A predictive complaint handling decision support method and system based on machine learning

Abstract

The present invention discloses a predictive complaint handling decision support method and system based on machine learning, which relates to the field of complaint data processing technology, including collecting complaint management system data, preprocessing the data, determining the main entities and relationships in the knowledge graph to construct a knowledge graph; extracting node and edge features from the knowledge graph, extracting keywords and sentiment tendencies in text data; predicting complaint handling demand scores based on machine learning algorithms, and generating priority rankings and processing suggestions for each complaint. The method of the present invention structures and visualizes complex complaint data through the construction of a knowledge graph, improves the efficiency and accuracy of complaint handling, improves the accuracy of complaint handling and customer satisfaction through deep feature extraction and sentiment analysis, and improves the efficiency and response speed of complaint handling through the application of machine learning algorithms, reduces manual intervention, and reduces processing costs.

Description

Technical Field

[0001] The present invention relates to the technical field of complaint data processing, and specifically to a predictive complaint processing decision support method and system based on machine learning. Background Art

[0002] In modern society, customer complaint management systems play a vital role in corporate services. With the rapid development of big data and artificial intelligence technologies, companies have gradually realized the importance of efficiently handling customer complaints and have begun to use advanced technologies to improve customer satisfaction and operational efficiency. Traditional complaint handling methods mainly rely on manual processing, which is not only inefficient but also prone to human errors. In order to meet these challenges, more and more companies have begun to introduce machine learning and knowledge graph technologies to optimize the complaint handling process through automation and intelligent means. These technologies can quickly analyze large amounts of complaint data, identify key issues and trends, and provide companies with valuable decision-making support, thereby improving the response speed and resolution quality of customer complaints.

[0003] Although machine learning and knowledge graph technologies have shown great potential in customer complaint data management, existing related technologies still have many shortcomings in practical applications. Most current complaint data processing systems still rely on manual operations in data preprocessing and feature extraction, and lack effective automated solutions, resulting in inefficient and error-prone data processing. When constructing and utilizing knowledge graphs, existing technologies are often unable to accurately extract and associate important entities and relationships in complaint data, making it difficult to give full play to the role of knowledge graphs in decision support. In addition, existing machine learning models usually only make simple predictions based on historical data, and fail to combine knowledge graphs for in-depth analysis, thus failing to provide accurate complaint processing demand scores and reasonable priority sorting. These technical deficiencies make it difficult for enterprises to achieve efficient and accurate processing when faced with a large amount of complaint data. In contrast, our invention can effectively overcome the above problems and improve the efficiency and accuracy of complaint data processing through automated data preprocessing, knowledge graph construction and feature extraction, as well as complaint demand score prediction based on deep learning. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing complaint data processing methods have the problems of low data preprocessing efficiency, inaccurate knowledge graph construction, inaccurate complaint demand scoring, and how to achieve efficient and accurate complaint processing.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a predictive complaint handling decision support method based on machine learning, comprising collecting complaint management system data, preprocessing the data, determining the main entities and relationships in the knowledge graph to construct a knowledge graph; extracting node and edge features from the knowledge graph, extracting keywords and sentiment tendencies in text data; predicting complaint handling demand scores based on machine learning algorithms, and generating priority rankings and handling suggestions for each complaint.

[0007] As a preferred scheme of the predictive complaint handling decision support method based on machine learning described in the present invention, wherein: the preprocessing includes collecting complaint management system data, the complaint management system data includes customer complaint content, complaint time, customer information, product information, complaint handling history data, and performing data cleaning, data integration, and text preprocessing on the complaint management system data; data cleaning includes processing missing values, outliers, unifying data formats, and performing data standardization; data integration includes integrating data from different data sources to generate a unified data set; text preprocessing performs word segmentation, stop word removal, and stem extraction on the complaint text data.

