Customer complaint analysis method and system based on large model, and storage medium

Abstract

The invention provides a customer complaint analysis method and system based on a large model and a storage medium, and the method comprises the steps: inputting a customer complaint training sample into the large model for granularity decomposition, and obtaining word granularity representation and character granularity representation; performing local feature extraction on the character granularity representation to obtain a local feature vector, performing word embedding processing on the word granularity representation to obtain a context sensing vector, and performing inner convolution processing on the context sensing vector to obtain an inner convolution vector; fusing the inner convolution vector and the local feature vector to obtain a fusion vector, and decoding the fusion vector to obtain a sample classification result; determining model loss according to a sample classification result, and performing parameter updating on the large model according to the model loss; and inputting to-be-analyzed customer complaint information into the converged large model for customer complaint analysis to obtain a customer complaint processing plan. According to the embodiment of the invention, the most suitable customer complaint handling plan can be automatically matched based on the large model, and the efficiency and accuracy of generation of the customer complaint handling plan are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a customer complaint analysis method, system and storage medium based on a large model. Background Art

[0002] Customer complaints reflect dissatisfaction with products or services, and the reasons mainly cover product quality and service quality. When handling customer complaints, it is necessary to have a deep understanding of the problem and customer needs, clarify customer expectations, and implement customer complaint handling plans in a timely manner. In order to prevent the negative effects of customer complaints from expanding, the issue of customer complaint analysis is receiving more and more attention.

[0003] In the existing customer complaint analysis process, customer complaint analysis is generally performed manually to obtain customer complaint handling plans, resulting in a low accuracy rate in the generation of customer complaint handling plans. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a customer complaint analysis method, system and storage medium based on a large model to solve the problem of low accuracy in generating customer complaint handling plans in the prior art.

[0005] The embodiment of the present invention is implemented as follows: a customer complaint analysis method based on a large model, the method comprising:

[0006] Obtain customer complaint training samples, and input the customer complaint training samples into the large model for granularity decomposition to obtain word granularity representation and character granularity representation;

[0007] Extracting local features from the character granularity representation to obtain a local feature vector, and encoding the word granularity representation;

[0008] Performing word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and performing inner convolution processing on the context-aware vector to obtain an inner convolution vector;

[0009] Fusing the inner convolution vector and the local feature vector to obtain a fused vector, and decoding the fused vector to obtain a sample classification result;

[0010] Determine the model loss according to the sample classification result, and update the parameters of the large model according to the model loss until the large model converges;

[0011] The customer complaint information to be analyzed is input into the converged large model to perform customer complaint analysis, obtain customer complaint classification results, and determine a customer complaint handling plan based on the customer complaint classification results.

[0012] Preferably, word embedding processing is performed on the word granularity representation according to the encoding result to obtain a context-aware vector, including:

[0013] Encoding the words in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encoding the words in the word granularity representation according to the position encoding in the encoding result to obtain a position vector;

[0014] The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector.

[0015] Preferably, performing inner convolution processing on the context-aware vector to obtain an inner convolution vector includes:

[0016] Obtain a dimensionality reduction transformation matrix and a dimensionality increase transformation matrix, and determine an inner convolution kernel according to a first activation function, the context perception vector, the dimensionality reduction transformation matrix, and the dimensionality increase transformation matrix;

[0017] Generating inner convolution layers of different dimensions according to the inner convolution kernels, and stacking the inner convolution layers according to the dimensions to obtain stacked inner convolutions;

[0018] The context-aware vector is input into the stacked inner convolution for convolution processing to obtain a global convolution vector, and the global convolution vector is subjected to maximum pooling processing to obtain the inner convolution vector.

[0019] Preferably, the inner convolution vector and the local feature vector are fused to obtain a fused vector, comprising:

[0020] Acquire an alignment transformation matrix, and perform feature alignment on the inner convolution vector and the local feature vector according to the alignment transformation matrix;

[0021] Obtaining an update gate weight and an update gate bias, and determining a degree of feature fusion according to a second activation function, the update gate weight, the update gate bias, the inner convolution vector after feature alignment, and the local feature vector;

[0022] A weighted operation is performed on the inner convolution vector and the local feature vector after feature alignment according to the feature fusion degree to obtain the fusion vector.

