An e-commerce export goods information search cloud platform based on big data

Abstract

The present invention proposes an e-commerce export goods information search cloud platform based on big data. The cloud platform includes a multi-dimensional dynamic data preprocessing system, a distributed blockchain traceability system, a multi-agent collaborative optimization system, an emotion-aware search optimization system, a user interface and interaction design system, and an interface integration system. The present invention provides a reliable data basis for subsequent intelligent analysis and decision-making through efficient data processing. By applying blockchain technology and a multi-agent collaborative optimization system in the e-commerce export goods information search cloud platform, it provides efficient order processing, ensures the intelligence and optimization effect of the supply chain system, optimizes search results and customer service quality through emotion awareness, improves customer satisfaction and loyalty, and thus better serves the overall goal of the patent. The present invention provides an intelligent search cloud platform for the service of e-commerce export goods through efficient data collection and preprocessing methods, multi-agent reinforcement learning technology, and blockchain technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of e-commerce export goods information search, and particularly relates to a cloud platform for e-commerce export goods information search based on big data. Background Art

[0002] In recent years, with the rapid development of the global e-commerce industry, the business scale and transaction volume of e-commerce platforms have increased significantly. However, the information management and search of e-commerce export goods still face many challenges. Traditional e-commerce platforms usually rely on decentralized data management systems, which lead to problems such as information inconsistency and low data processing efficiency. In the prior art, many e-commerce platforms adopt rule-based search engines, which cannot meet the needs of users for diverse and personalized search results. In addition, existing logistics scheduling and supply chain management systems mostly adopt static and single optimization methods, lacking the ability to dynamically adapt to complex and changing market environments, resulting in low logistics efficiency and high costs.

[0003] In terms of customer service, the prior art mainly relies on keyword-based automatic reply systems, which cannot accurately understand the emotions and needs of users, resulting in low customer satisfaction. Although some e-commerce platforms have tried to introduce sentiment analysis technology, it is usually limited to simple sentiment recognition and fails to fully utilize the sentiment analysis results to optimize customer service strategies. In addition, traditional supply chain management systems lack transparency and traceability, making it difficult to ensure the authenticity and integrity of goods information, which is particularly prominent in cross-border e-commerce. Although blockchain technology has been introduced into some supply chain systems, it has not been widely applied due to its complexity and high cost.

[0004] In summary, the prior art has the following main problems:

[0005] 1. The data management system is decentralized, resulting in information inconsistency and low processing efficiency;

[0006] 2. Traditional search engines cannot meet the needs of users for diverse and personalized search results;

[0007] 3. Existing logistics scheduling and supply chain management systems lack the ability to dynamically adapt, resulting in low logistics efficiency and high costs;

[0008] 4. The customer service system cannot accurately understand and respond to the emotions and needs of users, resulting in low customer satisfaction;

[0009] 5. The supply chain management system lacks transparency and traceability, making it difficult to ensure the authenticity and integrity of goods information. Summary of the Invention

[0010] The object of the present invention is to design an e-commerce export goods information search cloud platform based on big data, which provides an intelligent search cloud platform for the services of e-commerce export goods through efficient data collection and preprocessing methods, multi-agent reinforcement learning technology, and blockchain technology.

[0011] To achieve the above object, the present invention provides an e-commerce export goods information search cloud platform based on big data. The cloud platform includes a multi-dimensional dynamic data preprocessing system, a distributed blockchain traceability system, a multi-agent collaborative optimization system, an emotion-aware search optimization system, a user interface and interaction design system, and an interface integration system. Among them,

[0012] The multi-dimensional dynamic data preprocessing system is used to collect multi-dimensional data of e-commerce and preprocess the multi-dimensional data, integrate and synchronize the preprocessed data, fuse the integrated and synchronized data to obtain a data matrix, and store the fused data matrix in a distributed database. Among them, the multi-dimensional data includes user behavior data, order data, and logistics data. The user behavior data includes browsing records, click-through rates, and purchase histories. The order data includes order numbers, product IDs, quantities, prices, order times, and user IDs. The logistics data includes transportation routes, transportation times, warehouse locations, goods statuses, and delivery times.

[0013] The distributed blockchain traceability system is used to perform full-process traceability and verification processing on the data preprocessed by the multi-dimensional dynamic data preprocessing system.

[0014] The multi-agent collaborative optimization system is used to construct a multi-agent reinforcement learning model to optimize the supply chain management process of the e-commerce platform.

[0015] The emotion-aware search optimization system is used to construct an emotion analysis model to analyze and feedback the customer emotions of the e-commerce platform.

[0016] The user interface and interaction design system is used to perform interactive design on the interface of the cloud platform.

[0017] The interface integration system is used to perform system integration on the cloud platform.

[0018] Among them, the multi-agent reinforcement learning model is constructed as follows:

[0019] Model the supply chain management problem of the e-commerce platform as a multi-agent environment, where each agent represents a node in the supply chain, and define the state space as S i S i represents the state space of the i-th agent, and define the action space as A i A iThe action space of the $i$-th agent, with the state transition function $P(s'|s,a)$, represents the probability of transitioning from state $s$ to state $s'$ after performing action $a$; where, $S$ i $=\{s$ i1 , $s$ i2 , $\cdots s$ ij , $s$ in}$\}$, where $s$ ij represents the $j$-th state variable, and $s$ in represents the $n$-th state variable, $A$ i $=\{a$ i1 , $a$ i2 , $\cdots a$ ij , $a$ im}$\}$, where $a$ ij represents the $j$-th action, and $a$ im represents the $m$-th action;

[0020] Initialize the parameters of each agent, including the state space, action space, and reward value, and use a deep neural network to represent the agent's policy function $\pi$ i (a|s), the policy function $\pi$ i (a|s) represents the probability of selecting action $a$ in state $s$, and $\pi$ i (a|s; $\theta$ i ) represents the policy function of the $i$-th agent when selecting action $a$ in state $s$, and $\theta$ i is the parameter of the policy network;