[0008] As a preferred solution of the predictive complaint handling decision support method based on machine learning described in the present invention, wherein: the construction of the knowledge graph includes determining the main entities and relationships in the knowledge graph, the main entities include customers, products, complaints, solutions, and the relationships include customers and complaints, complaints and products, complaints and solutions. Each entity represents a node, and the edge represents the relationship between entities. Different categories of relationship information are integrated to construct a knowledge graph, which is expressed as:

[0009] ,

[0010] in, Nodes in the knowledge graph With Node The weighted adjacency matrix elements between is the category index of the relation, is the total number of relationship categories, For the The weight of the class relationship, Indicates Node under class relationship With Node The connection status, Representation Node With Node In the Connected under class relationship, Representation Node With Node In the Not connected under class relationship; calculate the Weight of class relationship It is expressed as:

[0011] ,

[0012] in, Indicates The frequency of class relations, Indicates The importance of class relationships, Indicates The frequency of class relations, Indicates The importance of class relationships, Indicates the difference from The index of the relationship category.

[0013] As a preferred solution of the predictive complaint handling decision support method based on machine learning described in the present invention, the extracting node and edge features includes extracting and adjusting the node and edge features in the complaint management system data by multi-level propagation and combination of node features and combining centrality indicators, which is expressed as:

[0014] ,

[0015] in, For the The node feature matrix of the layer, is the ReLU activation function, is the number of different layers, For the The weight coefficient of the layer, is the degree matrix, is the inverse matrix of the square root of the degree matrix, For the The adjacency matrix of the layer, For the The node characteristics of the layer, For the The weight matrix of the layer, is the adjustment coefficient, is the centrality matrix; the degree matrix is ​​a diagonal matrix with diagonal elements Representation Node The degree of the node The number of edges is expressed as:

[0016] ,

[0017] in, Nodes in the knowledge graph With Node weighted adjacency matrix elements between; construct the centrality matrix using different centrality indicators and calculate the node The centrality measure , expressed as:

[0018] ,

[0019] in, Through the intermediate node The number of shortest paths, For Node With Node The total number of shortest paths between .

[0020] As a preferred solution of the predictive complaint handling decision support method based on machine learning described in the present invention, wherein: the extraction of keywords and sentiment tendencies in text data includes calculating the contribution of each keyword in the complaint content to the sentiment score based on the hidden state and attention mechanism of the BERT model, and adjusting the sentiment impact through the time decay factor, calculating the text In time The sentiment score of is expressed as:

[0021] ,

[0022] in, For text In time The sentiment score, Indicates the number of the current keyword in the text sequence. is the number of keywords, is the attention weight, is the weight matrix of sentiment analysis, The first Keyword hidden status, is the bias of sentiment analysis, For keywords In text The number of occurrences in To contain keywords The number of texts, is the total number of texts, is the current time, Indicates The timestamp of the first occurrence of the keyword, is the time decay factor.

[0023] As a preferred solution of the predictive complaint handling decision support method based on machine learning described in the present invention, the predictive complaint handling demand scoring includes predicting and analyzing the complaint management system data through a machine learning algorithm, using a sentiment score matrix and the node feature matrix The combined features of are used to predict the handling requirements of complaints, which can be expressed as:

[0024] ,

[0025] in, represents the predicted complaint handling demand score, is the comprehensive feature transformation matrix, is the sentiment feature transformation matrix, is the sentiment score matrix, Build sentiment scores based on different texts, is the graph convolution feature transformation matrix, Scoring bias for complaint handling needs.

