Large-model-driven customer complaint attribution system
Through a large-model-driven customer complaint attribution system, semantic coding and feature interaction processing technology, the problems of low efficiency of cross-language customer feedback processing and difficulty in handling exceptions in the existing technology are solved, and the customer complaint classification is achieved higher and system adaptability.
Patent Information
- Application Number
- CN202411586898.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-11-08
AI Technical Summary
Existing customer complaint attribution systems are inefficient and cannot flexibly handle exceptions when dealing with cross-language customer feedback, and the traditional way of pre-training classification models limits the flexibility and accuracy of the system.
Using a large-scale model-driven customer complaint attribution system, the system enhances the system's system by semantic encoding of customer feedback text data and significantly guiding semantic feature interaction processing with the knowledge base records. Flexibility and adaptability.
It improves the accuracy of customer complaint classification and systematic flexibility, can automatically match relevant records, provide better services and a basis for handling customer feedback.
Smart Images

Figure CN119088907A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of large model-driven systems, and more particularly to a large model-driven customer complaint attribution system in the embodiments of this application. Background Art
[0002] With the development of e-commerce, intelligent customer service systems have become an important tool for merchants to handle customer inquiries. This is especially true for cross-border e-commerce, where dealing with a large volume of customer feedback from different regions around the world and speaking multiple languages presents a significant challenge. Traditional methods typically rely on human customer service representatives, but due to language barriers and the sheer volume of feedback, this approach is inefficient and prone to errors.
[0003] Currently, some existing technologies for classifying customer complaints primarily operate by pre-training classification models. This means that a large amount of labeled data is required to train the model before deployment, enabling it to identify and classify different customer complaint scenarios. This approach limits the flexibility and accuracy of existing customer complaint attribution systems. Furthermore, it cannot flexibly communicate specific customer complaints as exceptions to the model, instructing the model to handle these exceptions specially.
[0004] In recent years, the development of Large Language Models (LLM) has provided a new approach to solving the aforementioned technical problems. LLMs possess the ability to handle complex cross-linguistic natural language descriptions, understand customer feedback, and perform attribution analysis, such as identifying issues like "poor product quality" or "delayed logistics." Compared to traditional methods, the powerful understanding capabilities of LLMs eliminate the need for pre-training specialized classification models and better handle non-standardized language expressions such as interjections and dialectal slang. By applying Large Language Models, the accuracy of customer complaint classification and the system's adaptability can be improved.
[0005] Furthermore, to further improve the accuracy and adaptability of classification, Retrieval Augmentation (RAG) technology has been introduced. RAG utilizes an embedding model to convert natural language customer feedback text data into vector form and stores it in a knowledge base. When a new query request is received, the system also converts this request into vector form and compares it with the vectors in the knowledge base, finding the most relevant records as contextual information to add to the LLM's prompts to obtain prompt information, thereby optimizing the customer complaint classification results. This approach is particularly suitable for complex cases requiring special handling and can serve as a method to correct or intervene in the LLM's expected results. Specifically, in the technical solution of this application, the knowledge base is mainly used to store some particularly difficult or special customer complaint cases. If these prompts are not provided to the LLM in the prompts, then using only the LLM's existing knowledge will not yield the expected customer complaint classification results. Therefore, applying RAG technology can serve as a method to correct or intervene in the LLM's expected results. For ordinary customer complaints whose results can be easily inferred through common sense, it is unnecessary to store their results in the RAG.
[0006] Therefore, an optimized customer complaint attribution system is desired. Summary of the Invention
[0007] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a large model-driven customer complaint attribution system. This system encodes each record in the knowledge base using the same encoding method as the customer feedback text data. Then, it performs significant guided semantic feature interaction processing between the customer feedback semantics and the semantics of each record to evaluate the matching degree between the customer feedback semantics and the semantics of each record in the knowledge base. This returns a number of records with a high degree of matching with the customer feedback semantics as the matching result. In this way, records related to the customer feedback semantics can be automatically matched from the knowledge base and provided as contextual information to the LLM to ensure the accuracy of customer complaint classification. This enhances the flexibility and adaptability of the customer complaint attribution system, providing a foundation for merchants to promptly handle customer feedback information and provide better services.
