Merchant management method and related equipment

Abstract

The invention provides a merchant management method, and the method comprises the steps: obtaining complaint information of an abnormal complaint work order and obtaining historical multi-dimensional data of a merchant when a target merchant having the abnormal complaint work order is detected, and the multi-dimensional data comprises historical commodity transaction data, historical user evaluation data and historical risk score data; based on the complaint information and the historical multi-dimensional data, determining the complaint category and the complaint confidence of the target merchant; based on the complaint category and the complaint confidence, adjusting the risk score of the target merchant, and when the adjustment of the risk score exceeds a first threshold, generating a risk report of the target merchant; and sending the risk report to the target merchant. According to the method, the risk score of the merchant is evaluated and adjusted by analyzing the abnormal complaint work order and the historical multi-dimensional data of the merchant, so that the risk condition of the merchant can be accurately evaluated, the problem can be found in time, the opportunity of improvement and development is provided for the merchant, and meanwhile, the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a merchant management method and related equipment. Background Art

[0002] With the rapid development of e-commerce and local service platforms, the number of merchants has exploded, and the merchant management challenges faced by the platforms are becoming increasingly severe. Currently, existing merchant management models typically use keyword matching or manual review to screen abnormal complaints, but this is extremely inefficient when faced with massive work orders, and it is difficult to identify semantically complex or hidden malicious complaints. Risk assessments are only based on a single dimension, such as the number of complaints or user ratings, ignoring the correlation of multi-dimensional data; and fixed rules are used to update risk scores, making it difficult to respond to sudden abnormal events in a timely manner. Therefore, existing merchant management methods have problems such as low complaint identification efficiency and high misjudgment rate, risk assessment relying on single-dimensional data, and delayed risk score updates. They cannot accurately assess the risk status of merchants, resulting in a poor customer experience. Summary of the Invention

[0003] Embodiments of the present invention provide a merchant management method designed to address existing merchant management issues, such as reliance on single-dimensional data for risk assessment, delayed risk score updates, and inaccurate assessment of merchants' risk profiles, leading to a poor customer experience. By analyzing abnormal complaint tickets and merchants' historical multi-dimensional data to assess and adjust their risk scores, the present invention accurately assesses merchants' risk profiles, identifies issues promptly, and provides opportunities for improvement and development, while also enhancing the user experience.

[0004] In a first aspect, an embodiment of the present invention provides a merchant management method, the method comprising the following steps:

[0005] When detecting a target merchant with an abnormal complaint ticket, obtaining the complaint information of the abnormal complaint ticket and the merchant's historical multi-dimensional data, the multi-dimensional data including historical commodity transaction data, historical user evaluation data, and historical risk score data;

[0006] Determining the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data;

[0007] Adjusting the risk score of the target merchant based on the complaint category and the complaint confidence, and generating a risk report for the target merchant when the risk score adjustment exceeds a first threshold;

[0008] The risk report is sent to the target merchant.

[0009] Optionally, before obtaining the complaint information of the abnormal complaint work order and obtaining the merchant's historical multi-dimensional data, the method further includes:

[0010] Performing semantic clustering on the received complaint tickets according to the complaint content to obtain multiple ticket clusters, each of which includes at least one complaint ticket;

[0011] Based on the similarity between the cluster center vectors of each of the work order clusters and the preset normal complaint vector, an abnormal work order cluster is determined from the multiple work order clusters, where the normal complaint vector is the cluster center vector corresponding to the normal complaint work order;

[0012] The complaint work orders in the abnormal work order cluster are determined as abnormal complaints, and the merchants of the abnormal complaints are determined as target merchants.

[0013] Optionally, determining an abnormal work order cluster from the plurality of work order clusters based on the similarity between the cluster center vector of each of the work order clusters and a preset normal complaint vector includes:

[0014] Calculating the cosine similarity between the cluster center vector of each work order cluster and the preset normal complaint vector;

[0015] Determine the cluster center vector of the work order cluster whose cosine similarity is less than the similarity threshold corresponding to each complaint vector as an abnormal vector, and different normal complaint vectors correspond to different similarity thresholds;

[0016] The complaint work order in the work order cluster corresponding to the abnormal vector is determined as an abnormal complaint work order.

[0017] Optionally, different normal complaint vectors correspond to different similarity thresholds, including:

[0018] Obtaining historical complaint work orders without abnormalities, performing semantic clustering on the historical complaint work orders according to the complaint content, and obtaining multiple historical work order clusters;

[0019] Determining the cluster center of the historical work order cluster as a normal complaint vector, where each of the historical work order clusters corresponds to one normal complaint vector;

[0020] Based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, a similarity threshold corresponding to each of the normal complaint vectors is determined.

[0021] Optionally, determining a similarity threshold corresponding to each normal complaint vector based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters includes:

[0022] For each of the historical work order clusters, semantic extraction is performed on each of the historical complaint work orders in the historical work order cluster to obtain a complaint vector, where each of the historical complaint work orders corresponds to one complaint vector;

[0023] Calculating the distance between each of the complaint vectors and the normal complaint vector;

[0024] Based on the gradient distribution of the distance values, a similarity threshold corresponding to the normal complaint vector is determined.

[0025] Optionally, determining the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data includes:

[0026] Quantifying the historical commodity transaction data in time series to obtain a commodity transaction sequence, wherein each sequence value in the commodity transaction sequence includes commodity type, commodity price, and transaction time;

[0027] Furthermore, the historical user evaluation data is used to form a user evaluation graph by user, wherein each node in the user evaluation graph includes a user portrait and an evaluation value, and the nodes are connected by the similarity between the user portraits;

[0028] and arranging the historical risk score data in chronological order to obtain a risk score sequence, wherein each sequence value in the risk score sequence includes a risk score value;

[0029] Based on the commodity transaction sequence, the user evaluation map and the risk score sequence, the complaint category and complaint confidence of the target merchant are determined.