[0026] As a preferred embodiment of the predictive complaint handling decision support method based on machine learning described in the present invention, the generating of priority ranking and handling suggestions includes prioritizing the complaint handling needs based on the complaint handling needs scoring output result; when the predicted complaint handling needs score is When the complaint handling demand is low priority, the complaint content is judged as positive emotional feedback, the user is satisfied with the service, and the complaint management system records and automatically responds; when the predicted complaint handling demand score is When the predicted complaint handling demand score is , the complaint content is judged as neutral emotional feedback, and the user has no negative emotions about the service. The complaint management system records and automatically responds to the complaint and provides guidance on problem solving related to the complaint. When the predicted complaint handling demand score is , the complaint content is judged as general emotional feedback, and the user is confused about the service, the complaint management system notifies the staff to communicate online and records the communication process; when the predicted complaint handling demand score is When the complaint handling needs are urgent and the complaint content is judged to be serious emotional feedback, the user is strongly dissatisfied with the service. The complaint management system records and uploads the complaint content, notifies the staff to handle it immediately based on the user's specific needs, and provides solutions and guidance to the user until the customer ends the complaint.

[0027] Another object of the present invention is to provide a predictive complaint handling decision support system based on machine learning, which can extract the features of nodes and edges from the knowledge graph, extract keywords and sentiment tendencies in text data, and solve the problem of low accuracy in current complaint data processing.

[0028] As a preferred solution of the predictive complaint handling decision support system based on machine learning described in the present invention, it includes: a data collection module, a keyword processing module, and a demand scoring module; the data collection module is used to collect complaint management system data, pre-process the data, determine the main entities and relationships in the knowledge graph to construct a knowledge graph; the keyword processing module is used to extract node and edge features from the knowledge graph, and extract keywords and sentiment tendencies in text data; the demand scoring module is used to predict complaint handling demand scores based on machine learning algorithms, and generate priority rankings and processing suggestions for each complaint.

[0029] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a predictive complaint handling decision support method based on machine learning.

[0030] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a predictive complaint handling decision support method based on machine learning.

[0031] Beneficial effects of the present invention: The predictive complaint handling decision support method based on machine learning provided by the present invention improves the quality and consistency of data through systematic data collection and preprocessing, ensures the accuracy and reliability of subsequent analysis, and the construction of the knowledge graph enables complex complaint data to be structured and visualized, providing strong support for further analysis and decision-making, and improving the efficiency and accuracy of complaint handling. Through deep feature extraction and sentiment analysis, a comprehensive analysis and understanding of complaint data is achieved. The combination of knowledge graph features and text sentiment features makes the analysis results more accurate and comprehensive, which can better support subsequent predictions and decisions, and improves the accuracy of complaint handling and customer satisfaction. Through the application of machine learning algorithms, the automation and intelligence of complaint handling is achieved, and priority sorting and processing suggestion generation are performed according to the predicted demand scores, which can improve the efficiency and response speed of complaint handling, reduce manual intervention, and reduce processing costs. At the same time, it ensures that emergency complaints are given priority and improves customer satisfaction and service quality. The present invention achieves better results in processing efficiency, accuracy and cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0033] Figure 1An overall flow chart of a predictive complaint handling decision support method based on machine learning provided for the first embodiment of the present invention.

[0034] Figure 2 A schematic diagram of module functions of a predictive complaint handling decision support system based on machine learning provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0036] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a predictive complaint handling decision support method based on machine learning, comprising:

[0037] S1: Collect complaint management system data, pre-process the data, determine the main entities and relationships in the knowledge graph to construct the knowledge graph.

[0038] Furthermore, preprocessing includes collecting complaint management system data, which includes customer complaint content, complaint time, customer information, product information, and complaint processing history data, and performing data cleaning, data integration, and text preprocessing on the complaint management system data; data cleaning includes processing missing values, outliers, unifying data formats, and performing data standardization; data integration includes integrating data from different data sources to generate a unified data set; text preprocessing performs word segmentation, stop word removal, and stem extraction on complaint text data.