[0021] Define the comprehensive reward function $R$ i (s,a) according to the optimization goal of the e-commerce platform's supply chain. $R$ i (s,a) represents the immediate reward obtained by the $i$-th agent after performing action $a$ in state $s$, which is expressed as follows:

[0022] $R$ i (s,a) = -$\alpha$C transport - $\beta$C inventory - rT delivery + $\delta$S customer

[0023] where, $C$ transport represents the transportation cost, $C$ inventory represents the inventory holding cost, $T$ delivery represents the delivery time, $S$ customer represents the customer satisfaction score, and $\alpha$, $\beta$, $r$, $\delta$ represent weight parameters used to adjust the importance of different goals;

[0024] Construct a globally shared experience replay pool Store the historical states, actions, rewards, and next states of all agents, which are expressed as follows:

[0025]

[0026] Among them, (s t , a t , r t , s t+1 ) represents the experience sample at time step t, s t represents the state at time step t, a t the action at time step t, r t represents the reward at time step t, s t+1 represents the state at time step t + 1;

[0027] Use the multi-agent deep Q-network to update the Q function Q i (s, a; θ i ), which represents the expected return for the agent to execute action a in state s, is expressed as follows:

[0028] Q i (s, a; θ i ) = E[r t + γ max a, Q i (s t+1 , a′; θ i )|s t = s, a t = a]

[0029] Among them, γ is the discount factor, representing the discount coefficient of future rewards; E represents the expectation, and a′ represents the action a′;

[0030] Use the multi-agent policy gradient algorithm to update the policy network parameters θ i for each agent, which is expressed as follows:

[0031]

[0032] Among them is the learning rate, J(θ i ) is the target value of the policy function, represents taking the gradient with respect to the parameter θ i ;

[0033] Design the objective function to maximize the expected rewards of all agents to obtain the optimal supply chain strategy for the e-commerce platform; among them, the objective function is expressed as follows:

[0034]

[0035] Among them, i represents the agent, and π i represents the policy of the i-th agent, represents the expected policy of all agents.

[0036] Furthermore, the browsing record includes the click time, page stay time, user ID, and product ID.

[0037] Furthermore, in the multi-dimensional dynamic data preprocessing system, real-time data flow technology is used to collect the real-time data flows of each data source; the preprocessing includes data deduplication and missing value processing, and data transformation.

[0038] Among them, for the data deduplication and missing values, an adaptive interpolation method is used to predict and fill according to the trend of the time series, which is expressed as follows:

[0039]

[0040] where x ij represents the current value; represents the value predicted based on the time series model, and the calculation is as follows:

[0041]

[0042] Among them, the weight parameters ρ, δ, τ of the prediction model are determined by minimizing the prediction error and satisfy ρ + δ + τ = 1; x i-1,j represents the data value of the previous time step, x i-2,j represents the data value of the two previous time steps, x i-3,j represents the data value of the three previous time steps;

[0043] The data transformation adopts the range normalization method, which is expressed as follows:

[0044]

[0045] where x ij ′ represents the normalized data value, min(x j ) and max(x j 0 are the minimum and maximum values of the feature x j respectively.

[0046] Furthermore, the preprocessed data is integrated and synchronized, and the integrated and synchronized data is fused to obtain a data matrix, and the fused data matrix is stored in a distributed database. Specifically, it includes:

[0047] Use the order ID as the primary key to uniquely identify each record, and define the order ID as ID;

[0048] Define the priority of the data sources: order data O′ > user behavior data U′ > logistics data L′, and merge each record according to the priority, which is expressed as follows:

[0049] D = U′ ∪ O′ ∪ L′

[0050] Among them, O′ represents order data, U′ represents user behavior data, L′ represents logistics data, and D represents a data matrix;

[0051] The integrated data is stored in a distributed database, which is expressed as follows:

[0052] S = Cassandra(D)

[0053] Among them, S represents the final integrated data stored in the distributed database, and Cassandra represents the distributed database system for storing data.

[0054] Furthermore, the full-process traceability and verification process includes the following steps:

[0055] Generate blocks based on the preprocessed data. Among them, each block contains multiple records, and each record corresponds to a link in the supply chain;

[0056] Generate a hash value for the generated block data through an encryption algorithm and upload it to the chain;

[0057] Design and deploy a smart contract to automatically execute the business logic in the supply chain; among them, the business logic is that when a new product is put into storage, the inventory data is automatically updated, and at the same time, according to the order information, the inventory is automatically allocated and a transportation plan is generated. According to the real-time logistics information and the transportation plan, the transportation route is dynamically adjusted;

[0058] Regularly compare the data hash value stored on the blockchain with the current data hash value to verify the data consistency. When the data changes, the data on the blockchain is automatically updated through the smart contract;

[0059] Users query the supply chain information of specific products through a blockchain browser or API interface and / or the cloud platform automatically queries and analyzes the supply chain data regularly to generate reports and visualization charts.

[0060] Furthermore, the construction of the sentiment analysis model uses a convolutional neural network combined with a bidirectional long short-term memory network BiLSTM to perform sentiment classification training on text data; among them, the structure of the constructed sentiment analysis model is expressed as follows:

[0061] Sentiment classifier(d i ) = σ(W2 · (BiLSTM(W1 · x i )) + b20

[0062] Among them, d i represents the i-th piece of customer text data, x i represents the text vector, W1 and W2 respectively represent model parameters, σ represents the activation function, and b2 represents the bias vector.

[0063] Furthermore, in the emotion-aware search optimization system, for each piece of customer text data d i perform an emotion scoring S(d i ), and calculate its emotion tendency value, which is expressed as follows:

[0064]

[0065] where score pos (t) and score neg (t) respectively represent the positive and negative scores of the term t, and TF-IDF(t, d i ) represents the frequency of the term t appearing in the customer text data d i ;

[0066] Introduce a multi-emotion dimension analysis model, divide the customer emotion into multiple dimensions and calculate the comprehensive emotion tendency value S multi (d i ), which is expressed as follows:

[0067]

[0068] where score k (t) represents the score of the term t on the emotion dimension k, and w k represents the weight of the emotion dimension;

[0069] Monitor the emotion changes of customer text data in real time and identify emotion mutation points, which is expressed as follows:

[0070] ΔS(d i ) = S(d i ) - S(d i-1 )

[0071] When ΔS(d i ) > χ, trigger an emotion mutation alarm, where χ is a preset threshold.