[0023] Preferably, the customer complaint training samples are input into a large model for granularity decomposition to obtain word granularity representation and character granularity representation, including:

[0024] Segmenting the sample text in the customer complaint training sample into words and characters according to the large model to obtain a word sequence and a character sequence;

[0025] Matrix conversion is performed on the word sequence and the character sequence to obtain the word granularity representation and the character granularity representation.

[0026] Preferably, after decoding the fusion vector to obtain the sample classification result, the method further includes:

[0027] According to the sample classification result, judging whether the classification of the sample text by the large model is correct;

[0028] If the large model correctly classifies the sample text, the sample text is successfully labeled, and a set is divided according to the classification label of the sample text to obtain a classification set;

[0029] If the number of the successful markings in any of the classification sets is greater than the quantity threshold, performing text mutation processing on the sample text in the classification set to obtain a mutant text;

[0030] The sample text is replaced according to the variant text.

[0031] Preferably, a customer complaint handling plan is determined based on the customer complaint classification results, including:

[0032] The classification number in the customer complaint classification result is matched with the plan query table to obtain the customer complaint handling plan. The plan query table stores the corresponding relationship between different classification numbers and the corresponding customer complaint handling plans.

[0033] Another object of an embodiment of the present invention is to provide a customer complaint analysis system based on a large model, the system comprising:

[0034] A granularity decomposition module is used to obtain customer complaint training samples and input the customer complaint training samples into a large model for granularity decomposition to obtain word granularity representation and character granularity representation;

[0035] An encoding module, used for extracting local features from the character granularity representation to obtain a local feature vector, and encoding the word granularity representation;

[0036] An inner convolution module, used for performing word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and performing inner convolution processing on the context-aware vector to obtain an inner convolution vector;

[0037] A decoding module, used for fusing the inner convolution vector and the local feature vector to obtain a fused vector, and decoding the fused vector to obtain a sample classification result;

[0038] An updating module, used to determine a model loss according to the sample classification result, and update parameters of the large model according to the model loss until the large model converges;

[0039] The customer complaint analysis module is used to input the customer complaint information to be analyzed into the converged large model to perform customer complaint analysis, obtain customer complaint classification results, and determine the customer complaint handling plan based on the customer complaint classification results.

[0040] Preferably, the inner convolution module is also used for:

[0041] Encoding the words in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encoding the words in the word granularity representation according to the position encoding in the encoding result to obtain a position vector;

[0042] The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector.

[0043] The embodiments of the present invention perform customer complaint analysis on the customer complaint information to be analyzed through a converged large model, and can automatically match the most suitable customer complaint handling plan, thereby improving the pertinence and effectiveness of the customer complaint handling plan, avoiding the tediousness and inefficiency of manual analysis, and improving the efficiency and accuracy of generating customer complaint handling plans. Through efficient and accurate customer complaint analysis, it is helpful to quickly respond to and effectively handle customer complaints, thereby reducing the impact of events and the complaint rate, and improving customer satisfaction and brand image. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of a customer complaint analysis method based on a large model provided in the first embodiment of the present invention;

[0045] Figure 2 is a structural diagram of a customer complaint analysis system based on a large model provided in a second embodiment of the present invention;

[0046] Figure 3 It is a schematic diagram of the structure of a terminal device provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0049] Embodiment 1

[0050] See also Figure 1, is a flow chart of a customer complaint analysis method based on a big model provided in a first embodiment of the present invention. The customer complaint analysis method based on a big model can be applied to any device or system. The customer complaint analysis method based on a big model comprises the following steps:

[0051] Step S10, obtaining customer complaint training samples, and inputting the customer complaint training samples into the large model for granularity decomposition to obtain word granularity representation and character granularity representation;

[0052] Among them, historical customer complaints and standard customer complaint plans are obtained, data annotation is performed on historical customer complaints and standard customer complaint plans, and customer complaint training samples are obtained. By inputting the customer complaint training samples into the large model for granularity decomposition, the word granularity representation of the word dimension and the character granularity representation of the character can be effectively obtained.