[0039] It should be noted that constructing a knowledge graph includes determining the main entities and relationships in the knowledge graph. The main entities include customers, products, complaints, and solutions. The relationships include customers and complaints, complaints and products, and complaints and solutions. Each entity represents a node, and the edge represents the relationship between entities. Different categories of relationship information are integrated to construct a knowledge graph, which can be expressed as:

[0040] ,

[0041] in, Nodes in the knowledge graph With Node The weighted adjacency matrix elements between is the category index of the relation, is the total number of relationship categories, For the The weight of the class relationship, Indicates Node under class relationship With Node The connection status, Representation Node With Node In the Connected under class relationship, Representation Node With Node In the Not connected under class relationship; calculate the Weight of class relationship It is expressed as:

[0042] ,

[0043] in, Indicates The frequency of class relations, Indicates The importance of class relationships, Indicates The frequency of class relations, Indicates The importance of class relationships, Indicates the difference from The index of the relationship category.

[0044] It should also be noted that through systematic data collection and preprocessing, the present invention ensures the quality and consistency of the data, provides a reliable data basis for subsequent analysis and modeling, data cleaning removes noise and redundant information, data integration solves the compatibility problem of multi-source data, and text preprocessing lays the foundation for text data analysis, which not only improves the efficiency of data processing, but also ensures the accuracy of data analysis, and helps to improve the overall effect of complaint handling.

[0045] S2: Extract node and edge features from the knowledge graph, and extract keywords and sentiment tendencies from text data.

[0046] Furthermore, extracting the features of nodes and edges includes extracting and adjusting the features of nodes and edges in the complaint management system data by propagating and combining node features at multiple levels and combining centrality indicators, which can be expressed as:

[0047] ,

[0048] in, For the The node feature matrix of the layer, is the ReLU activation function, is the number of different layers, For the The weight coefficient of the layer, is the degree matrix, is the inverse matrix of the square root of the degree matrix, For the The adjacency matrix of the layer, For the The node characteristics of the layer, For the The weight matrix of the layer, is the adjustment coefficient, is the centrality matrix; the degree matrix is ​​a diagonal matrix with diagonal elements Representation Node The degree of the node The number of edges is expressed as:

[0049] ,

[0050] in, Nodes in the knowledge graph With Node weighted adjacency matrix elements between; construct the centrality matrix using different centrality indicators and calculate the node The centrality measure , expressed as:

[0051] ,

[0052] in, Through the intermediate node The number of shortest paths, For Node With Node The total number of shortest paths between .

[0053] It should be noted that extracting keywords and sentiment trends from text data includes calculating the contribution of each keyword in the complaint content to the sentiment score based on the hidden state and attention mechanism of the BERT model, adjusting the sentiment impact through the time decay factor, and calculating the text In time The sentiment score of is expressed as:

[0054] ,

[0055] in, For text In time The sentiment score, Indicates the number of the current keyword in the text sequence. is the number of keywords, is the attention weight, is the weight matrix of sentiment analysis, The first Keyword hidden status, is the bias of sentiment analysis, For keywords In text The number of occurrences in To contain keywords The number of texts, is the total number of texts, is the current time, Indicates The timestamp of the first occurrence of the keyword, is the time decay factor.

[0056] It should also be noted that the present invention realizes a structured representation of complaint data by constructing a knowledge graph, which can capture the complex relationships and features in the data. By extracting the features of nodes and edges through a graph convolutional network and combining it with centrality indicators, it can deeply analyze the relationship between customers, products, complaints and solutions. The application of the BERT model makes sentiment analysis more accurate and can effectively capture the emotional tendencies in customer complaints. These technologies not only improve the depth and breadth of data analysis, but also provide strong support for subsequent predictions.

[0057] S3: Predict complaint handling needs scores based on machine learning algorithms and generate priority ranking and handling suggestions for each complaint.

[0058] Furthermore, predictive complaint handling needs scoring involves predictive analysis of complaint management system data through machine learning algorithms, using sentiment score matrices and the node feature matrix The combined features of are used to predict the handling requirements of complaints, which can be expressed as:

[0059] ,

[0060] in, represents the predicted complaint handling demand score, is the comprehensive feature transformation matrix, is the sentiment feature transformation matrix, is the sentiment score matrix, Build sentiment scores based on different texts, is the graph convolution feature transformation matrix, Scoring bias for complaint handling needs.