[0030] Optionally, determining the complaint category and complaint confidence of the target merchant based on the product transaction sequence, the user evaluation graph, and the risk score sequence includes:

[0031] Constructing a complaint prediction model to be trained and a training data set, wherein the complaint prediction model to be trained includes a first time series network, a graph convolutional network, a second time series network, a feature fusion network, and an output network, and the training data set includes sample data and label data, wherein the sample data includes a sample product transaction sequence, a sample user evaluation map, and a sample risk score value sequence, and the label data includes a complaint category label;

[0032] Input the sample product transaction sequence, the sample user evaluation graph, and the sample risk score value sequence into the first temporal network, the graph convolutional network, and the second temporal network, respectively, to obtain the first eigenvector, the second eigenvector, and the third eigenvector respectively;

[0033] Inputting the first feature vector, the second feature vector, and the third feature vector into the feature fusion network for feature fusion to obtain fused features, and classifying the fused features through the output network to obtain a predicted complaint category;

[0034] The error loss between the predicted complaint category and the complaint category label is calculated by a loss function, and the model parameters of the complaint prediction model to be trained are adjusted with the minimum error loss as the optimization goal. The parameter adjustment process is iterated until the error loss reaches a preset target value or the number of iterations reaches a preset number, then the training is stopped to obtain a trained complaint prediction model;

[0035] The commodity transaction sequence, the user evaluation map and the risk score sequence corresponding to the target merchant are input into the trained complaint prediction model, and the complaint category and complaint confidence of the target merchant are output.

[0036] In a second aspect, an embodiment of the present invention further provides a merchant management device, the merchant management device comprising:

[0037] An acquisition module is used to, when detecting a target merchant with an abnormal complaint work order, obtain the complaint information of the abnormal complaint work order and obtain the merchant's historical multi-dimensional data, the multi-dimensional data including historical product transaction data, historical user evaluation data, and historical risk score data;

[0038] A first determination module is configured to determine the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data;

[0039] a first processing module, configured to adjust the risk score of the target merchant based on the complaint category and the complaint confidence level, and generate a risk report for the target merchant when the risk score adjustment exceeds a first threshold;

[0040] A sending module is used to send the risk report to the target merchant.

[0041] In a third aspect, an embodiment of the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the merchant management method provided in an embodiment of the present invention when executing the computer program.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the merchant management method provided in the embodiment of the invention are implemented.

[0043] In an embodiment of the present invention, when a target merchant with an abnormal complaint work order is detected, the complaint information of the abnormal complaint work order and the merchant's historical multi-dimensional data are obtained, the multi-dimensional data including historical commodity transaction data, historical user evaluation data and historical risk score data; based on the complaint information and historical multi-dimensional data, the complaint category and complaint confidence of the target merchant are determined; based on the complaint category and complaint confidence, the risk score of the target merchant is adjusted, and when the risk score adjustment exceeds a first threshold, a risk report for the target merchant is generated; and the risk report is sent to the target merchant. The present invention evaluates and adjusts the merchant's risk score by analyzing the abnormal complaint work order and the merchant's historical multi-dimensional data. The merchant's risk score can accurately assess the merchant's risk status, discover problems in a timely manner, provide merchants with opportunities for improvement and development, and improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flow chart of a merchant management method provided by an embodiment of the present invention;

[0046] Figure 2 This is a schematic structural diagram of a merchant management device provided by an embodiment of the present invention;

[0047] Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] like Figure 1 As shown, Figure 11 is a flowchart of a merchant management method provided by an embodiment of the present invention, the merchant management method comprising the steps of:

[0050] 101. When a target merchant with an abnormal complaint work order is detected, the complaint information of the abnormal complaint work order and the historical multi-dimensional data of the merchant are obtained.

[0051] In an embodiment of the present invention, the merchant management method may be applied to a server, and the server is communicatively connected to a user terminal. The user terminal includes a terminal interface, and the terminal interface is used to display merchant information of an online shopping mall.

[0052] The aforementioned multi-dimensional data includes historical product transaction data, historical user review data, and historical risk score data. The aforementioned historical transaction data can be understood as the merchant's sales records for products over a period of time, including information such as sales volume, sales amount, sales time, and purchasing users. The aforementioned historical user review data can be understood as user evaluations and feedback on purchased products, including ratings, review content, and review time. The aforementioned historical risk score data can be understood as scores or records that record whether the merchant has engaged in past violations or risk events, including credit scores, quality scores, and safety scores.

[0053] The above-mentioned abnormal complaint tickets can be understood as complaint records in customer feedback that appear abnormal or prominent compared to normal complaint tickets due to product quality issues, service attitude issues, order processing errors, etc.

[0054] Complaint information can be understood as customer feedback and opinions expressed when dissatisfied with a merchant's products or services. Complaint information includes customer information, complaint content, and complaint date. By analyzing complaint information, we can understand customer dissatisfaction with the goods or services, as well as their expectations and suggestions for the merchant. Furthermore, based on the content and frequency of complaint information, we can assess the merchant's product quality and service level, enabling timely improvements and enhancements.

[0055] It should be noted that the complaint information of abnormal complaint tickets can reflect customer dissatisfaction, sales product quality, and service problems.