[0072] Introduce emotion trend analysis, predict the customer emotion trend through the time series model ARIMA, and identify emotion fluctuations in advance, which is expressed as follows:

[0073]

[0074] When , trigger an emotion trend alarm, where θ trend is the trend warning threshold, and ARIMA represents the time series model.

[0075] Furthermore, in the emotion-aware search optimization system, for the customer text data with emotion mutation, automatically generate a feedback processing plan, which is expressed as follows:

[0076] Processing solution (d i ) = Policy library (S(d i ))

[0077] Among them, the policy library contains preset processing policies, and the best policy is selected for response according to the sentiment score;

[0078] Utilize the sentiment analysis result to optimize the search algorithm, and adjust the search result sorting and recommended content according to the customer sentiment, which is expressed as follows:

[0079] Ranking score (q i ) = Relevance score (q i ) + λS(d i )

[0080] Among them, q i is the search query, and λ is the sentiment weight parameter;

[0081] For customers with sudden sentiment changes, automatically push personalized customer service support and services to improve customer satisfaction, which is expressed as follows:

[0082] Customer service solution (d i ) = Customer service policy library (S multi (d i ))

[0083] Among them, the customer service policy library contains the best customer service response solutions in different sentiment states.

[0084] Furthermore, in the user interface and interaction design system, design the user path so that the user can reach the required page with the fewest clicks, and optimize the path length L to minimize the click count C, which is expressed as follows:

[0085]

[0086] Among them, C i represents the click count in the i-th path, and n represents the total number of paths;

[0087] Use the grid layout to design the interface layout, divide the interface into several areas, and each area displays different types of information. The calculation of the number of areas N of the grid layout is as follows:

[0088] N = Rows × Columns

[0089] Among them, Rows represents the number of rows, and Columns represents the number of columns.

[0090] Furthermore, the system integration of the cloud platform uses the HTTP / HTTPS protocol for inter-system communication.

[0091] The beneficial technical effects of the present invention are at least as follows:

[0092] (1) The present invention proposes a method called the Multi-dimensional Dynamic Data Preprocessing System (MDDP), which conducts systematic preprocessing for the special data structure and characteristics of e-commerce platforms. Through efficient data collection and preprocessing methods, it provides a reliable data foundation for subsequent intelligent analysis and decision-making. The present invention forms an organic whole from data collection, cleaning, transformation, integration to storage, and provides a reliable data foundation for subsequent intelligent analysis and decision-making through efficient data processing, meeting the special data processing requirements of e-commerce platforms and laying a solid foundation for the implementation of this patent.

[0093] (2) The present invention forms an organic whole from data block generation, blockchain uploading, smart contract execution to data verification and query. Through blockchain technology, it ensures the transparency and immutability of data, provides full-process traceability and authenticity verification for the supply chain management system, meets the special data processing requirements of e-commerce platforms, and lays a solid foundation for the implementation of this patent.

[0094] (3) The adaptive learning model of the present invention can be dynamically adjusted according to the commodity characteristics and user behavior data in different markets. Through the adaptive learning algorithm, it self-optimizes the classification criteria, improves the accuracy and adaptability of classification, and thus better meets the dynamic needs of cross-border e-commerce platforms. This innovation solves the problem that existing methods lack the ability of dynamic adjustment and cannot be optimized according to market changes.

[0095] (4) The multi-agent collaborative optimization system of the present invention can provide precise logistics scheduling, intelligent inventory management and efficient order processing in the e-commerce export goods information search cloud platform, ensuring the intelligence and optimization effect of the entire supply chain system.

[0096] (5) The emotion analysis-driven customer service system of the present invention can better respond to customer needs and optimize the overall service quality and customer satisfaction of the e-commerce export goods information search cloud platform. The emotion-aware customer service system can provide real-time emotion monitoring and intelligent feedback processing in the e-commerce export goods information search cloud platform based on big data, optimize search results and customer service quality, enhance customer satisfaction and loyalty, and thus better serve the overall goal of the patent. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0098] Figure 1 It is a framework diagram of an e-commerce export goods information search cloud platform based on big data according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0099] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0100] In one or more embodiments, as Figure 1 shown, a big data-based e-commerce export goods information search cloud platform is disclosed, including the following systems:

[0101] The multi-dimensional dynamic data preprocessing system 101 is used to collect multi-dimensional data of e-commerce, preprocess the multi-dimensional data, integrate and synchronize the preprocessed data, fuse the integrated and synchronized data to obtain a data matrix, and store the fused data matrix in a distributed database; wherein, the multi-dimensional data includes user behavior data, order data and logistics data; the user behavior data includes browsing records, click-through rates and purchase histories; the order data includes order numbers, product IDs, quantities, prices, order times and user IDs; the logistics data includes transportation routes, transportation times, warehouse locations, goods statuses and delivery times.

[0102] Among them, the browsing records include click times, page stay times, user IDs and product IDs.

[0103] Specifically, real-time data stream technology is used to collect real-time data streams from each data source; the preprocessing includes data deduplication and missing value processing and data conversion. The specific data stream schematic diagram is as follows:

[0104] Kafka→{U,O,L}(1)

[0105] Among them, Kafka is responsible for the real-time transmission and processing of data streams, and U, O, and L respectively represent user behavior data, order data and logistics data.