[0053] Optionally, the customer complaint training sample is input into a large model for granularity decomposition to obtain word granularity representation and character granularity representation, including:

[0054] The sample text in the customer complaint training sample is segmented into words and characters according to the large model to obtain a word sequence and a character sequence; wherein, the sample text in the customer complaint training sample is detected for stop words according to the large model, the sample text is filtered for stop words according to the stop word detection result, and the sample text after the stop word filtering is segmented into words and characters to obtain a word sequence and a character sequence, wherein the stop word filtering is used to delete stop words without actual meaning in the sample text, thereby improving the quality of the sample text;

[0055] The word sequence and the character sequence are matrix-converted to obtain the word granularity representation and the character granularity representation; wherein, the word sequence and the character sequence are matrix-mapped according to a preset mapping relationship to obtain the word granularity representation and the character granularity representation.

[0056] Step S20, extracting local features from the character granularity representation to obtain a local feature vector, and encoding the word granularity representation;

[0057] Among them, the local feature extraction model can be used to extract local features of the character granularity representation to obtain a local feature vector, and the word granularity can be respectively encoded by word, position and sentence.

[0058] Step S30, performing word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and performing inner convolution processing on the context-aware vector to obtain an inner convolution vector;

[0059] Among them, by performing word embedding processing on the word granularity representation through the encoding result, the context-aware vector corresponding to the word granularity representation can be effectively obtained, and by performing inner convolution processing on the context-aware vector, the inner convolution vector corresponding to the word granularity representation can be effectively obtained.

[0060] Optionally, word embedding processing is performed on the word granularity representation according to the encoding result to obtain a context-aware vector, including:

[0061] Encode the words in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encode the words in the word granularity representation according to the position encoding in the encoding result to obtain a position vector; wherein the word encoding is used to map the vocabulary to the corresponding vector, the word vector is used to represent the semantic information of the corresponding vocabulary, and the position encoding is used to represent the position information of the vocabulary in the word granularity representation. The words in the word granularity representation can be encoded by using a cosine function or the like;

[0062] The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector; wherein, division marks are added between adjacent sentences in the sample text, and the division marks are used to mark different sentences, and the context-aware vector is obtained by vector addition of the word vector, the position vector and the sentence vector.

[0063] Further, the context-aware vector is subjected to inner convolution processing to obtain an inner convolution vector, including:

[0064] Obtain a dimensionality reduction transformation matrix and a dimensionality increase transformation matrix, and determine an inner convolution kernel according to a first activation function, the context perception vector, the dimensionality reduction transformation matrix, and the dimensionality increase transformation matrix; wherein the dimensionality reduction transformation matrix, the dimensionality increase transformation matrix, and the first activation function can all be set according to requirements, and the context perception vector is activated by the first activation function, the dimensionality reduction transformation matrix, and the dimensionality increase transformation matrix to obtain an inner convolution kernel;

[0065] Generating inner convolution layers of different dimensions according to the inner convolution kernel, and stacking the inner convolution layers according to the dimensions to obtain stacked inner convolution; wherein the inner convolution layers of different dimensions are stacked from high to low dimensions to obtain stacked inner convolution;

[0066] The context-aware vector is input into the stacked inner convolution for convolution processing to obtain a global convolution vector, and the global convolution vector is subjected to maximum pooling processing to obtain the inner convolution vector; wherein, by stacking multiple inner convolution layers of different dimensions, local feature information of different dimensions of the context-aware vector is obtained in a descending dimensional order, and the dimension of the feature vector is gradually reduced while extracting feature information of a higher level.

[0067] Step S40, fusing the inner convolution vector and the local feature vector to obtain a fused vector, and decoding the fused vector to obtain a sample classification result;

[0068] Among them, by fusing the inner convolution vector and the local feature vector, a fusion vector containing global features and local features can be effectively obtained, which improves the quality of data features in the large model training process. By decoding the fusion vector, the customer complaint training samples can be effectively classified to obtain sample classification results.