[0008] According to one aspect of this application, a large model-driven customer complaint attribution system is provided, comprising: a customer feedback data acquisition module for acquiring customer feedback text data; a customer feedback text semantic encoding module for performing semantic embedding encoding on the customer feedback text data to obtain a customer feedback semantic encoding vector; a record matching module for matching several records from a knowledge base based on the customer feedback semantic encoding vector, the records including customer feedback record text data and classification labels; a large model prompt word filling module for filling the customer feedback text data and the several records as context information into prompt words of a large language model to obtain prompt information; and a customer complaint attribution classification module for determining the customer complaint classification result of the customer feedback text data based on the prompt information.
[0009] In the aforementioned large-model-driven customer complaint attribution system, the record matching module includes: a knowledge base record semantic embedding encoding unit, used to perform semantic embedding encoding on each record in the knowledge base to obtain a set of record semantic embedding encoding vectors; a semantic feature joint saliency fusion unit, used to perform semantic feature joint saliency fusion on each record semantic embedding encoding vector in the set of customer feedback semantic encoding vectors and record semantic embedding encoding vectors to obtain a customer feedback-record semantic joint saliency fusion representation, including: a joint implicit feature analysis and activation processing subunit, used to perform joint implicit feature analysis and activation processing on the customer feedback semantic encoding vectors and record semantic embedding encoding vectors to obtain customer feedback-record semantic conditional features; a feature interaction fusion subunit, used to perform feature interaction fusion on the customer feedback semantic encoding vectors and record semantic embedding encoding vectors using the customer feedback-record semantic conditional features as conditional information to obtain the customer feedback-record semantic joint saliency fusion representation; and a matching degree estimation value analysis unit, used to perform matching degree evaluation based on the customer feedback-record semantic joint saliency fusion representation, determine the matching degree estimate, and determine whether to return a certain record as the matching result.
[0010] In the aforementioned large-model-driven customer complaint attribution system, the joint implicit feature analysis activation processing subunit includes: a joint implicit feature capture convolutional encoding secondary subunit, used to input the customer feedback semantic encoding vector and the record semantic embedding encoding vector into the joint implicit feature capture network to obtain a customer feedback-record semantic joint implicit feature vector; and a customer feedback-record semantic joint implicit feature activation secondary subunit, used to perform Sigmoid function-based feature activation on the customer feedback-record semantic joint implicit feature vector to obtain a customer feedback-record semantic conditional feature vector as the customer feedback-record semantic conditional feature.
[0011] In the aforementioned large-model-driven customer complaint attribution system, the joint implicit feature capture convolutional coding second-level subunit is used to: add the customer feedback semantic encoding vector and the record semantic embedding encoding vector positionally, multiply the resulting customer feedback-record semantic sum vector with the weight matrix, and then add it positionally with the bias vector to obtain the customer feedback-record semantic joint interaction vector; and process the customer feedback-record semantic joint interaction vector using the tanh function to obtain the customer feedback-record semantic joint implicit feature vector.
[0012] In the aforementioned large-model-driven customer complaint attribution system, the feature interaction fusion subunit includes: a secondary subunit for calculating customer feedback semantic contribution, used to calculate the customer feedback semantic contribution of the customer feedback semantic encoding vector relative to the customer feedback-record semantic conditional feature vector; a secondary subunit for calculating record semantic contribution, used to calculate the record semantic contribution of the record semantic embedding encoding vector relative to the customer feedback-record semantic conditional feature vector; and a secondary subunit for weighted modulation, used to normalize the customer feedback semantic contribution and the record semantic contribution, and use the normalized customer feedback semantic contribution and the normalized record semantic contribution to the customer feedback semantic encoding vector and the record semantic conditional feature vector. The semantic embedding encoding vector is weighted and modulated to obtain the modulated customer feedback semantic encoding vector and the modulated record semantic embedding encoding vector. The feature saliency guided interaction secondary subunit is used to input the modulated customer feedback semantic encoding vector, the modulated record semantic embedding encoding vector, and the customer feedback-record semantic conditional feature vector into the feature saliency guided interaction module based on the converter structure, using the modulated customer feedback semantic encoding vector as the query vector, the modulated record semantic embedding encoding vector as the key vector, and the customer feedback-record semantic conditional feature vector as the value vector, to obtain the customer feedback-record semantic joint saliency fusion representation vector as the customer feedback-record semantic joint saliency fusion representation.