[0106] Among them, the data deduplication and missing values are predicted and filled using an adaptive interpolation method according to the trend of the time series, as shown below:

[0107]

[0108] where x ij represents the current value; represents the value predicted based on the time series model, and the calculation is as follows:

[0109]

[0110] Among them, the weight parameters ρ, δ, and τ of the prediction model are determined by minimizing the prediction error and satisfy ρ + δ + τ = 1; x i-+1,j represents the data value at the previous time step, x i-2,j represents the data value at the two previous time steps, x i-3,j represents the data value at the three previous time steps;

[0111] The data conversion adopts the Range Normalization Method (RNM), which is expressed as follows:

[0112]

[0113] Among them, x ij ′ represents the normalized data value, min(x j ) and max(x j ) are the minimum and maximum values of the feature x j respectively. Through data normalization, the influence brought by the difference in dimension is eliminated.

[0114] Furthermore, the preprocessed data is integrated and synchronized, and the integrated and synchronized data is fused to obtain a data matrix, and the fused data matrix is stored in a distributed database, and the processed multi-source data is integrated into a unified database to ensure data consistency. The multi-source data fusion algorithm (Multi-source Data Fusion Algorithm, MDFA) is used to integrate the data from different data sources into a unified data matrix, specifically including:

[0115] Use the order ID as the primary key to uniquely identify each record, and define the order ID as ID;

[0116] Define the priority of the data sources: order data O′ > user behavior data U′ > logistics data L′. For different fields of the same record, select the latest update or merge the field values according to the preset rules, and merge each record according to the priority, which is expressed as follows:

[0117] D = U′ ∪ O′ ∪ L′ (5)

[0118] Among them, O′ represents the order data, U′ represents the user behavior data, L′ represents the logistics data, and D represents the data matrix; in the specific implementation, the normalized user behavior data, order data, and logistics data matrices are integrated to ensure the unity and consistency of the data.

[0119] Specifically, the data merging matrix:

[0120] User behavior data matrix U′:

[0121]

[0122] Order data matrix O′:

[0123]

[0124] Logistics data matrix L′:

[0125]

[0126] Use the unique identifier ID to merge the data and construct the final data matrix D. If the same record exists in multiple data sources, the data is retained according to the priority. For different fields of the same record, select the latest updated value or merge according to the rules. It is expressed as follows:

[0127]

[0128] Data fusion formula:

[0129] D = U′ ∪ O′ ∪ L′(10)

[0130] The data matrix D represents:

[0131]

[0132] where d ij is the final value selected according to the merging rules.

[0133] Finally, store the integrated data in a distributed database to ensure data scalability and efficient query. Use the distributed database Apache Cassandra to store the processed data:

[0134] S = Cassandra(D)(12)

[0135] where S represents the final integrated data stored in the distributed database, and Cassandra represents the distributed database system used to store the data. In the specific implementation, by defining the data storage architecture, ensure data redundancy and high availability, and achieve efficient storage and query of data through the replication factor and data sharding technology.

[0136] Through the above steps, the present invention forms an organic whole from data collection, cleaning, transformation, integration to storage, provides a reliable data basis for subsequent intelligent analysis and decision-making through efficient data processing, meets the special requirements of data processing on the e-commerce platform, and lays a solid foundation for the implementation of this patent.

[0137] The distributed blockchain traceability system 102 is used to perform full-process traceability and verification processing on the data preprocessed by the multi-dimensional dynamic data preprocessing system.

[0138] Specifically, the preprocessed data in the multi-dimensional dynamic data preprocessing system 101 is further organized into blocks for chain uploading. Each block contains multiple records, and each record corresponds to a link in the supply chain, such as product production, warehousing, transportation, etc. Among them, the generation of blocks follows the following format:

[0139] B k ={R k1 ,R k2 ,…,R kn}(13)

[0140] Among them, B k represents the k-th block, and R ki represents the i-th record in the block. The i-th record contains the following fields:

[0141] t i represents the timestamp, recording the time when the data is generated;

[0142] d i represents the data content, such as product information, location, etc.;

[0143] h k-1 represents the hash value of the previous block for chained connection;

[0144] Each block is associated with the hash value of a previous block to ensure the integrity and immutability of the data chain.

[0145] Furthermore, the generated block data is hashed through an encryption algorithm and uploaded to the chain. The specific steps are as follows:

[0146] Perform a hash operation on the data of each block to obtain the hash value H(B k );

[0147] Through digital signature technology, sign the block to ensure the immutability of the data, which is expressed as follows:

[0148] σ k =Sign private (H(B k ))(14)

[0149] Among them, σ k represents the signature of the k-th block, and Sign private represents the private key signature operation.

[0150] Then upload the block data with the hash value and signature to the chain to ensure the public transparency of the data. The specific operations include using a consensus algorithm (such as PoW or PoS) to verify the block to ensure its legality and adding the block to the blockchain to form a chained structure.

[0151] Next, design and deploy smart contracts to automatically execute the business logic in the supply chain. For example, inventory updates, order processing, transportation scheduling, etc., are as follows:

[0152] Inventory update: When new products are put into storage, automatically update the inventory data;

[0153] Order processing: According to the order information, automatically allocate inventory and generate a transportation plan;

[0154] Transportation scheduling: Dynamically adjust the transportation route according to real-time logistics information.

[0155] Furthermore, in order to ensure the authenticity and integrity of the data, it is necessary to regularly verify and update the data on the blockchain. The main steps are as follows:

[0156] Data verification: Verify the consistency of the data by comparing the data hash value stored on the blockchain with the current data hash value.

[0157] Data update: When the data changes, automatically update the data on the blockchain through the smart contract. The specific steps are as follows:

[0158] Generate a new data record and calculate the hash value, sign the new data record, and add it to the blockchain:

[0159]

[0160] Furthermore, users or systems can obtain the full-process traceability information of the supply chain by querying the blockchain, including

[0161] User query: Users query the supply chain information of specific products through a blockchain browser or API interface;

[0162] System query: The system automatically queries and analyzes the supply chain data regularly, generates reports and visualization charts.

[0163] Through the above steps, the present invention forms an organic whole from data block generation, uploading to the chain, smart contract execution to data verification and query. Through blockchain technology, it ensures the transparency and immutability of the data, provides full-process traceability and authenticity verification for the supply chain management system, meets the special requirements of e-commerce platform data processing, and lays a solid foundation for the implementation of this patent.