[0056] 102. Based on complaint information and historical multi-dimensional data, determine the complaint category and complaint confidence of the target merchant.

[0057] In an embodiment of the present invention, the above-mentioned complaint categories are complaint categories obtained by classifying complaints according to complaint information, such as product problems, service problems, etc.

[0058] The above complaint confidence level is used to assess and quantify the credibility, severity or importance of a complaint. The above complaint confidence level reflects the severity of the complaint.

[0059] Specifically, the complaint category and corresponding complaint confidence of the target merchant can be determined through the collected complaint information, historical product transaction data, historical user evaluation data, and historical risk score data.

[0060] It should be noted that for each complaint category, the system also assesses the confidence level of each complaint category. If a product category has a large number of complaints, a high frequency of complaints, and a significant impact on consumers, then the confidence level of this product category will be higher. If a product category has a small number of complaints or is not serious, then the confidence level of this product category will be lower.

[0061] 103. Adjust the risk score of the target merchant based on the complaint category and the complaint confidence level, and generate a risk report for the target merchant when the risk score adjustment exceeds a first threshold.

[0062] In this embodiment of the present invention, the target merchant's risk score is a comprehensive evaluation of the merchant. The risk score can help the system determine the merchant's risk level during its operations. For example, a low risk score indicates that the merchant is operating at a low risk; a high risk score indicates that the merchant is at a high risk, which can improve monitoring and management of the merchant.

[0063] The first threshold mentioned above is a risk score threshold preset by the system, which is used to determine whether the magnitude of the merchant risk score adjustment has reached a level that requires attention.

[0064] The above-mentioned risk report can be understood as a document that analyzes risks and records risk status, including risk identification, risk analysis, and response measures.

[0065] In one possible implementation, a merchant's risk score is adjusted based on the types of complaints received and the confidence level of the complaints. When the merchant's risk score adjustment exceeds a pre-set threshold, the system deems the merchant's risk level to have reached a level requiring attention and generates a risk report for the merchant. This report may include information such as the merchant's current complaints, changes in risk score, and recommended improvement measures.

[0066] 104. Send the risk report to the target merchant.

[0067] In an embodiment of the present invention, when the risk score of the target merchant exceeds a set first threshold, the system generates a risk report for the target merchant and sends the risk report to the target merchant.

[0068] In an embodiment of the present invention, when a target merchant with an abnormal complaint work order is detected, complaint information of the abnormal complaint work order and historical multi-dimensional data of the merchant are obtained, the multi-dimensional data including historical commodity transaction data, historical user evaluation data, and historical risk score data; based on the complaint information and historical multi-dimensional data, the complaint category and complaint confidence of the target merchant are determined; based on the complaint category and complaint confidence, the risk score of the target merchant is adjusted, and when the risk score adjustment exceeds a first threshold, a risk report of the target merchant is generated; and the risk report is sent to the target merchant. The present invention evaluates and adjusts the risk score of the merchant by analyzing the abnormal complaint work order and the merchant's historical multi-dimensional data. It can accurately assess the merchant's risk status, identify problems in a timely manner, provide opportunities for improvement and development for the merchant, and improve the user experience.

[0069] It is understandable that in the specific implementation of this application, related data such as product information, evaluation information, user information, purchase information, risk score information, etc. are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data, as well as the training, deployment and calling of algorithm models, must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0070] Optionally, before obtaining the complaint information of the abnormal complaint work order and obtaining the historical multi-dimensional data of the merchant, the received complaint work order can also be semantically clustered according to the complaint content to obtain multiple work order clustering clusters; based on the similarity between the cluster center vector of each work order clustering cluster and the preset normal complaint vector, an abnormal work order clustering cluster is determined among the multiple work order clustering clusters, and the normal complaint vector is the cluster center vector corresponding to the normal complaint work order; the complaint work order in the abnormal work order clustering cluster is determined as an abnormal complaint, and the merchant of the abnormal complaint is determined as the target merchant.

[0071] In an embodiment of the present invention, the work order cluster includes at least one complaint work order. The work order cluster can be understood as a grouping of work orders with similar features or contents, and the texts in each cluster have semantic similarity.

[0072] The above-mentioned complaint ticket can be understood as a record of customers making complaints through telephone, email, etc. due to dissatisfaction with services or product issues.

[0073] The above complaints include users' dissatisfaction with the product, service attitude, service quality, etc., and detailed descriptions of the consumption process and reasons for the complaint.

[0074] The semantic clustering process described above can be understood as the process of analyzing the semantic information in the text data of received complaint tickets based on the complaint content, and classifying text data with similar semantics into the same group. For example, suppose a batch of complaint tickets are received, and the complaint tickets include customers' evaluations of the quality of the purchased goods, delivery speed, service attitude, and other aspects. First, after semantic clustering is performed on these tickets, multiple ticket clusters are obtained, including "product quality issues," "slow delivery speed," and "poor service attitude," etc. Natural language processing technology can be used to semantically cluster complaint tickets based on the complaint content. The above natural language processing technology aims to enable machines to understand, interpret, and generate human language, achieve effective communication between humans and machines, and enable computers to perform tasks such as language translation, sentiment analysis, and text summarization.

[0075] The cluster center vector of the above work order cluster can be understood as a vector representing the characteristics of the work order cluster.

[0076] The similarity between the cluster center vectors of each of the aforementioned work order clusters and the preset normal complaint vector is predefined and reflects the proximity of the cluster center vectors of the work order clusters to the preset normal complaint vector in feature space. Specifically, abnormal work order clusters can be identified from multiple work order clusters based on the similarity between the cluster center vectors of each work order cluster and the preset normal complaint vector.