[0013] In the aforementioned large-model-driven customer complaint attribution system, the secondary subunit for calculating the semantic contribution of customer feedback is used to: calculate the positional division of the semantic encoding vector of customer feedback and the semantic condition feature vector of customer feedback-record to obtain the semantic contribution vector of customer feedback; calculate the base-2 logarithmic function value of the absolute value of each feature value of the semantic contribution vector of customer feedback to obtain the logarithmic vector of customer feedback semantic contribution; calculate the positional dot product of the semantic encoding vector of customer feedback and the logarithmic vector of customer feedback semantic contribution, and perform positional addition on the resulting dot product vector to obtain the semantic contribution value of customer feedback; and calculate the exponential function with the natural constant e as the base, where the semantic contribution value of customer feedback is the exponent, to obtain the semantic contribution degree of customer feedback.
[0014] In the aforementioned large-model-driven customer complaint attribution system, the weighted modulation secondary subunit is used to: calculate the sum of the customer feedback semantic contribution and the record semantic contribution to obtain the total sum of customer feedback-record semantic contributions; divide the customer feedback semantic contribution and the record semantic contribution by the total sum of customer feedback-record semantic contributions to obtain the normalized customer feedback semantic contribution and the normalized record semantic contribution; perform a positional multiplication of the customer feedback semantic encoding vector with the normalized customer feedback semantic contribution to obtain the modulated customer feedback semantic encoding vector; and perform a positional multiplication of the record semantic embedding encoding vector with the normalized record semantic contribution to obtain the modulated record semantic embedding encoding vector.
[0015] In the aforementioned large-model-driven customer complaint attribution system, the second-level subunit for significant guided interaction between features is used to: multiply the modulated customer feedback semantic encoding vector with the transpose of the modulated record semantic embedding encoding vector; divide the resulting modulated customer feedback-record semantic association matrix by the square root of the length of the modulated record semantic embedding encoding vector to obtain the modulated customer feedback-record semantic association scaling matrix; activate the modulated customer feedback-record semantic association scaling matrix using the softmax function; and multiply the resulting modulated customer feedback-record semantic association scaling activation matrix with the customer feedback-record semantic conditional feature vector to obtain the customer feedback-record semantic joint significant fusion representation vector.
[0016] In the aforementioned large model-driven customer complaint attribution system, the matching degree estimation analysis unit is used to: input the customer feedback-record semantic joint saliency fusion representation vector into the decoder-based matching degree evaluation module to obtain a matching degree estimate; and determine whether to return a certain record as a matching result based on the comparison between the matching degree estimate and a preset threshold.
[0017] Compared to existing technologies, this application provides a large model-driven customer complaint attribution system. It encodes each record in the knowledge base using the same encoding method as the customer feedback text data, and then performs significant guided semantic feature interaction processing between the customer feedback semantics and the semantics of each record. This process evaluates the matching degree between the customer feedback semantics and the semantics of each record in the knowledge base, returning a few records with a high degree of matching with the customer feedback semantics as the matching results. In this way, it can automatically match records related to the customer feedback semantics from the knowledge base and provide them as contextual information to the LLM to ensure the accuracy of customer complaint classification. This enhances the flexibility and adaptability of the customer complaint attribution system, providing a foundation for businesses to handle customer feedback information promptly and provide better services. Attached Figure Description
[0018] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.
[0019] Figure 1 This is a system block diagram of a large model-driven customer complaint attribution system according to an embodiment of this application.
[0020] Figure 2 This is a block diagram of the recording matching module in a large model-driven customer complaint attribution system according to an embodiment of this application.
[0021] Figure 3 This is a schematic diagram of the data flow of the recording matching module in a large model-driven customer complaint attribution system according to an embodiment of this application.
[0022] Figure 4 This is a block diagram of a semantic feature joint saliency fusion unit in a large model-driven customer complaint attribution system according to an embodiment of this application.
[0023] Figure 5 This is a block diagram of the joint implicit feature analysis activation processing subunit in a large model-driven customer complaint attribution system according to an embodiment of this application.
[0024] Figure 6 This is a block diagram of a feature interaction fusion subunit in a large model-driven customer complaint attribution system according to an embodiment of this application. Detailed Implementation
[0025] Various exemplary embodiments, features, and aspects of this application will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.
[0026] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.
[0027] Furthermore, to better illustrate this application, numerous specific details are provided in the following detailed description. Those skilled in the art should understand that this application can be implemented without certain specific details. In some instances, methods, means, components, and circuits well-known to those skilled in the art have not been described in detail in order to highlight the main points of this application.