[0069] Optionally, fusing the inner convolution vector and the local feature vector to obtain a fused vector includes:

[0070] Obtain an alignment transformation matrix, and perform feature alignment on the inner convolution vector and the local feature vector according to the alignment transformation matrix; wherein the alignment transformation matrix can be set as required, and the alignment transformation matrix is ​​used to align the feature dimension of the inner convolution vector and the local feature vector;

[0071] Obtain an update gate weight and an update gate bias, and determine the degree of feature fusion according to the second activation function, the update gate weight, the update gate bias, the inner convolution vector after feature alignment, and the local feature vector; wherein the second activation function, the update gate weight, and the update gate bias can all be set according to demand, and based on the second activation function, the update gate weight, and the update gate bias, the degree of feature fusion of the inner convolution vector after feature alignment and the local feature vector is determined by using an update gate;

[0072] The inner convolution vector and the local feature vector after feature alignment are weighted according to the feature fusion degree to obtain the fusion vector; wherein, during the weighted operation, the feature fusion degree is the weighting coefficient of the inner convolution vector, and the sum of the weighting coefficient of the local feature vector and the feature fusion degree is 1.

[0073] Furthermore, after decoding the fusion vector to obtain the sample classification result, the method further includes:

[0074] According to the sample classification result, judging whether the classification of the sample text by the large model is correct;

[0075] If the large model correctly classifies the sample text, the sample text is successfully marked, and the set is divided according to the classification label of the sample text to obtain a classification set; wherein, if the large model correctly classifies the sample text, it is determined that the large model can effectively classify customer complaints for the sample text during the model training process, and the sample texts of the same type can be effectively divided into the same classification set by dividing the set by the classification label of the sample text;

[0076] If the number of successful markings in any of the classification sets is greater than the quantity threshold, the sample text in the classification set is subjected to text mutation processing to obtain a mutated text, and the sample text is replaced according to the mutated text; wherein, the quantity threshold can be set as required. If the number of successful markings in any of the classification sets is greater than the quantity threshold, it is determined that during the model training process, the large model can effectively classify customer complaints for the sample texts of the corresponding type of the classification set. Therefore, by performing text mutation processing on the sample texts in the classification set to improve the training ability of the large model, the text mutation processing is used to replace similar words in the sample text.

[0077] Step S50, determining a model loss according to the sample classification result, and updating parameters of the large model according to the model loss until the large model converges;

[0078] The difference between the sample classification result and the classification label of the customer complaint training sample is calculated according to the loss function to obtain the model loss, and the model loss is back-propagated in the large model until the large model converges.

[0079] Step S60, inputting the customer complaint information to be analyzed into the converged large model to perform customer complaint analysis, obtain a customer complaint classification result, and determine a customer complaint handling plan based on the customer complaint classification result;

[0080] Among them, by using the converged large model to conduct customer complaint analysis on the customer complaint information to be analyzed, it can automatically match the most suitable customer complaint handling plan and obtain a customer complaint handling plan.

[0081] Optionally, determining a customer complaint handling plan based on the customer complaint classification result includes: matching the classification number in the customer complaint classification result with a plan query table to obtain the customer complaint handling plan, wherein the plan query table stores the correspondence between different classification numbers and corresponding customer complaint handling plans.

[0082] In this embodiment, by using the converged large model to perform customer complaint analysis on the customer complaint information to be analyzed, the most suitable customer complaint handling plan can be automatically matched, thereby improving the pertinence and effectiveness of the customer complaint handling plan, avoiding the tediousness and inefficiency of manual analysis, and improving the efficiency and accuracy of generating customer complaint handling plans. Through efficient and accurate customer complaint analysis, it helps to quickly respond to and effectively handle customer complaints, thereby reducing the impact of events and the complaint rate, and improving customer satisfaction and brand image.