[0061] It should be noted that generating priority ranking and handling suggestions includes prioritizing complaint handling needs based on the complaint handling needs score output; when the predicted complaint handling needs score is When the complaint handling demand is low priority, the complaint content is judged as positive emotional feedback, the user is satisfied with the service, and the complaint management system records and automatically responds; when the predicted complaint handling demand score is When the predicted complaint handling demand score is , the complaint content is judged as neutral emotional feedback, and the user has no negative emotions about the service. The complaint management system records and automatically responds to the complaint and provides guidance on problem solving related to the complaint. When the predicted complaint handling demand score is , the complaint content is judged as general emotional feedback, and the user is confused about the service, the complaint management system notifies the staff to communicate online and records the communication process; when the predicted complaint handling demand score is When the complaint handling needs are urgent and the complaint content is judged to be serious emotional feedback, the user is strongly dissatisfied with the service. The complaint management system records and uploads the complaint content, notifies the staff to handle it immediately based on the user's specific needs, and provides solutions and guidance to the user until the customer ends the complaint.

[0062] It should also be noted that the present invention achieves accurate prediction of complaint handling demand scores through the application of machine learning algorithms. The comprehensive feature transformation matrix combines sentiment features and node features to make the prediction results more accurate and comprehensive. Based on the priority sorting of the prediction results, complaint handling resources can be effectively allocated to ensure that emergency complaints are given priority, thereby improving the efficiency of complaint handling and customer satisfaction. Compared with traditional manual processing methods, the application of machine learning algorithms has improved the automation and intelligence level of complaint handling.

[0063] Example 2, an embodiment of the present invention, provides a predictive complaint handling decision support method based on machine learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0064] First, a batch of complaint management system data from a large e-commerce platform was collected. These data include customer complaint content, complaint time, customer information, product information and complaint processing history data. After data collection, data preprocessing began. Data preprocessing includes data cleaning, data integration and text preprocessing. In the data cleaning step, missing values ​​and outliers are first processed by using mean filling and outlier filtering methods. Then, the data format is unified and standardized to ensure that data from different sources have consistent formats and dimensions. In the data integration step, customer information, product information and complaint content from different data sources are merged to generate a unified For the first dataset, text preprocessing involves word segmentation, stop word removal, and stem extraction of complaint text data to extract effective key information. After data preprocessing is completed, the knowledge graph is constructed to determine the main entities and relationships in the knowledge graph, including customers, products, complaints, and solutions. Each entity represents a node, and the edge represents the relationship between entities. By integrating relationship information of different categories, a weighted adjacency matrix is ​​constructed to ensure that the weight of each node and its relationship is reasonably calculated. The weight calculation is based on the frequency and importance of the relationship to ensure that the knowledge graph can accurately reflect the actual situation. After the knowledge graph is constructed, the features of the nodes and edges are extracted. Using the graph convolutional network (GCN), the node features are propagated and combined at multiple levels. Combined with the centrality index, the features of the nodes and edges in the complaint management system data are extracted and adjusted. Through multi-level propagation and combination, the complex relationships and features between nodes can be captured, thereby generating a new node feature matrix. These features can better reflect the complex relationship between customers, products and complaints. Subsequently, based on the hidden state and attention mechanism of the BERT model, the keywords and sentiment tendencies in the text data are extracted. The contribution of each keyword to the sentiment score is calculated by the BERT model, and the sentiment impact is adjusted in combination with the time decay factor to generate an accurate sentiment score. By allocating attention weights to keywords, the sentiment analysis results are ensured to be more in line with the actual situation. Based on the machine learning algorithm, the complaint handling demand score is predicted. The combined features of the sentiment score matrix and the node feature matrix are used to generate the predicted complaint handling demand score. According to the predicted demand score, priority sorting and processing suggestions are generated for each complaint. According to different demand score ranges, corresponding priority sorting and processing suggestions are generated for each complaint. From low priority to urgent needs, the processing intensity and response speed are gradually increased to ensure that each complaint can be handled in a timely and appropriate manner. Refer to Table 1 to record and analyze some experimental data.