[0028] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.
[0029] To address the aforementioned technical issues, this application proposes a large-model-driven customer complaint attribution system. Figure 1 This is a system block diagram of a large-model-driven customer complaint attribution system according to an embodiment of this application. Figure 1 As shown, the large model-driven customer complaint attribution system 100 according to an embodiment of this application includes: a customer feedback data acquisition module 110, used to acquire customer feedback text data; a customer feedback text semantic encoding module 120, used to perform semantic embedding encoding on the customer feedback text data to obtain a customer feedback semantic encoding vector; a record matching module 130, used to match several records from a knowledge base based on the customer feedback semantic encoding vector, the records including customer feedback record text data and classification labels; a large model prompt word filling module 140, used to fill the customer feedback text data and the several records as context information into the prompt words of the large language model to obtain prompt information; and a customer complaint attribution classification module 150, used to determine the customer complaint classification result of the customer feedback text data based on the prompt information.
[0030] In the aforementioned large-scale model-driven customer complaint attribution system 100, the customer feedback data acquisition module 110 is used to acquire customer feedback text data. It should be understood that customer feedback is an important basis for enterprises to understand user needs, identify problems, and determine improvement directions. By analyzing customer feedback, enterprises can promptly identify shortcomings in their products or services, thereby making targeted improvements to meet user expectations and enhance customer loyalty. Specifically, in the technical solution of this application, customer feedback is acquired through a customer service system.
[0031] In the aforementioned large-model-driven customer complaint attribution system 100, the customer feedback text semantic encoding module 120 is used to perform semantic embedding encoding on the customer feedback text data to obtain a customer feedback semantic encoding vector. It should be understood that semantic embedding encoding maps text to a continuous vector space, thereby converting the original text data into a high-dimensional vector representation. This ensures that texts with similar meanings have similar representations in the vector space, better capturing the semantic information of the text. Specifically, firstly, the customer feedback text data undergoes data preprocessing, including noise removal (such as punctuation marks and special characters), word segmentation, and part-of-speech tagging, to standardize the customer feedback text and reduce unnecessary interference. Next, the preprocessed customer feedback text data is input into the BERT semantic embedding model. The input data needs to be converted into a format that the model can process, including segmenting the text into tokens, adding special tokens (such as [CLS] and [SEP]), and constructing corresponding attention masks. Afterward, the processed input data is passed to the model, which outputs the contextual embedding representation of each token through multiple levels of transformation. Finally, by extracting the vector corresponding to the [CLS] marker output by the model, or by averaging or weighted summing the vectors of all markers in the text, a fixed-dimensional semantic encoding vector is obtained. This vector can be regarded as the semantic embedding representation of customer feedback, i.e., the customer feedback semantic encoding vector, to capture the semantic information within it and improve the usability of the text data.
[0032] In the aforementioned large-model-driven customer complaint attribution system 100, the record matching module 130 is used to match several records from the knowledge base based on the semantic encoding vector of customer feedback. These records include customer feedback record text data and classification tags. It should be understood that matching several records from the knowledge base can more effectively associate customer feedback with existing data, enabling rapid and accurate response and decision-making. Specifically, in one embodiment of this application, the knowledge base records customer complaints and their classifications in a Q&A format, for example: Q: The desk I received is already broken.
[0033] A: Classification, quality, materials, damage.
[0034] Q: Thank you.
[0035] A: Category, service evaluation, thank you.
[0036] This structured recording method allows for the rapid retrieval and acquisition of answers and handling suggestions related to specific customer complaints, helping customer service personnel respond to customer issues more accurately. Simultaneously, the clear categorization enables the quick summarization and analysis of similar customer complaints, providing businesses with data-driven insights to optimize products and services.
[0037] In particular, in the aforementioned large model-driven customer complaint attribution system, the step of matching several records from the knowledge base based on the semantic encoding vector of customer feedback is crucial. This is because by converting the customer feedback text into a semantic encoding vector and comparing it with records in the knowledge base, the system can identify historical cases most similar to the current customer feedback. These historical cases and their classification labels are then added as contextual information to the prompts of the Large Language Model (LLM), thereby enhancing the LLM's understanding of the current feedback. This enhancement helps the LLM to classify customer complaints more accurately when processing new feedback. It is worth noting that the knowledge base stores processed and classified customer complaint cases that are particularly difficult or require special handling; these cases, to some extent, represent past problems and their solutions. By matching the semantics of the current customer feedback with records in the knowledge base, the system can draw on past successful experiences to guide the handling of the current problem and the generation of customer complaint classification results, avoiding repeated errors and ensuring the consistency of customer complaint classification results.