[0083] Embodiment 2

[0084] See also Figure 2 , is a schematic diagram of the structure of a customer complaint analysis system 100 based on a large model provided in a second embodiment of the present invention, including:

[0085] The granularity decomposition module 10 is used to obtain customer complaint training samples and input the customer complaint training samples into the large model for granularity decomposition to obtain word granularity representation and character granularity representation.

[0086] Optionally, the granularity decomposition module 10 is further used to: perform word segmentation and character segmentation on the sample text in the customer complaint training sample according to the large model to obtain a word sequence and a character sequence;

[0087] Matrix conversion is performed on the word sequence and the character sequence to obtain the word granularity representation and the character granularity representation.

[0088] The encoding module 11 is used to extract local features of the character granularity representation to obtain a local feature vector and encode the word granularity representation.

[0089] The inner convolution module 12 is used to perform word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and perform inner convolution processing on the context-aware vector to obtain an inner convolution vector.

[0090] Optionally, the inner convolution module 12 is further used to: encode the word in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encode the word in the word granularity representation according to the position encoding in the encoding result to obtain a position vector;

[0091] The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector.

[0092] Furthermore, the inner convolution module 12 is also used to: obtain a dimensionality reduction transformation matrix and a dimensionality increase transformation matrix, and determine an inner convolution kernel according to the first activation function, the context perception vector, the dimensionality reduction transformation matrix and the dimensionality increase transformation matrix;

[0093] Generating inner convolution layers of different dimensions according to the inner convolution kernels, and stacking the inner convolution layers according to the dimensions to obtain stacked inner convolutions;

[0094] The context-aware vector is input into the stacked inner convolution for convolution processing to obtain a global convolution vector, and the global convolution vector is subjected to maximum pooling processing to obtain the inner convolution vector.

[0095] The decoding module 13 is used to fuse the inner convolution vector and the local feature vector to obtain a fused vector, and decode the fused vector to obtain a sample classification result.

[0096] Optionally, the decoding module 13 is further used to: obtain an alignment transformation matrix, and perform feature alignment on the inner convolution vector and the local feature vector according to the alignment transformation matrix;

[0097] Obtaining an update gate weight and an update gate bias, and determining a degree of feature fusion according to a second activation function, the update gate weight, the update gate bias, the inner convolution vector after feature alignment, and the local feature vector;

[0098] A weighted operation is performed on the inner convolution vector and the local feature vector after feature alignment according to the feature fusion degree to obtain the fusion vector.

[0099] Furthermore, the decoding module 13 is also used to: determine whether the classification of the sample text by the large model is correct according to the sample classification result;

[0100] If the large model correctly classifies the sample text, the sample text is successfully labeled, and a set is divided according to the classification label of the sample text to obtain a classification set;

[0101] If the number of the successful markings in any of the classification sets is greater than the quantity threshold, performing text mutation processing on the sample text in the classification set to obtain a mutant text;

[0102] The sample text is replaced according to the variant text.

[0103] The updating module 14 is used to determine the model loss according to the sample classification result, and update the parameters of the large model according to the model loss until the large model converges.

[0104] The customer complaint analysis module 15 is used to input the customer complaint information to be analyzed into the converged large model to perform customer complaint analysis, obtain customer complaint classification results, and determine a customer complaint handling plan based on the customer complaint classification results.

[0105] Optionally, the customer complaint analysis module 15 is further used to match the classification number in the customer complaint classification result with a plan query table to obtain the customer complaint handling plan, wherein the plan query table stores the correspondence between different classification numbers and corresponding customer complaint handling plans.

[0106] This embodiment integrates past customer complaints and builds a large model to achieve automatic classification and experience learning, forming a rich knowledge base. When a customer complaint occurs, it combines multi-modal and multi-dimensional data for comprehensive judgment and quickly matches the most suitable plan to efficiently assist in handling complaints and reduce the impact of the incident and the complaint rate. The specific plan is as follows:

[0107] Build a customer complaint case knowledge base: By collecting and organizing historical customer complaint cases, using big model technology to automatically classify, annotate and extract features from these cases, we build a comprehensive customer complaint case knowledge base. This knowledge base not only contains basic case information, but also covers multi-dimensional data such as the processing process, result feedback and lessons learned.