[0164] The multi-agent collaborative optimization system 103 is used to construct a multi-agent reinforcement learning model to optimize the strategy of the supply chain management process of the e-commerce platform.

[0165] Specifically, the multi-agent reinforcement learning model is constructed as follows:

[0166] Model the supply chain management problem of an e-commerce platform as a multi-agent environment, where each agent represents a node in the supply chain, and define the state space as S i , S i represents the state space of the i-th agent, and define the action space as A i , A i the action space of the i-th agent, and the state transition function is P(s′|s,a), which represents the probability of transitioning from state s to state s′ after performing action a; where, S i = {s i1 , s i2 , … s ij , s in}, where s ij represents the j-th state variable, s in represents the n-th state variable, A i = {a i1 , a i2 , … a ij , a im}, where a ij represents the j-th action, a im represents the m-th action;

[0167] Initialize the parameters of each agent, including the state space, action space, and reward value, and use a deep neural network to represent the policy function π i (a|s), the policy function π i (a|s) represents the probability of choosing action a in state s, π i (a|s; θ i ) represents the policy function of the i-th agent when choosing action a in state s, θ i is the parameter of the policy network;

[0168] Define the comprehensive reward function R i (s,a) according to the optimization goal of the supply chain of the e-commerce platform, R i (s,a) represents the immediate reward obtained by the i-th agent after performing action a in state s, which is expressed as follows:

[0169] R i (s,a) = -αC transport - βC inventory - rT delivery + δS customer (16)

[0170] where, C transport represents the transportation cost, C inventory represents the inventory holding cost, T delivery represents the delivery time, S customerDenote the customer satisfaction score, and use α, β, r, δ as weight parameters to adjust the importance of different objectives;

[0171] Construct a globally shared experience replay pool Store the historical states, actions, rewards, and next states of all agents, which are represented as follows:

[0172]

[0173] Among them, (s t , a t , r t , s t+1 ) represents the experience sample at time step t, s t represents the state at time step t, a t the action at time step t, r t represents the reward at time step t, s t+1 represents the state at time step t + 1;

[0174] Use the multi-agent deep Q-network to update the Q-function Q i (s, a; θ i ), which represents the expected return for the agent to execute action a in state s, and is represented as follows:

[0175] Q i (s, a; θ i ) = E[r t + γ max a′ Q i (s t+1 , a′; θ i ) | s t = s, a t = a]

[0176] Among them, γ is the discount factor, representing the discount coefficient of future rewards; E represents the expectation, and a′ represents the action a′;

[0177] Use the multi-agent policy gradient algorithm to update the policy network parameters θ i for each agent, which is represented as follows:

[0178]

[0179] Among them is the learning rate, J(θ i ) is the target value of the policy function, represents taking the gradient with respect to the parameter θ i ;

[0180] Design the objective function to maximize the expected rewards of all agents to obtain the optimal supply chain strategy for the e-commerce platform; among them, the objective function is represented as follows:

[0181]

[0182] Among them, \(i\) represents an agent, and \(\pi\) i represents the policy of the \(i\)-th agent, and \(\overline{\pi}\) represents the expected policy of all agents.

[0183] The emotion-aware search optimization system 104 is used to build an emotion analysis model to analyze and feedback the emotions of customers on the e-commerce platform.

[0184] Specifically, first collect text data such as search queries, comments, chat records, feedback forms, and social media interactions of customers on the e-commerce platform as the basic data for emotion analysis, \(D = \{d_1, d_2, \ldots, d_n\}\), where \(d_i\) n represents the \(i\)-th piece of customer text data. i

[0185] Furthermore, clean, tokenize, and remove stop words from the text data. Use TF-IDF to represent the text in vector form:

[0186] TF-IDF(t, d) = TF(t, d) × IDF(t) (17)

[0187] where TF(t, d) is the frequency of the term \(t\) in the document \(d\), and IDF(t) is the inverse document frequency of the term \(t\).

[0188] Furthermore, the built emotion analysis model uses a convolutional neural network combined with a bidirectional long short-term memory network BiLSTM to perform emotion classification training on the text data; first build an emotion dictionary, including positive and negative words:

[0189] L_p pos = \{w_1, w_2, \ldots, w_m\}, L_n m = \{w_1, w_2, \ldots, w_k\} neg = \{w_1, w_2, \ldots, w_m\}, L_n n = \{w_1, w_2, \ldots, w_k\}

[0190] L_p pos and \(L_n\) neg represent the sets of positive and negative words respectively.

[0191] Build the emotion analysis model structure, which is expressed as follows:

[0192] Emotion classifier(d_i) i ) = \(\sigma(W_2 \cdot (BiLSTM(W_1 \cdot x_i)) + b_2)\) (18) i

[0193] where \(d_i\) i represents the \(i\)-th piece of customer text data, and \(x_i\)​​i Represents the text vector, W1 and W2 represent the model parameters respectively, σ represents the activation function, and b2 represents the bias vector.

[0194] Specifically, the structure of the sentiment classifier is as follows:

[0195] Convolutional layer: Extracts local features in the text to improve the accuracy of sentiment classification.

[0196] Bidirectional LSTM layer: Captures the context information in the text to enhance the model's understanding of sentiment.

[0197] Fully connected layer: Combines the outputs of the convolutional layer and the LSTM layer for final sentiment classification.

[0198] Furthermore, in the sentiment-aware search optimization system, the sentiment of the customer is detected and feedback is provided:

[0199] For each piece of customer text data d i A sentiment score S(d i ) is calculated, and its sentiment tendency value is expressed as follows:

[0200]

[0201] where score pos (t) and score neg (t) represent the positive and negative scores of the term t respectively, and TF-IDF(t, d i ) represents the frequency of the term t appearing in the customer text data d i ;

[0202] To improve the accuracy and multi-dimensionality of the sentiment score, a multi-sentiment dimension analysis model is introduced. The customer sentiment is divided into multiple dimensions (such as positive, negative, neutral), and the comprehensive sentiment tendency value S multi (d i ) is calculated, which is expressed as follows:

[0203]

[0204] where score k (t) represents the score of the term t on the sentiment dimension k, and w k represents the weight of the sentiment dimension;

[0205] The sentiment changes in the customer text data are monitored in real time to identify the sentiment mutation points, which are expressed as follows:

[0206] ΔS(d i ) = S(d i ) - S(d i-1 ) (21)

[0207] When ΔS(d i ) > χ, an emotional mutation alarm is triggered, where χ is a preset threshold.