[0038] Specifically, in the process of matching several records from the knowledge base based on the semantic encoding vector of customer feedback, the technical concept of this application is to encode each record in the knowledge base using the same encoding method as the customer feedback text data, and then perform significant guided semantic feature interaction processing between the customer feedback semantics and the semantics of each record. This process evaluates the matching degree between the customer feedback semantics and the semantics of each record in the knowledge base, returning several records with a high degree of matching with the customer feedback semantics as the matching result. In this way, records related to the customer feedback semantics can be automatically matched from the knowledge base and provided as contextual information to the LLM to ensure the accuracy of customer complaint classification. This approach avoids the shortcomings of traditional methods, enhances the flexibility and adaptability of the customer complaint attribution system, and provides a foundation for merchants to handle customer feedback information in a timely manner and provide better services.
[0039] Figure 2 This is a block diagram of the recording matching module in a large model-driven customer complaint attribution system according to an embodiment of this application. Figure 3 This is a schematic diagram illustrating the data flow of the recording and matching module in a large-model-driven customer complaint attribution system according to an embodiment of this application. Figure 2 and Figure 3As shown, the record matching module 130 includes: a knowledge base record semantic embedding encoding unit 131, used to perform semantic embedding encoding on each record in the knowledge base to obtain a set of record semantic embedding encoding vectors; a semantic feature joint saliency fusion unit 132, used to perform semantic feature joint saliency fusion on each record semantic embedding encoding vector in the set of customer feedback semantic encoding vectors and the set of record semantic embedding encoding vectors to obtain a customer feedback-record semantic joint saliency fusion representation; and a matching degree estimation value analysis unit 133, used to perform matching degree evaluation based on the customer feedback-record semantic joint saliency fusion representation, determine the matching degree estimation value, and determine whether to return a certain record as a matching result.
[0040] Specifically, in the knowledge base record semantic embedding encoding unit 131, semantic embedding encoding is performed on each record in the knowledge base to obtain a set of record semantic embedding encoding vectors. It should be understood that traditional text matching methods often rely on keyword matching, making it difficult to capture deep semantic relationships between texts. However, through semantic embedding encoding, text data can be transformed into high-dimensional vectors, thereby preserving the basic information of the text, reflecting its semantic features, and achieving more accurate similarity comparison and retrieval in the vector space. Specifically, by using the Word2Vec model, word vectors are trained by predicting context words using the concept of context windows, obtaining a vector representation for each word to reflect its position in the semantic space. Next, to obtain the semantic embedding encoding vector for each record, an aggregation operation is needed on the word vectors contained in the text. Specifically, the vectors of all words in each record are extracted, and then the arithmetic mean of these vectors is calculated for aggregation. Finally, the semantic embedding encoding vectors of all records are collected together to form a vector set, i.e., the set of record semantic embedding encoding vectors. This set can be used for subsequent similarity calculations and information retrieval.
[0041] Specifically, in the semantic feature joint saliency fusion unit 132, semantic feature joint saliency fusion is performed on each of the record semantic embedding encoding vectors in the set of customer feedback semantic encoding vectors and the set of record semantic embedding encoding vectors to obtain a customer feedback-record semantic joint saliency fusion representation. It should be understood that the customer feedback semantic encoding vector contains semantic embedding encoding features related to customer feedback text data, and each of the record semantic embedding encoding vectors in the set of record semantic embedding encoding vectors contains semantic embedding encoding features of several records matched from the knowledge base. In order to perform saliency feature interaction fusion on the semantic information of these two entities, thereby generating an enhanced and more discriminative feature representation, and providing a basis for subsequent matching degree estimation and record return, in the technical solution of this application, semantic feature joint saliency fusion is further performed on each of the record semantic embedding encoding vectors in the set of customer feedback semantic encoding vectors and the set of record semantic embedding encoding vectors to obtain a customer feedback-record semantic joint saliency fusion representation. The customer feedback-record semantic joint saliency fusion representation can better reflect the essential relationship and saliency interaction fusion features between customer feedback and knowledge base records. This vector not only contains the original semantic information but also integrates the potential complex relationships between the two, enhancing the importance of the features and reducing interference from irrelevant information. This provides a more accurate foundation for subsequent matching evaluation and classification. This is crucial for improving the accuracy of customer complaint classification because it ensures the system can make decisions based on the most relevant information.