[0108] Automatic classification and experience learning of large models: Using advanced algorithms such as deep learning, the customer complaint case knowledge base is trained to enable the large model to automatically identify the categories, key issues and potential risks of new customer complaint cases, and give preliminary treatment suggestions based on historical experience. At the same time, by continuously learning new cases, the model performance is continuously optimized to improve the accuracy and timeliness of analysis.

[0109] Comprehensive judgment of multimodal and multidimensional data: When a customer complaint occurs, the system can automatically collect and integrate multimodal and multidimensional data (such as text descriptions, pictures, videos, voice, etc.) from different channels and use large models for comprehensive analysis. By comparing similar cases in the knowledge base, the system can more accurately judge the nature, scope of impact and possible development trends of the customer complaint.

[0110] Plan matching and intelligent recommendation: Based on the comprehensive judgment results, the system can automatically match the most suitable customer complaint handling plan and intelligently recommend it to the handling personnel for reference. The above plan not only includes specific handling steps and measures, but also provides risk assessment and response strategies, which helps to respond quickly and effectively reduce the impact of the incident.

[0111] In this embodiment, through automatic classification and experience learning of large models, the present invention can quickly and accurately identify the core issues and impact scope of customer complaints, avoiding the tediousness and inefficiency of manual analysis. Based on the comprehensive analysis of historical cases and real-time data, the system can automatically match the most suitable customer complaint handling plan, improving the pertinence and effectiveness of the plan. By continuously collecting new cases and optimizing model performance, the present invention can continuously enrich and improve the customer complaint case knowledge base, providing more powerful support for future customer complaint handling. Through efficient and accurate customer complaint analysis and intelligent plan matching, the present invention helps to quickly respond to and effectively handle customer complaints, thereby reducing the impact of events and complaint rates, and improving customer satisfaction and brand image.

[0112] Embodiment 3

[0113] Figure 3 2 is a block diagram of a terminal device 2 provided in the third embodiment of the present application. Figure 3As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program of a customer complaint analysis method based on a large model. When the processor 20 executes the computer program 22, the steps in each embodiment of the customer complaint analysis method based on a large model are implemented.

[0114] Exemplarily, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, a processor 20 and a memory 21.

[0115] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0116] The memory 21 may be an internal storage unit of the terminal device 2, such as a hard disk or memory of the terminal device 2. The memory 21 may also be an external storage device of the terminal device 2, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 2. Further, the memory 21 may also include both an internal storage unit and an external storage device of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 may also be used to temporarily store data that has been output or is to be output.

[0117] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional unit.

[0118] If the integrated module 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. Among them, the computer-readable storage medium can be non-volatile or volatile. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0119] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A customer complaint analysis method based on a large model, characterized in that: The method comprises: Obtain customer complaint training samples, and input the customer complaint training samples into the large model for granularity decomposition to obtain word granularity representation and character granularity representation; Extracting local features from the character granularity representation to obtain a local feature vector, and encoding the word granularity representation; Performing word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and performing inner convolution processing on the context-aware vector to obtain an inner convolution vector; Fusing the inner convolution vector and the local feature vector to obtain a fused vector, and decoding the fused vector to obtain a sample classification result; Determine the model loss according to the sample classification result, and update the parameters of the large model according to the model loss until the large model converges; The customer complaint information to be analyzed is input into the converged large model to perform customer complaint analysis, obtain customer complaint classification results, and determine a customer complaint handling plan based on the customer complaint classification results.

2. The customer complaint analysis method based on a large model as claimed in claim 1, characterized in that: According to the encoding result, the word granularity representation is subjected to word embedding processing to obtain a context-aware vector, including: Encoding the words in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encoding the words in the word granularity representation according to the position encoding in the encoding result to obtain a position vector; The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector.