[0077] The above abnormal work order clusters can be understood as cluster groups of abnormal or erroneous work orders, and the abnormal work order clusters exhibit characteristics different from normal work orders.

[0078] The above-mentioned abnormal complaints can be understood as complaints that do not conform to routine and standards, including abnormal complaint content, unreasonable complaints, frequent complaints, repeated complaints, etc.

[0079] In one possible embodiment, for example, assuming there are 100 complaint tickets, they are divided into five clusters through semantic analysis: Cluster A, Cluster B, Cluster C, Cluster D, and Cluster E. Based on the similarity between the cluster center vector of each ticket cluster and the preset normal complaint vector, if the cluster center vector of Cluster D or Cluster E in Cluster A, Cluster B, Cluster C, Cluster D, and Cluster E is lower than the preset normal complaint vector, then Cluster D and Cluster E are identified as abnormal ticket clusters. The complaint tickets in the abnormal ticket clusters are then determined to be abnormal complaints, and the merchants corresponding to Cluster D and Cluster E are determined to be target merchants.

[0080] Optionally, in the step of determining an abnormal work order clustering cluster among multiple work order clustering clusters based on the similarity between the cluster center vectors of each work order clustering cluster and the preset normal complaint vector, the cosine similarity between the cluster center vectors of each work order clustering cluster and the preset normal complaint vector can be calculated; the cluster center vectors of the work order clustering clusters whose cosine similarity is less than the similarity threshold corresponding to each complaint vector are determined as abnormal vectors; and the complaint work order in the work order clustering cluster corresponding to the abnormal vector is determined as an abnormal complaint work order.

[0081] In the embodiment of the present invention, different normal complaint vectors correspond to different similarity thresholds.

[0082] The above cosine similarity measures the cosine value of the angle between two vectors and evaluates the similarity between the two vectors.

[0083] In one possible embodiment, abnormal work order clusters can be identified by calculating the cosine similarity between the cluster center vector of the work order cluster and a preset normal complaint vector. If the cosine similarity between the cluster center vector of work order cluster B and the preset normal complaint vector is less than a preset similarity threshold, then work order cluster B is determined to be an abnormal vector.

[0084] In another possible embodiment, for example, suppose that when analyzing customer service requests, it is discovered that clusters B, C, and D all inquire about product usage issues, but cluster A primarily contains complaints about payment errors. The cluster center vector of cluster A differs significantly in content or emphasis from the vector containing regular product usage inquiries. After calculating the cosine similarity between the two, if the cosine similarity falls below a preset similarity threshold, cluster A is determined to be an anomalous vector, and the complaint ticket corresponding to cluster A is determined to be an anomalous complaint ticket.

[0085] Optionally, in the step where different normal complaint vectors correspond to different similarity thresholds, historical complaint tickets without abnormalities can be obtained, and semantic clustering processing can be performed on the historical complaint tickets according to the complaint content to obtain multiple historical ticket clusters; the cluster center of the historical ticket cluster is determined as the normal complaint vector, and each historical ticket cluster corresponds to a normal complaint vector; based on the density distribution between each historical complaint ticket and the normal complaint vector in each historical ticket cluster, the similarity threshold corresponding to each normal complaint vector is determined.

[0086] In an embodiment of the present invention, the above-mentioned historical complaint work orders without abnormalities can be understood as complaint records that do not contain any abnormal or erroneous situations in the historical complaint work orders.

[0087] The above-mentioned complaints include a detailed description of the consumption process and reasons for the complaint due to customers' dissatisfaction with service quality, product quality, after-sales service, etc. during the consumption process.

[0088] The semantic clustering process described above can be understood as analyzing the semantic information within the text data within historical complaint tickets and grouping text data with similar semantics into the same group. For example, if the complaints in two tickets are semantically similar or related, they will be classified into the same cluster. Natural language processing technology can be used to perform semantic clustering on historical complaint tickets based on their content. This natural language processing technology aims to enable machines to understand, interpret, and generate human language, enabling effective communication between humans and machines and enabling computers to perform tasks such as language translation, sentiment analysis, and text summarization.

[0089] The above-mentioned work order clustering can be understood as grouping work orders with similar features or contents together, and the texts in each cluster are semantically similar.

[0090] The cluster center of the above historical work order cluster can be understood as the representative point of all work orders in the cluster in the semantic space. The cluster center reflects the main characteristics and trends of the cluster.

[0091] Furthermore, for each historical work order cluster, the cluster center point of the cluster can be determined as a normal complaint vector.

[0092] The density distribution between each historical complaint work order and the normal complaint vector in each of the above historical work order clusters can be understood as the relative density of the historical complaint work orders and the normal complaint vector in the cluster.

[0093] The similarity threshold corresponding to the normal complaint vector is used to distinguish normal complaint vectors from abnormal complaint vectors. For example, if the similarity between a complaint ticket 1 and a normal complaint vector exceeds the similarity threshold corresponding to the normal complaint vector, the complaint ticket 1 can be judged as a normal complaint ticket. If the similarity between the complaint ticket 1 and the normal complaint vector is lower than the similarity threshold corresponding to the normal complaint vector, the complaint ticket 1 can be judged as an abnormal complaint ticket.

[0094] Optionally, in the step of determining the similarity threshold corresponding to each normal complaint vector based on the density distribution between each historical complaint work order and the normal complaint vector in each historical work order clustering cluster, semantic extraction can be performed on each historical complaint work order in the historical work order clustering cluster to obtain a complaint vector, with each historical complaint work order corresponding to a complaint vector; the distance value between each complaint vector and the normal complaint vector is calculated; and the similarity threshold corresponding to the normal complaint vector is determined based on the gradient distribution of the distance value.