[0042] Figure 4 This is a block diagram of a semantic feature joint saliency fusion unit in a large model-driven customer complaint attribution system according to an embodiment of this application. Figure 4 As shown, the semantic feature joint saliency fusion unit 132 includes: a joint implicit feature analysis and activation processing subunit 1321, used to perform joint implicit feature analysis and activation processing on the customer feedback semantic encoding vector and the record semantic embedding encoding vector to obtain customer feedback-record semantic conditional features; and a feature interaction fusion subunit 1322, used to perform feature interaction fusion on the customer feedback semantic encoding vector and the record semantic embedding encoding vector with the customer feedback-record semantic conditional features as conditional information to obtain the customer feedback-record semantic joint saliency fusion representation.
[0043] Figure 5 This is a block diagram of the joint implicit feature analysis activation processing subunit in a large model-driven customer complaint attribution system according to an embodiment of this application. Figure 5As shown, the joint implicit feature analysis activation processing subunit 1321 includes: a joint implicit feature capture convolutional encoding secondary subunit 1321-1, used to input the customer feedback semantic encoding vector and the record semantic embedding encoding vector into the joint implicit feature capture network to obtain a customer feedback-record semantic joint implicit feature vector; and a customer feedback-record semantic joint implicit feature activation secondary subunit 1321-2, used to perform Sigmoid function-based feature activation on the customer feedback-record semantic joint implicit feature vector to obtain a customer feedback-record semantic conditional feature vector as the customer feedback-record semantic conditional feature.
[0044] Specifically, in the joint implicit feature capture convolutional coding secondary subunit 1321-1, the customer feedback semantic encoding vector and the record semantic embedding encoding vector are input into the joint implicit feature capture network to obtain a customer feedback-record semantic joint implicit feature vector. It should be understood that by inputting the semantic embedding encoding vectors of customer feedback and knowledge base records into the joint implicit feature capture network, a customer feedback-record semantic joint implicit feature vector can be generated. This new vector not only contains information from the original features but also captures the intrinsic relationship between customer feedback and knowledge base records, thereby generating a richer and more discriminative customer feedback-record semantic joint implicit feature representation.
[0045] In the embodiments of this application, the joint implicit feature capture convolutional coding secondary subunit 1321-1 includes: adding the customer feedback semantic coding vector and the record semantic embedding coding vector positionally, multiplying the resulting customer feedback-record semantic sum vector with the weight matrix, and then adding it positionally with the bias vector to obtain the customer feedback-record semantic joint interaction vector; and processing the customer feedback-record semantic joint interaction vector using the tanh function to obtain the customer feedback-record semantic joint implicit feature vector.
[0046] Specifically, in the secondary subunit 1321-2 of the joint implicit feature activation of customer feedback-record semantics, the joint implicit feature vector of customer feedback-record semantics is activated using the Sigmoid function to obtain the customer feedback-record semantic conditional feature vector as the customer feedback-record semantic conditional feature. It should be understood that using the Sigmoid function to activate the joint implicit feature vector of customer feedback-record semantics can amplify the parts most critical to describing the comprehensive conditions of customer feedback-record semantics, while suppressing irrelevant noise information. This ensures that the model focuses on features more important for subsequent matching degree estimation and customer complaint classification.
[0047] Figure 6This is a block diagram of a feature interaction fusion subunit in a large model-driven customer complaint attribution system according to an embodiment of this application. Figure 6 As shown, the feature interaction fusion subunit 1322 includes: a secondary subunit 1322-1 for calculating customer feedback semantic contribution, used to calculate the customer feedback semantic contribution of the customer feedback semantic encoding vector relative to the customer feedback-record semantic conditional feature vector; a secondary subunit 1322-2 for calculating record semantic contribution, used to calculate the record semantic contribution of the record semantic embedding encoding vector relative to the customer feedback-record semantic conditional feature vector; and a secondary subunit 1322-3 for normalizing the customer feedback semantic contribution and the record semantic contribution, and using the normalized customer feedback semantic contribution and the normalized record semantic contribution to calculate the customer feedback semantic encoding vector and the record semantic contribution. The semantic embedding encoding vector is weighted and modulated to obtain the modulated customer feedback semantic encoding vector and the modulated record semantic embedding encoding vector. The feature saliency guided interaction secondary subunit 1322-4 is used to input the modulated customer feedback semantic encoding vector, the modulated record semantic embedding encoding vector, and the customer feedback-record semantic conditional feature vector into the feature saliency guided interaction module based on the converter structure, using the modulated customer feedback semantic encoding vector as the query vector, the modulated record semantic embedding encoding vector as the key vector, and the customer feedback-record semantic conditional feature vector as the value vector, to obtain the customer feedback-record semantic joint saliency fusion representation vector as the customer feedback-record semantic joint saliency fusion representation.