3. The customer complaint analysis method based on a large model as claimed in claim 1, characterized in that: Performing inner convolution processing on the context-aware vector to obtain an inner convolution vector includes: Obtain a dimensionality reduction transformation matrix and a dimensionality increase transformation matrix, and determine an inner convolution kernel according to a first activation function, the context perception vector, the dimensionality reduction transformation matrix, and the dimensionality increase transformation matrix; Generating inner convolution layers of different dimensions according to the inner convolution kernels, and stacking the inner convolution layers according to the dimensions to obtain stacked inner convolutions; The context-aware vector is input into the stacked inner convolution for convolution processing to obtain a global convolution vector, and the global convolution vector is subjected to maximum pooling processing to obtain the inner convolution vector.

4. The customer complaint analysis method based on a large model as claimed in claim 1, characterized in that: The inner convolution vector and the local feature vector are fused to obtain a fused vector, including: Acquire an alignment transformation matrix, and perform feature alignment on the inner convolution vector and the local feature vector according to the alignment transformation matrix; Obtaining an update gate weight and an update gate bias, and determining a degree of feature fusion according to a second activation function, the update gate weight, the update gate bias, the inner convolution vector after feature alignment, and the local feature vector; A weighted operation is performed on the inner convolution vector and the local feature vector after feature alignment according to the feature fusion degree to obtain the fusion vector.

5. The customer complaint analysis method based on a large model as claimed in claim 1, characterized in that: The customer complaint training samples are input into the large model for granularity decomposition to obtain word granularity representation and character granularity representation, including: Segmenting the sample text in the customer complaint training sample into words and characters according to the large model to obtain a word sequence and a character sequence; Matrix conversion is performed on the word sequence and the character sequence to obtain the word granularity representation and the character granularity representation.

6. The customer complaint analysis method based on a large model as claimed in claim 5, characterized in that: After decoding the fusion vector to obtain the sample classification result, the method further includes: According to the sample classification result, judging whether the classification of the sample text by the large model is correct; If the large model correctly classifies the sample text, the sample text is successfully labeled, and a set is divided according to the classification label of the sample text to obtain a classification set; If the number of the successful markings in any of the classification sets is greater than the quantity threshold, performing text mutation processing on the sample text in the classification set to obtain a mutant text; The sample text is replaced according to the variant text.

7. The customer complaint analysis method based on a large model as claimed in claim 1, characterized in that: Determine the customer complaint handling plan based on the customer complaint classification results, including: The classification number in the customer complaint classification result is matched with the plan query table to obtain the customer complaint handling plan. The plan query table stores the corresponding relationship between different classification numbers and the corresponding customer complaint handling plans.

8. A customer complaint analysis system based on a large model, characterized in that: The system comprises: A granularity decomposition module is used to obtain customer complaint training samples and input the customer complaint training samples into a large model for granularity decomposition to obtain word granularity representation and character granularity representation; An encoding module, used for extracting local features from the character granularity representation to obtain a local feature vector, and encoding the word granularity representation; An inner convolution module, used for performing word embedding processing on the word granularity representation according to the encoding result to obtain a context-aware vector, and performing inner convolution processing on the context-aware vector to obtain an inner convolution vector; A decoding module, used for fusing the inner convolution vector and the local feature vector to obtain a fused vector, and decoding the fused vector to obtain a sample classification result; An updating module, used to determine a model loss according to the sample classification result, and update parameters of the large model according to the model loss until the large model converges; The customer complaint analysis module is used to input the customer complaint information to be analyzed into the converged large model to perform customer complaint analysis, obtain customer complaint classification results, and determine the customer complaint handling plan based on the customer complaint classification results.

9. The customer complaint analysis system based on a large model as claimed in claim 8, characterized in that: The inner convolution module is also used to: Encoding the words in the word granularity representation according to the word encoding in the encoding result to obtain a word vector, and encoding the words in the word granularity representation according to the position encoding in the encoding result to obtain a position vector; The words in the word granularity representation are encoded according to the sentence encoding in the encoding result to obtain a sentence vector, and the word vector, the position vector and the sentence vector are combined to obtain the context-aware vector.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images