[0208] Emotional trend analysis is introduced to predict the customer emotional trend through a time series model and identify emotional fluctuations in advance, as shown below:

[0209]

[0210] When , an emotional trend alarm is triggered, where θ trend is the trend warning threshold, and ARIMA represents the time series model.

[0211] As a preferred example of the present invention, it further includes:

[0212] For the customer text data with emotional mutations, a feedback processing solution is automatically generated:

[0213] Processing solution(d i ) = Policy library(S(d i )) (23)

[0214] Among them, the policy library contains preset processing strategies, and appropriate strategies are selected for response according to the emotional score.

[0215] The search algorithm is optimized using the emotional analysis results, and the search result ranking and recommended content are adjusted according to the customer's emotion:

[0216] Ranking score(q i ) = Relevance score(q i ) + λS(d i )(24)

[0217] Among them, q i is the search query, and λ is the emotional weight parameter.

[0218] For customers with emotional mutations, personalized customer service support and services are automatically pushed to improve customer satisfaction:

[0219] Customer service solution(d i ) = Customer service policy library 9S multi (d i ))(25)

[0220] Among them, the customer service policy library contains the best customer service response solutions in different emotional states to ensure quick response and personalized services.

[0221] Through these innovative measures, the customer service system driven by emotional analysis can better respond to customer needs and optimize the overall service quality and customer satisfaction of the e-commerce export goods information search cloud platform.

[0222] Through the above steps, the emotion-aware customer service system can provide real-time emotion monitoring and intelligent feedback processing in the e-commerce export goods information search cloud platform based on big data, optimize the search results and customer service quality, enhance customer satisfaction and loyalty, and thus better serve the overall goal of the patent.

[0223] The user interface and interaction design system 105 is used to perform interaction design on the interface of the cloud platform.

[0224] Specifically, it is necessary to classify and organize the information on the platform so that users can quickly find the information they need.

[0225] First, divide the information into several main categories: search results, product details, inventory information, logistics information, and customer feedback. Define the content and display methods of each category to ensure a clear hierarchy of information. Use a tree structure to represent the information hierarchy, enabling users to drill down to the required information through the navigation menu layer by layer. Design the user path so that users can reach the required page with the fewest clicks. Optimize the path length L to minimize the click count C:

[0226]

[0227] where C i represents the click count in the i-th path, and n is the total number of paths.

[0228] Then, use a grid layout to divide the interface into several areas, and each area displays different types of information. The calculation formula for the number of areas N in the grid layout:

[0229] N = Rows × Columns (27)

[0230] where Rows is the number of rows and Columns is the number of columns, and the specific values are adjusted according to the screen size and the amount of information.

[0231] Furthermore, design the key areas:

[0232] Top navigation bar: Place the main navigation menu and search bar;

[0233] Left sidebar: Display the classification navigation;

[0234] Central content area: Display the main information, such as search results, product details, etc.;

[0235] Right sidebar: Display additional information, such as recommended products and customer feedback.

[0236] Then, design the interactive elements so that users can obtain and process information through simple operations. Design clear and easy-to-click buttons and place them in the key operation areas. The button size Bs and the spacing B m Calculation formula:

[0237] B s = 1.5 × F w (28)

[0238] B m = 0.5 × F w (29)

[0239] Wherein, F w is the width of the user's finger, usually about 10 mm. Design a search bar with an auto-complete function to help users quickly input and find information. The calculation formula for the auto-complete recommendation item A r is:

[0240] A r = Match(I, K) × Freq(K)(30)

[0241] Wherein, Match(I, K) is the matching degree between the input I and the words in the thesaurus K, and Freq(K) is the occurrence frequency of the words in the thesaurus.

[0242] Use a card-style design to display search results. Each card contains key information such as product pictures, names, prices, etc. The card size C s Calculation formula:

[0243]

[0244] Wherein, W is the screen width, C is the number of columns, and M is the card spacing. The specific values are adjusted according to the screen size and the amount of information.

[0245] Finally, conduct usability tests on the designed interface and interaction to ensure that users can use it efficiently. Select representative users to cover different ages, occupations, and usage habits. Define test tasks such as searching for products, viewing details, and querying inventory.

[0246] Observe the user's operation process and record the operation time, number of clicks, and operation errors. The calculation formula for the user operation efficiency E is:

[0247]

[0248] Wherein, T s is the number of successful tasks, and T t is the total number of tasks.

[0249] Analyze the test data to identify common problems and bottlenecks. Optimize the interface and interaction design based on the analysis results to enhance the user experience. Through the above steps, the user interface and interaction design solution can help users of the e-commerce export goods information search cloud platform efficiently obtain and process information, thus better serving the overall goal of the patent.

[0250] The interface integration system 106 is used for system integration of the cloud platform.

[0251] Specifically, define the interfaces between each module to ensure smooth communication and data exchange between modules. Use a unified data format D for interface definition:

[0252] D = {d1, d2, …, d n}(33)

[0253] where d i represents the i-th data item.

[0254] Preferably, use the HTTP / HTTPS protocol for inter-module communication to ensure the security and efficiency of data transmission. Use JSON for serialization and deserialization of the request and response formats. Define the interface functions of each module to ensure clear data interaction and call logic between modules. Example interface function:

[0255] {

[0256] "function": "get_data",

[0257] "parameters": {

[0258] "query": "product_id",

[0259] "filters": ["date_range", "location"]

[0260] }

[0261] }

[0262] Preferably, integrate each system into a unified system framework to achieve seamless connection and data flow between systems. Adopt a microservices architecture and deploy and manage each system as an independent service.