[0095] In the embodiment of the present invention, the semantic extraction can be understood as a process of extracting and understanding semantic information from each historical complaint ticket. Natural language processing technology can be used to vectorize each historical complaint ticket.

[0096] The complaint vector can be understood as a vector formed by extracting features from a complaint ticket. The complaint vector can be used to represent the overall status of a complaint ticket. The complaint vector includes features such as the content, cause, and resolution of the complaint.

[0097] The distance value between each of the above complaint vectors and the normal complaint vector can be calculated using Euclidean distance, cosine similarity, etc. The distance value can be used to evaluate the similarity between the complaint vector and the normal complaint vector.

[0098] The gradient distribution can be understood as the rate or direction at which the distance between each complaint vector and a normal complaint vector changes with position in space. Gradient distribution is often used to characterize the spatial variation of the distance between each complaint vector and a normal complaint vector.

[0099] The similarity threshold corresponding to the above normal complaint vector is used to distinguish normal complaint vectors from abnormal complaint vectors.

[0100] Optionally, in the step of determining the complaint category and complaint confidence of the target merchant based on complaint information and historical multi-dimensional data, the historical commodity transaction data can be quantified in time series to obtain a commodity transaction sequence; and the historical user evaluation data can be formed into a user evaluation map by user; and the historical risk score data can be arranged in time series to obtain a risk score sequence; based on the commodity transaction sequence, the user evaluation map and the risk score sequence, the complaint category and complaint confidence of the target merchant are determined.

[0101] In the embodiment of the present invention, each sequence value in the commodity transaction sequence includes commodity type, commodity price and transaction time.

[0102] The above-mentioned quantitative processing can be understood as converting historical commodity transaction data into a calculable and analyzable form in chronological order.

[0103] The above-mentioned chronological order can be understood as according to chronological order.

[0104] The above commodity transaction sequence is the commodity sales history data arranged in chronological order.

[0105] Each node in the above user evaluation graph includes a user profile and an evaluation value, and the nodes are connected by the similarity between the user profiles.

[0106] Each sequence value in the above risk score sequence includes a risk score value.

[0107] The above complaint categories are obtained by classifying according to the product transaction sequence, user evaluation map and risk score sequence, such as product problems, service problems, etc.

[0108] The above complaint confidence level is used to assess and quantify the credibility, severity or importance of a complaint. The above complaint confidence level reflects the severity of the complaint.

[0109] Optionally, in the step of determining the complaint category and complaint confidence of the target merchant based on the product transaction sequence, user evaluation map and risk score sequence, a complaint prediction model to be trained and a training data set can be constructed; the sample product transaction sequence, the sample user evaluation map and the sample risk score value sequence are respectively input into the first time series network, the graph convolution network and the second time series network to obtain the first eigenvector, the second eigenvector and the third eigenvector respectively; the first eigenvector, the second eigenvector and the third eigenvector are input into the feature fusion network for feature fusion to obtain fusion features, and the fusion features are classified and processed through the output network to obtain the predicted complaint category; the error loss between the predicted complaint category and the complaint category label is calculated by the loss function, and the model parameters of the complaint prediction model to be trained are adjusted with the minimum error loss as the optimization goal, and the parameter adjustment process is iterated until the error loss reaches the preset target value or the number of iterations reaches the preset number, then the training is stopped to obtain the trained complaint prediction model; the product transaction sequence, user evaluation map and risk score sequence corresponding to the target merchant are input into the trained complaint prediction model, and the complaint category and complaint confidence of the target merchant are output.

[0110] In an embodiment of the present invention, the complaint prediction model to be trained may be a complaint prediction model based on deep learning or machine learning, such as a convolutional neural network (CNN), a recurrent neural network (RNN), etc. The complaint prediction model to be trained includes a first time series network, a graph convolutional network, a second time series network, a feature fusion network, and an output network.

[0111] The training dataset includes sample data and labeled data. The sample data includes sample product transaction sequences, sample user review graphs, and sample risk score sequences. The sample product transaction sequences contain the sales history of the products in chronological order. Each node in the sample user review graph includes a user profile and a review value, with nodes connected by the similarity between user profiles. Each sequence value in the sample risk score sequence includes a risk score value. The labeled data includes complaint category labels. These complaint category labels can be understood as categorized labels for product or service complaints.

[0112] The aforementioned time series network is used to process time series data and can capture temporal dependencies in the data. The aforementioned graph convolutional network is used to combine the graph's topological structure with node features, extracting relationships and feature information between nodes through graph convolution operations. The aforementioned feature fusion network aims to integrate feature information from different levels and sources. By effectively fusing multiple feature representations, the feature fusion network can help the complaint prediction model better capture the complex relationships in the data. The aforementioned output network predicts the complaint category based on the input features.

[0113] The above-mentioned feature fusion can be understood as the process of fusing the first, second, and third eigenvectors. This feature fusion aims to effectively combine feature information from different levels, sources, or modalities, thereby improving the model's performance on complex visual tasks.

[0114] The above classification process can be understood as the process of mapping the fused features to predefined category labels through the output network.

[0115] The above-mentioned predicted complaint categories can be understood as classifying complaints into different types according to the content of the complaints, such as product quality, service attitude, etc.

[0116] The above loss function is used to measure the difference between the model's prediction and the actual result. By minimizing the loss function, the model parameters are optimized to make the predicted value as close to the actual value as possible. The above loss function can be a mean squared error loss function, a cross entropy loss function, etc.