[0048] Specifically, in the secondary subunit 1322-1 for calculating the semantic contribution of customer feedback and the secondary subunit 1322-2 for calculating the semantic contribution of records, the contributions of the customer feedback and record semantic encoding vectors to the customer feedback-record semantic conditional feature vector are calculated. This determines the importance of each semantic feature to the final representation, thus obtaining the semantic contribution of customer feedback and the semantic contribution of records. Through this evaluation, the system can identify which features are more important for the final feature representation and for subsequent matching degree estimation and customer complaint classification tasks.
[0049] In an embodiment of this application, the secondary subunit 1322-1 for calculating the customer feedback semantic contribution includes: calculating the positional division of the customer feedback semantic encoding vector and the customer feedback-record semantic condition feature vector to obtain a customer feedback semantic contribution vector; calculating the base-2 logarithmic function value of the absolute value of each feature value of the customer feedback semantic contribution vector to obtain a customer feedback semantic contribution logarithmic vector; calculating the positional dot product of the customer feedback semantic encoding vector and the customer feedback semantic contribution logarithmic vector, and performing positional addition on the resulting dot product vector to obtain a customer feedback semantic contribution value; and calculating an exponential function with the natural constant e as the base and the customer feedback semantic contribution value as the exponent to obtain the customer feedback semantic contribution degree.
[0050] Specifically, in the weighted modulation secondary subunit 1322-3, the customer feedback semantic contribution and the record semantic contribution are normalized. Then, the normalized customer feedback semantic contribution and the normalized record semantic contribution are used to perform weighted modulation on the customer feedback semantic encoding vector and the record semantic embedding encoding vector to obtain the modulated customer feedback semantic encoding vector and the modulated record semantic embedding encoding vector. It should be understood that by normalizing the customer feedback semantic contribution and the record semantic contribution, and adjusting the original customer feedback semantic encoding vector and the record semantic embedding encoding vector accordingly, it can be ensured that the more important customer feedback semantic features and record semantic features occupy a greater weight in the final feature representation. This helps the model better understand and process customer feedback information and interact and match the semantic features of these two elements.
[0051] In embodiments of this application, the weighted modulation secondary subunit 1322-3 includes: calculating the sum of the customer feedback semantic contribution and the recording semantic contribution to obtain a total sum of customer feedback-recording semantic contributions; dividing the customer feedback semantic contribution and the recording semantic contribution by the total sum of customer feedback-recording semantic contributions to obtain the normalized customer feedback semantic contribution and the normalized recording semantic contribution; performing a positional multiplication of the customer feedback semantic encoding vector with the normalized customer feedback semantic contribution to obtain the modulated customer feedback semantic encoding vector; and performing a positional multiplication of the recording semantic embedding encoding vector with the normalized recording semantic contribution to obtain the modulated recording semantic embedding encoding vector.
[0052] Specifically, in the secondary subunit 1322-4 of the feature-based saliency-guided interaction, the modulated customer feedback semantic encoding vector is used as the query vector, the modulated record semantic embedding encoding vector as the key vector, and the customer feedback-record semantic conditional feature vector as the value vector. The modulated customer feedback semantic encoding vector, the modulated record semantic embedding encoding vector, and the customer feedback-record semantic conditional feature vector are input into the feature-based saliency-guided interaction module based on the converter structure to obtain the customer feedback-record semantic joint saliency fusion representation vector as the customer feedback-record semantic joint saliency fusion representation. It should be understood that by using the feature-based saliency-guided interaction module based on the converter structure to interactively fuse the modulated feature vectors, this adaptive feature fusion method allows the model to simultaneously examine the interaction between customer feedback semantic features and record semantic features from multiple perspectives, and discover the most valuable information flow path.