[0263] Preferably, ensure smooth data flow between systems without data loss and delay. Use a queue mechanism (such as RabbitMQ) to manage data transmission between modules.

[0264] Preferably, use a load balancer (such as NGINX) to distribute user requests and balance the load of each server. Load balancing algorithm:

[0265]

[0266] Among them, W i is the weight of the i-th server, R i is the number of requests of the server, and C i is the processing capacity of the server.

[0267] Furthermore, use a caching mechanism (such as Redis) to store frequently accessed data, reduce the number of database queries, and improve the response speed. The formula for the cache hit rate H is:

[0268]

[0269] Among them, C hit is the number of cache hits, and C total is the total number of accesses.

[0270] Furthermore, optimize the database query statements, create indexes, and improve the query efficiency. Index selection algorithm:

[0271]

[0272] Among them, I s is the index selection, Q t is the query time, and T r is the number of records.

[0273] Finally, conduct stability tests on the system to ensure the stable operation of the system under different load conditions.

[0274] Furthermore, use a stress testing tool (such as JMeter) to simulate a large number of user requests and test the performance of the system under high load. Record the system response time T r and the success rate S r :

[0275]

[0276] Among them, R s is the number of successful requests, and R t is the total number of requests.

[0277] Preferably, simulate system failures and test the system's fault recovery ability. Record the system recovery time T f and the data integrity D i :

[0278]

[0279] Among them, D r is the amount of recovered data, and D t is the total amount of data.

[0280] Result analysis: Analyze the test data to identify system bottlenecks and failure points. Optimize the system architecture and modules based on the analysis results to improve the stability and performance of the system. Through the above steps, the system integration and optimization solution can ensure the efficient and stable operation of the e-commerce export goods information search cloud platform based on big data, thus better serving the overall goal of the patent.

[0281] The above are the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. An e-commerce export goods information search cloud platform based on big data, characterized in that, The cloud platform includes a multi-dimensional dynamic data preprocessing system, a distributed blockchain traceability system, a multi-agent collaborative optimization system, an emotion-aware search optimization system, a user interface and interaction design system, and an interface integration system. Among them, The multi-dimensional dynamic data preprocessing system is used to collect multi-dimensional data of e-commerce, preprocess the multi-dimensional data, integrate and synchronize the preprocessed data, fuse the integrated and synchronized data to obtain a data matrix, and store the fused data matrix in a distributed database. Among them, the multi-dimensional data includes user behavior data, order data, and logistics data. The user behavior data includes browsing records, click-through rates, and purchase histories. The order data includes order numbers, product IDs, quantities, prices, order times, and user IDs. The logistics data includes transportation routes, transportation times, warehouse locations, cargo statuses, and delivery times. The distributed blockchain traceability system is used to perform full-process traceability and verification processing on the data preprocessed by the multi-dimensional dynamic data preprocessing system. The multi-agent collaborative optimization system is used to build a multi-agent reinforcement learning model to optimize the supply chain management process of the e-commerce platform. The emotion-aware search optimization system is used to build an emotion analysis model to analyze and feedback the customer emotions of the e-commerce platform. The user interface and interaction design system is used to perform interaction design on the interface of the cloud platform. The interface integration system is used to perform system integration on the cloud platform. Among them, the multi-agent reinforcement learning model is built as follows: Model the supply chain management problem of an e-commerce platform as a multi-agent environment, where each agent represents a node in the supply chain, and define the state space as S i , S i represents the state space of the i-th agent, and define the action space as A i , A i is the action space of the i-th agent, and the state transition function is P(s′|s,a), which represents the probability of transitioning from state s to state s′ after executing action a; where, S i = {s i1 , s i2 , … s ij , s in}, where s ij represents the j-th state variable, s in represents the n-th state variable, A i = {a i1 , a i2 , … a ij , a im}, where a ij represents the j-th action, a im represents the m-th action; Initialize the parameters of each agent, including the state space, action space, and reward value, and use a deep neural network to represent the policy function π of the agent i (a|s), the policy function π i (a|s) represents the probability of choosing action a in state s, π i (a|s; θ i ) represents the policy function of the i-th agent when choosing action a in state s, θ i are the parameters of the policy network; Define the comprehensive reward function \(R\) according to the optimization objectives of the supply chain of the e-commerce platform i \(R(s,a)\) i \(R(s,a)\) represents the immediate reward obtained by the \(i\)-th agent after executing action \(a\) in state \(s\), which is expressed as follows: R i (s,a) = -αC transport -βC inventory -rT delivery +δS customer Among them, C transport represents the transportation cost, C inventory represents the inventory holding cost, T delivery represents the delivery time, S customer represents the customer satisfaction score, and α, β, r, δ represent weight parameters used to adjust the importance of different objectives; Build a globally shared experience replay pool Store the historical states, actions, rewards, and next states of all agents, as shown below: Among them, (s t , a t , r t , s t+1 ) represents the experience sample at time step t, s t represents the state at time step t, a t the action at time step t, r t represents the reward at time step t, s t+1 represents the state at time step t + 1; Update the Q function Q using the multi-agent deep Q-network i (s, a; θ i ), which represents the expected return for the agent to execute action a in state s, is expressed as follows: Q i (s, a; θ i ) = E[r t + γ max a′ Q i (s t+1 , a′; θ i ) | s t = s, a t = a] Among them, γ is the discount factor, representing the discount coefficient of future rewards; E represents expectation, and a′ represents action a′. Using the multi-agent policy gradient algorithm, update the policy network parameters θ for each agent, which is expressed as follows: i Update as follows: where is the learning rate, and J(θ i ) is the target value of the policy function, denotes the gradient with respect to the parameter θ i ; The design objective function maximizes the expected rewards of all agents to obtain the optimal supply chain strategy for the e-commerce platform. Among them, the objective function is expressed as follows: where \(i\) represents the agent, and \(\pi\) i represents the policy of the \(i\)-th agent, and \(\overline{\pi}\) represents the expected policy of all agents; In the emotion perception search optimization system, for each piece of customer text data d i perform an emotion score S(d i ), and calculate its emotion tendency value, which is expressed as follows: Among them, score pos (t) and score neg (t) respectively represent the positive and negative scores of term t, and TF-IDF(t, d i ) represents the frequency of term t appearing in the customer text data d i ; Introduce a multi-emotional dimension analysis model, divide the customer's emotions into multiple dimensions and calculate the comprehensive emotional tendency value S multi (d i ) is expressed as follows: Among them, score k (t) represents the score of term t on the sentiment dimension k, and w k represents the weight of the sentiment dimension; Real-time monitor the emotional changes of customer text data and identify emotional mutation points, which are expressed as follows: ΔS(d i ) = S(d i ) - S(d i-1 ) When ΔS(d i ) > χ, an emotional mutation alarm is triggered, where χ is a preset threshold value; Introduce emotion trend analysis, predict the customer emotion trend through the time series model ARIMA, and identify emotional fluctuations in advance, which are expressed as follows: When , trigger the emotional trend alarm, where θ trend is the trend warning threshold, and ARIMA represents the time series model; In the emotion-aware search optimization system, for the customer text data with emotional mutations, automatically generate a feedback processing plan, which is expressed as follows: Processing solution (d i ) = Policy library (S(d i )) Among them, the policy library contains preset processing policies, and the best policy is selected according to the emotion score for response. Optimize the search algorithm using the emotion analysis results, and adjust the search result sorting and recommended content according to the customer emotions, which are expressed as follows: Ranking score(q i ) = Relevance score(q i ) + λS(d i ) where q i is the search query, and λ is the sentiment weight parameter; For customers with emotional mutations, automatically push personalized customer service support and services to improve customer satisfaction, which are expressed as follows: Customer service plan (d i ) = Customer service strategy library (S multi (d i )) Among them, the customer service policy library contains the best customer service response plans in different emotional states.