[0117] The above adjustments can be understood as the process of adjusting the parameters in the model and optimizing the model performance.

[0118] It's important to note that model parameters are optimized by minimizing the loss function, making the model's predictions closer to reality. During training, model parameters are continuously adjusted to reduce prediction error. This parameter adjustment process is repeated until the error is reduced to a preset target value or the preset number of iterations is reached. Training is then terminated, resulting in a trained complaint prediction model.

[0119] The above-trained complaint prediction model can predict the complaint category and complaint confidence of the target merchant based on the input target merchant's corresponding product transaction sequence, user evaluation map and risk score sequence.

[0120] like Figure 2 As shown, an embodiment of the present invention provides a merchant management device, which includes:

[0121] Acquisition module 201 is used to, when detecting a target merchant with an abnormal complaint work order, acquire the complaint information of the abnormal complaint work order and acquire the merchant's historical multi-dimensional data, the multi-dimensional data including historical product transaction data, historical user evaluation data, and historical risk score data;

[0122] A first determination module 202 is configured to determine the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data;

[0123] A first processing module 203 is configured to adjust the risk score of the target merchant based on the complaint category and the complaint confidence level, and generate a risk report for the target merchant when the risk score adjustment exceeds a first threshold;

[0124] The sending module 204 is configured to send the risk report to the target merchant.

[0125] Optionally, the device further includes:

[0126] A second processing module is configured to perform semantic clustering on the received complaint tickets according to the complaint content to obtain a plurality of ticket clusters, each of which includes at least one complaint ticket;

[0127] A third processing module is configured to determine an abnormal work order cluster from the plurality of work order clusters based on the similarity between the cluster center vector of each of the work order clusters and a preset normal complaint vector, wherein the normal complaint vector is the cluster center vector corresponding to the normal complaint work order;

[0128] The second determining module is used to determine the complaint work order in the abnormal work order cluster as an abnormal complaint, and determine the merchant of the abnormal complaint as the target merchant.

[0129] Optionally, the third processing module is also used to calculate the cosine similarity between the cluster center vector of each work order cluster and the preset normal complaint vector; the cluster center vector of the work order cluster whose cosine similarity is less than the similarity threshold corresponding to each complaint vector is determined as an abnormal vector, and different normal complaint vectors correspond to different similarity thresholds; the complaint work order in the work order cluster corresponding to the abnormal vector is determined as an abnormal complaint work order.

[0130] Optionally, the third processing module is also used to obtain historical complaint work orders without abnormalities, perform semantic clustering on the historical complaint work orders according to the complaint content, and obtain multiple historical work order clusters; determine the cluster center of the historical work order cluster as the normal complaint vector, and each historical work order cluster corresponds to one normal complaint vector; based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, determine the similarity threshold corresponding to each normal complaint vector.

[0131] Optionally, the third processing module is also used to perform semantic extraction on each of the historical complaint work orders in the historical work order cluster cluster to obtain a complaint vector, where each historical complaint work order corresponds to one complaint vector; calculate the distance value between each complaint vector and the normal complaint vector; and determine the similarity threshold corresponding to the normal complaint vector based on the gradient distribution of the distance value.

[0132] Optionally, the first determination module 202 is also used to quantify the historical commodity transaction data in time series to obtain a commodity transaction sequence, wherein each sequence value in the commodity transaction sequence includes the commodity type, commodity price and transaction time; and, to form a user evaluation graph according to the user, wherein each node in the user evaluation graph includes a user portrait and an evaluation value, and the nodes are connected through the similarity between the user portraits; and, to arrange the historical risk score data in time series to obtain a risk score sequence, wherein each sequence value in the risk score sequence includes a risk score value; based on the commodity transaction sequence, the user evaluation graph and the risk score sequence, the complaint category and complaint confidence of the target merchant are determined.

[0133] Optionally, the first determination module 202 is also used to construct a complaint prediction model to be trained and a training data set, wherein the complaint prediction model to be trained includes a first time series network, a graph convolutional network, a second time series network, a feature fusion network and an output network, and the training data set includes sample data and label data, and the sample data includes a sample product transaction sequence, a sample user evaluation map, and a sample risk score value sequence, and the label data includes a complaint category label; the sample product transaction sequence, the sample user evaluation map and the sample risk score value sequence are respectively input into the first time series network, the graph convolutional network and the second time series network, and the first eigenvector, the second eigenvector and the third eigenvector are correspondingly obtained; the first eigenvector, the second eigenvector and the third eigenvector ... The third feature vector is input into the feature fusion network for feature fusion to obtain a fused feature, and the fused feature is classified and processed by the output network to obtain a predicted complaint category; the error loss between the predicted complaint category and the complaint category label is calculated by the loss function, and the model parameters of the complaint prediction model to be trained are adjusted with the minimum error loss as the optimization goal, and the parameter adjustment process is iterated until the error loss reaches a preset target value or the number of iterations reaches a preset number, then the training is stopped to obtain a trained complaint prediction model; the product transaction sequence, the user evaluation map and the risk score sequence corresponding to the target merchant are input into the trained complaint prediction model, and the complaint category and complaint confidence of the target merchant are output.

[0134] like Figure 3 As shown, an embodiment of the present invention further provides an electronic device, including a processor, and the processor can execute any one of the above merchant management methods.