[0065] Table 1 Complaint management data analysis table

[0066] Complaint ID Customer ID Product ID Complaint handling needs rating Priority C001 C123 P456 0.2 Low Priority C002 C124 P457 0.4 generally C003 C125 P458 0.6 important C004 C126 P459 0.8 urgent C005 C127 P460 0.3 generally C006 C128 P461 0.7 important

[0067] Through the analysis of the table data, it can be seen that the present invention has advantages in complaint handling demand prediction and priority sorting. Compared with the existing technology, the present invention improves data processing efficiency and accuracy through automation and intelligent means, can better support the company's complaint management and decision-making process, and ultimately improves customer satisfaction and service quality.

[0068] Example 3, reference Figure 2 , which is an embodiment of the present invention, provides a predictive complaint handling decision support system based on machine learning, including a data collection module, a keyword processing module, and a demand scoring module.

[0069] The data collection module is used to collect complaint management system data, pre-process the data, determine the main entities and relationships in the knowledge graph to construct the knowledge graph; the keyword processing module is used to extract node and edge features from the knowledge graph, and extract keywords and sentiment tendencies in text data; the demand scoring module is used to predict the complaint handling demand score based on a machine learning algorithm, and generate priority ranking and processing suggestions for each complaint.

[0070] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0071] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0072] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk case (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0073] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logical function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A predictive complaint handling decision support method based on machine learning, characterized in that: include: Collect complaint management system data, pre-process the data, determine the main entities and relationships in the knowledge graph to build the knowledge graph; Extract node and edge features from the knowledge graph, and extract keywords and sentiment trends from text data; Predict complaint handling needs scores based on machine learning algorithms, and generate priority ranking and handling suggestions for each complaint; Constructing a knowledge graph includes determining the main entities and relationships in the knowledge graph. The main entities include customers, products, complaints, and solutions. The relationships include customers and complaints, complaints and products, and complaints and solutions. Each entity represents a node, and the edge represents the relationship between entities. Different categories of relationship information are integrated to construct a knowledge graph, which can be expressed as: Among them, A ij is the weighted adjacency matrix element between node i and node j in the knowledge graph, q is the category index of the relationship, Q is the total number of relationship categories, and w q is the weight of the qth type of relationship, Indicates the connection between node i and node j under the qth type of relationship, Indicates that node i is connected to node j under the qth type of relationship, It means that node i and node j are not connected under the qth type of relationship; Calculate the weight w of the qth relationship q It is expressed as: Among them, freq(q) represents the frequency of the qth relationship, imp(q) represents the importance of the qth relationship, freq(q') represents the frequency of the q'th relationship, imp(q') represents the importance of the q'th relationship, and q' represents the index of the relationship category different from q; Extracting node and edge features involves multi-level propagation and combination of node features, combined with centrality indicators, extracting and adjusting the features of nodes and edges in the complaint management system data, expressed as: Among them, H (l+1) is the node feature matrix of the l+1th layer, σ is the ReLU activation function, K is the number of different layers, β k is the weight coefficient of the kth layer, D is the degree matrix, D -1 / 2 is the inverse matrix of the square root of the degree matrix, A k is the adjacency matrix of the kth layer, H (l) is the node feature of the lth layer, W (l) is the weight matrix of the lth layer, γ is the adjustment coefficient, and C is the centrality matrix; The degree matrix is ​​a diagonal matrix, and the diagonal elements D ii represents the degree of node i, that is, the number of edges connected to node i, expressed as: D ii =S j A ij Among them, A ij is the weighted adjacency matrix element between node i and node j in the knowledge graph; Use different centrality indicators to construct the centrality matrix and calculate the centrality measure C of node i. i , expressed as: Among them, τ ij (m) is the number of shortest paths passing through the intermediate node m, τ ij is the total number of shortest paths between node i and node j; Extracting keywords and sentiment trends from text data involves calculating the contribution of each keyword in the complaint content to the sentiment score based on the hidden state and attention mechanism of the BERT model, and adjusting the sentiment impact through the time decay factor. The sentiment score of text d at time t is calculated as: Among them, S t,d is the sentiment score of text d at time t, n represents the number of the current keyword in the text sequence, N is the number of keywords, and α n is the attention weight, W s is the weight matrix of sentiment analysis, h n is the nth keyword hidden state of the BERT model, b s is the bias for sentiment analysis, TF ω,d is the number of occurrences of keyword ω in text d, DF(ω) is the number of texts containing keyword ω, M is the total number of texts, t is the current time, and t n It represents the timestamp of the first appearance of the nth keyword, and δ is the time decay factor; Predictive complaint handling demand scoring involves predictive analysis of complaint management system data through machine learning algorithms, using sentiment score matrix S' and node feature matrix H( l +1) to predict the handling requirements of complaints, expressed as: in, represents the predicted complaint handling demand score, W3 is the comprehensive feature transformation matrix, W1 is the sentiment feature transformation matrix, S' is the sentiment score matrix, S' is constructed based on the sentiment scores of different texts, W2 is the graph convolution feature transformation matrix, and B is the complaint handling demand score bias.