2. The cloud platform for searching e-commerce export goods information based on big data according to claim 1, wherein The browsing records include click times, page stay times, user IDs, and product IDs.

3. A cloud platform for searching e-commerce export goods information based on big data according to claim 1, characterized in that, In the multi-dimensional dynamic data preprocessing system, real-time data stream technology is used to collect real-time data streams from each data source. The preprocessing includes data deduplication and missing value processing and data conversion. Among them, the data deduplication and missing values are predicted and filled using an adaptive interpolation method according to the trend of the time series, which is expressed as follows: where x ij represents the current value; represents the value predicted based on the time series model and is calculated as follows: Among them, the weight parameters ρ, δ, τ of the prediction model are determined by minimizing the prediction error and satisfy ρ + δ + τ = 1; x i-1,j represents the data value at the previous time step, x i-2,j represents the data value at the two previous time steps, x i-3,j represents the data value at the three previous time steps; The data conversion uses the range normalization method, which is expressed as follows: where x ij ' represents the normalized data value, and min(x j ) and max(x j ) are the minimum and maximum values of the feature x j , respectively.

4. The e-commerce export goods information search cloud platform based on big data according to claim 3, characterized in that, Integrate and synchronize the preprocessed data, fuse the integrated and synchronized data to obtain a data matrix, and store the fused data matrix in a distributed database, which specifically includes: Use the order ID as the primary key to uniquely identify each record, and define the order ID as ID; Define the priority of data sources: order data O′ > user behavior data U′ > logistics data L′, and merge each record according to the priority, which is expressed as follows: D = U′ ∪ O′ ∪ L′ Where, O′ represents order data, U′ represents user behavior data, L′ represents logistics data, and D represents the data matrix; Store the integrated data in a distributed database, which is expressed as follows: S = Cassandra(D) Where, S represents the final integrated data stored in the distributed database, and Cassandra represents the distributed database system for storing data.

5. The e-commerce export goods information search cloud platform based on big data according to claim 1, characterized in that, The full-process traceability and verification process includes the following steps: Generate blocks based on the preprocessed data. Each block contains multiple records, and each record corresponds to a link in the supply chain; Generate a hash value for the generated block data through an encryption algorithm and upload it to the blockchain; Design and deploy a smart contract to automatically execute the business logic in the supply chain. Among them, the business logic is that when a new product is warehoused, the inventory data is automatically updated. At the same time, according to the order information, the inventory is automatically allocated and a transportation plan is generated. According to the real-time logistics information and the transportation plan, the transportation route is dynamically adjusted; Regularly compare the data hash value stored on the blockchain with the current data hash value to verify the data consistency. When the data changes, the data on the blockchain is automatically updated through the smart contract; Users query the supply chain information of specific products through a blockchain browser or API interface and / or the cloud platform automatically queries and analyzes the supply chain data regularly to generate reports and visualization charts.

6. The e-commerce export goods information search cloud platform based on big data according to claim 1, characterized in that, The constructed sentiment analysis model uses a convolutional neural network combined with a bidirectional long short-term memory network BiLSTM to perform sentiment classification training on text data. Among them, the structure of the constructed sentiment analysis model is expressed as follows: Emotion classifier (d i ) = σ(W2 · (BiLSTM(W1 · x i )) + b2) Among them, d i represents the i-th customer text data, x i represents the text vector, W1 and W2 respectively represent the model parameters, σ represents the activation function, and b2 represents the bias vector.

7. A cloud platform for searching e-commerce export goods information based on big data according to claim 1, characterized in that, In the user interface and interaction design system, design the user path so that users can reach the required page with the fewest clicks, and optimize the path length L to minimize the click count C, which is expressed as follows: Among them, C i represents the number of clicks in the i-th path, and n represents the total number of paths; Use a grid layout to design the interface layout, divide the interface into several areas, and each area displays different types of information. The calculation of the number of areas N in the grid layout is as follows: N = Rows × Columns Where, Rows represents the number of rows and Columns represents the number of columns.

8. A cloud platform for searching e-commerce export goods information based on big data according to claim 1, characterized in that, The system integration of the cloud platform uses the HTTP / HTTPS protocol for inter-system communication.

Images