[0135] Specifically, the system includes a processor 301, a memory 302, and a computer program for executing the merchant management method, which is stored in the memory 302 and can be run on the processor 301, wherein:

[0136] The processor 301 runs the computer program of the merchant management method stored in the memory 302 and performs the following steps:

[0137] When detecting a target merchant with an abnormal complaint ticket, obtaining the complaint information of the abnormal complaint ticket and the merchant's historical multi-dimensional data, the multi-dimensional data including historical commodity transaction data, historical user evaluation data, and historical risk score data;

[0138] Determining the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data;

[0139] Adjusting the risk score of the target merchant based on the complaint category and the complaint confidence, and generating a risk report for the target merchant when the adjustment of the risk score exceeds a first threshold;

[0140] The risk report is sent to the target merchant.

[0141] Optionally, before obtaining the complaint information of the abnormal complaint work order and obtaining the historical multi-dimensional data of the merchant, the method executed by the processor 301 further includes:

[0142] Performing semantic clustering on the received complaint tickets according to the complaint content to obtain multiple ticket clusters, each of which includes at least one complaint ticket;

[0143] Based on the similarity between the cluster center vectors of each of the work order clusters and the preset normal complaint vector, an abnormal work order cluster is determined from the multiple work order clusters, where the normal complaint vector is the cluster center vector corresponding to the normal complaint work order;

[0144] The complaint work orders in the abnormal work order cluster are determined as abnormal complaints, and the merchants of the abnormal complaints are determined as target merchants.

[0145] Optionally, the processor 301 determines an abnormal work order cluster from the plurality of work order clusters based on the similarity between the cluster center vector of each work order cluster and a preset normal complaint vector, including:

[0146] Calculating the cosine similarity between the cluster center vector of each work order cluster and the preset normal complaint vector;

[0147] Determine the cluster center vector of the work order cluster whose cosine similarity is less than the similarity threshold corresponding to each complaint vector as an abnormal vector, and different normal complaint vectors correspond to different similarity thresholds;

[0148] The complaint work order in the work order cluster corresponding to the abnormal vector is determined as an abnormal complaint work order.

[0149] Optionally, the different normal complaint vectors executed by the processor 301 correspond to different similarity thresholds, including:

[0150] Obtaining historical complaint work orders without abnormalities, performing semantic clustering on the historical complaint work orders according to the complaint content, and obtaining multiple historical work order clusters;

[0151] Determining the cluster center of the historical work order cluster as a normal complaint vector, where each of the historical work order clusters corresponds to one normal complaint vector;

[0152] Based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, a similarity threshold corresponding to each of the normal complaint vectors is determined.

[0153] Optionally, the processor 301 determines the similarity threshold corresponding to each normal complaint vector based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, including:

[0154] For each of the historical work order clusters, semantic extraction is performed on each of the historical complaint work orders in the historical work order cluster to obtain a complaint vector, where each of the historical complaint work orders corresponds to one complaint vector;

[0155] Calculating the distance between each of the complaint vectors and the normal complaint vector;

[0156] Based on the gradient distribution of the distance values, a similarity threshold corresponding to the normal complaint vector is determined.

[0157] Optionally, the processor 301 determines the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data, including:

[0158] Quantifying the historical commodity transaction data in time series to obtain a commodity transaction sequence, wherein each sequence value in the commodity transaction sequence includes commodity type, commodity price, and transaction time;

[0159] Furthermore, the historical user evaluation data is used to form a user evaluation graph by user, wherein each node in the user evaluation graph includes a user portrait and an evaluation value, and the nodes are connected by the similarity between the user portraits;

[0160] and arranging the historical risk score data in chronological order to obtain a risk score sequence, wherein each sequence value in the risk score sequence includes a risk score value;

[0161] Based on the commodity transaction sequence, the user evaluation map and the risk score sequence, the complaint category and complaint confidence of the target merchant are determined.

[0162] Optionally, the processor 301 determines the complaint category and complaint confidence of the target merchant based on the product transaction sequence, the user evaluation graph, and the risk score sequence, including:

[0163] Constructing a complaint prediction model to be trained and a training data set, wherein the complaint prediction model to be trained includes a first time series network, a graph convolutional network, a second time series network, a feature fusion network, and an output network, and the training data set includes sample data and label data, wherein the sample data includes a sample product transaction sequence, a sample user evaluation map, and a sample risk score value sequence, and the label data includes a complaint category label;

[0164] Input the sample product transaction sequence, the sample user evaluation graph, and the sample risk score value sequence into the first temporal network, the graph convolutional network, and the second temporal network, respectively, to obtain the first eigenvector, the second eigenvector, and the third eigenvector respectively;

[0165] Inputting the first feature vector, the second feature vector, and the third feature vector into the feature fusion network for feature fusion to obtain fused features, and classifying the fused features through the output network to obtain a predicted complaint category;

[0166] The error loss between the predicted complaint category and the complaint category label is calculated by a loss function, and the model parameters of the complaint prediction model to be trained are adjusted with the minimum error loss as the optimization goal. The parameter adjustment process is iterated until the error loss reaches a preset target value or the number of iterations reaches a preset number, then the training is stopped to obtain a trained complaint prediction model;

[0167] The commodity transaction sequence, the user evaluation map and the risk score sequence corresponding to the target merchant are input into the trained complaint prediction model, and the complaint category and complaint confidence of the target merchant are output.

[0168] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the merchant management method provided by the embodiment of the present invention are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0169] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0170] The above disclosure is merely a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A merchant management method, characterized in that: The method comprises the following steps: When detecting a target merchant with an abnormal complaint ticket, obtaining the complaint information of the abnormal complaint ticket and the merchant's historical multi-dimensional data, the multi-dimensional data including historical commodity transaction data, historical user evaluation data, and historical risk score data; Determining the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data; Adjusting the risk score of the target merchant based on the complaint category and the complaint confidence, and generating a risk report for the target merchant when the adjustment of the risk score exceeds a first threshold; The risk report is sent to the target merchant.

