semantic role labeling; Construct a multi-objective optimization function to balance the query response time, data integrity, and computing resource consumption; Based on the quantum particle swarm algorithm, search for the optimal query execution plan and dynamically generate the index call sequence of the distributed database.

Images