2. The predictive complaint handling decision support method based on machine learning according to claim 1, characterized in that: The preprocessing includes collecting complaint management system data, which includes customer complaint content, complaint time, customer information, product information, complaint processing history data, and performing data cleaning, data integration, and text preprocessing on the complaint management system data; Data cleaning includes processing missing values, outliers, unifying data formats, and standardizing data; Data integration involves integrating data from different data sources to generate a unified data set; Text preprocessing includes word segmentation, stop word removal, and stem extraction of complaint text data.

3. The predictive complaint handling decision support method based on machine learning as claimed in claim 1, characterized in that: The generating of the priority ranking and handling suggestions includes prioritizing the complaint handling needs based on the complaint handling needs scoring output; When predicted complaint handling needs are scored When the complaint handling demand is low priority, the complaint content is judged as positive emotional feedback, the user is satisfied with the service, and the complaint management system records and automatically responds; When predicted complaint handling needs are scored When the complaint handling demand is general, the complaint content is judged as neutral emotional feedback, and the user has no negative emotions about the service, the complaint management system records and automatically responds, and provides guidance content on problem solving related to the complaint; When predicted complaint handling needs are scored When the complaint handling demand is important, the complaint content is judged as general emotional feedback, and the user is confused about the service, the complaint management system notifies the staff to communicate online and records the communication process; When predicted complaint handling needs are scored When the complaint handling needs are urgent and the complaint content is judged to be serious emotional feedback, the user is strongly dissatisfied with the service. The complaint management system records and uploads the complaint content, notifies the staff to handle it immediately based on the user's specific needs, and provides solutions and guidance to the user until the customer ends the complaint.

4. A system using the predictive complaint handling decision support method based on machine learning as claimed in any one of claims 1 to 3, characterized in that: Including data collection module, keyword processing module, demand scoring module; The data collection module is used to collect complaint management system data, pre-process the data, determine the main entities and relationships in the knowledge graph to construct the knowledge graph; The keyword processing module is used to extract the features of nodes and edges from the knowledge graph, and extract keywords and sentiment tendencies from text data; The demand scoring module is used to predict the complaint handling demand score based on a machine learning algorithm and generate priority ranking and handling suggestions for each complaint.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the predictive complaint handling decision support method based on machine learning described in any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the predictive complaint handling decision support method based on machine learning described in any one of claims 1 to 3 are implemented.

Images