2. The merchant management method according to claim 1, wherein: Before obtaining the complaint information of the abnormal complaint work order and obtaining the historical multi-dimensional data of the merchant, the method further includes: Performing semantic clustering on the received complaint tickets according to the complaint content to obtain multiple ticket clusters, each of which includes at least one complaint ticket; Based on the similarity between the cluster center vectors of each of the work order clusters and the preset normal complaint vector, an abnormal work order cluster is determined from the multiple work order clusters, where the normal complaint vector is the cluster center vector corresponding to the normal complaint work order; The complaint work orders in the abnormal work order cluster are determined as abnormal complaints, and the merchants of the abnormal complaints are determined as target merchants.

3. The merchant management method according to claim 2, wherein: The determining of an abnormal work order cluster from the plurality of work order clusters based on the similarity between the cluster center vector of each of the work order clusters and a preset normal complaint vector includes: Calculating the cosine similarity between the cluster center vector of each work order cluster and the preset normal complaint vector; Determine the cluster center vector of the work order cluster whose cosine similarity is less than the similarity threshold corresponding to each complaint vector as an abnormal vector, and different normal complaint vectors correspond to different similarity thresholds; The complaint work order in the work order cluster corresponding to the abnormal vector is determined as an abnormal complaint work order.

4. The merchant management method according to claim 3, wherein: Different normal complaint vectors have different similarity thresholds, including: Obtaining historical complaint work orders without abnormalities, performing semantic clustering on the historical complaint work orders according to the complaint content, and obtaining multiple historical work order clusters; Determining the cluster center of the historical work order cluster as a normal complaint vector, where each of the historical work order clusters corresponds to one normal complaint vector; Based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, a similarity threshold corresponding to each of the normal complaint vectors is determined.

5. The merchant management method according to claim 4, wherein: The determining, based on the density distribution between each of the historical complaint work orders and the normal complaint vector in each of the historical work order clusters, a similarity threshold corresponding to each of the normal complaint vectors includes: For each of the historical work order clusters, semantic extraction is performed on each of the historical complaint work orders in the historical work order cluster to obtain a complaint vector, where each of the historical complaint work orders corresponds to one complaint vector; Calculating the distance between each of the complaint vectors and the normal complaint vector; Based on the gradient distribution of the distance values, a similarity threshold corresponding to the normal complaint vector is determined.

6. The merchant management method according to any one of claims 1 to 5, characterized in that: Determining the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data includes: Quantifying the historical commodity transaction data in time series to obtain a commodity transaction sequence, wherein each sequence value in the commodity transaction sequence includes commodity type, commodity price, and transaction time; Furthermore, the historical user evaluation data is used to form a user evaluation graph by user, wherein each node in the user evaluation graph includes a user portrait and an evaluation value, and the nodes are connected by the similarity between the user portraits; and arranging the historical risk score data in chronological order to obtain a risk score sequence, wherein each sequence value in the risk score sequence includes a risk score value; Based on the commodity transaction sequence, the user evaluation map and the risk score sequence, the complaint category and complaint confidence of the target merchant are determined.

7. The merchant management method according to claim 6, wherein: The determining of the complaint category and complaint confidence of the target merchant based on the product transaction sequence, the user evaluation graph, and the risk score sequence includes: Constructing a complaint prediction model to be trained and a training data set, wherein the complaint prediction model to be trained includes a first time series network, a graph convolutional network, a second time series network, a feature fusion network, and an output network, and the training data set includes sample data and label data, wherein the sample data includes a sample product transaction sequence, a sample user evaluation map, and a sample risk score value sequence, and the label data includes a complaint category label; Input the sample product transaction sequence, the sample user evaluation graph, and the sample risk score value sequence into the first temporal network, the graph convolutional network, and the second temporal network, respectively, to obtain the first eigenvector, the second eigenvector, and the third eigenvector respectively; Inputting the first feature vector, the second feature vector, and the third feature vector into the feature fusion network for feature fusion to obtain fused features, and classifying the fused features through the output network to obtain a predicted complaint category; The error loss between the predicted complaint category and the complaint category label is calculated by a loss function, and the model parameters of the complaint prediction model to be trained are adjusted with the minimum error loss as the optimization goal. The parameter adjustment process is iterated until the error loss reaches a preset target value or the number of iterations reaches a preset number, then the training is stopped to obtain a trained complaint prediction model; The commodity transaction sequence, the user evaluation map and the risk score sequence corresponding to the target merchant are input into the trained complaint prediction model, and the complaint category and complaint confidence of the target merchant are output.

8. A merchant management device, characterized in that: The merchant management device includes: An acquisition module is used to, when detecting a target merchant with an abnormal complaint work order, obtain the complaint information of the abnormal complaint work order and obtain the merchant's historical multi-dimensional data, the multi-dimensional data including historical product transaction data, historical user evaluation data, and historical risk score data; A first determination module determines the complaint category and complaint confidence of the target merchant based on the complaint information and the historical multi-dimensional data; a first processing module, configured to adjust the risk score of the target merchant based on the complaint category and the complaint confidence level, and generate a risk report for the target merchant when the risk score adjustment exceeds a first threshold; A sending module is used to send the risk report to the target merchant.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the merchant management method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the merchant management method according to any one of claims 1 to 7.

Images