[0053] In embodiments of this application, the secondary subunit 1322-4 of the significant guided interaction between features includes: multiplying the modulated customer feedback semantic encoding vector with the transpose of the modulated record semantic embedding encoding vector; dividing the resulting modulated customer feedback-record semantic association matrix by the square root of the length of the modulated record semantic embedding encoding vector to obtain a modulated customer feedback-record semantic association scaling matrix; activating the modulated customer feedback-record semantic association scaling matrix using the softmax function; and multiplying the resulting modulated customer feedback-record semantic association scaling activation matrix with the customer feedback-record semantic conditional feature vector to obtain the customer feedback-record semantic joint significant fusion representation vector.
[0054] Specifically, in the feature interaction fusion subunit 1322, using the customer feedback-record semantic conditional features as conditional information, feature interaction fusion is performed on the customer feedback semantic encoding vector and the record semantic embedding encoding vector to obtain the customer feedback-record semantic joint saliency fusion representation. It should be understood that by performing semantic feature joint saliency fusion processing on the customer feedback semantic encoding vector and the record semantic embedding encoding vector, the joint saliency key features and potential exercises between customer feedback semantics and record semantics can be integrated, strengthening the expressive ability of semantically related features with higher relevance to subsequent tasks and reducing interference from irrelevant information. This provides a foundation for subsequent matching degree evaluation and customer complaint classification, which is crucial for improving the accuracy of customer complaint classification.
[0055] More specifically, the semantic features of each record semantic embedding encoding vector in the set of customer feedback semantic encoding vectors and record semantic embedding encoding vectors are subjected to semantic feature joint saliency fusion using the following semantic feature joint formula to obtain the customer feedback-record semantic joint saliency fusion representation; wherein, the semantic feature joint formula is: .
[0056] in, and These are the customer feedback semantic encoding vector and the record semantic embedding encoding vector, respectively. To add based on location points, and These are the weight matrix and the bias vector, respectively. for function, For the customer feedback - record semantic joint implicit feature vector, for function, This represents the semantic conditional feature vector of the customer feedback record. For the feature values at each position in the semantic encoding vector of the customer feedback, The feature values at each position in the recorded semantic embedding encoding vector are... For the customer feedback - record the feature values at each position in the semantic conditional feature vector, The customer feedback semantic encoding vector and the number of feature values in the record semantic embedding encoding vector are given. This represents the value of the logarithmic function with base 2. This represents an exponential function with base e. Provide customers with feedback on semantic contribution. To record semantic contribution, and These are the normalized semantic contribution of customer feedback and the normalized semantic contribution of records, respectively. and These are the modulated client feedback semantic encoding vector and the modulated record semantic embedding encoding vector, respectively. yes The transpose of , yes Length, This represents vector multiplication. yes function, It is the semantic joint saliency fusion representation vector of the customer feedback-record.
[0057] In embodiments of this application, the matching degree estimation analysis unit 133 includes: inputting the customer feedback-record semantic joint saliency fusion representation vector into a decoder-based matching degree evaluation module to obtain a matching degree estimate; and determining whether to return a record as a matching result based on a comparison between the matching degree estimate and a preset threshold. It should be understood that decoding regression is performed using the joint saliency fusion features between customer feedback semantics and each record semantics to evaluate the matching degree between customer feedback semantics and the semantics of each record in the knowledge base. Furthermore, based on a comparison between the matching degree estimate and a preset threshold, it is determined whether to return a record as a matching result. Specifically, in response to the matching degree estimate being greater than the preset threshold, it is determined that a record is returned as a matching result. This allows for the automatic matching of records related to customer feedback semantics from the knowledge base and their provision as contextual information to the LLM to ensure the accuracy of customer complaint classification. This approach avoids the shortcomings of traditional methods, enhances the flexibility and adaptability of the customer complaint attribution system, and provides a foundation for merchants to promptly handle customer feedback information and provide better services.
[0058] In the aforementioned large-model-driven customer complaint attribution system 100, the large-model prompt word filling module 140 uses the customer feedback text data and several records as contextual information to fill the prompt words in the large language model to obtain prompt information. This generates more relevant and targeted prompt information. This method helps the model understand specific user needs, situations, and backgrounds, thereby providing more accurate and effective responses or suggestions.
[0059] As can be understood, a Large Language Model (LLM) is an artificial intelligence technology based on deep learning, designed to process and generate natural language text. In the technical solution of this application, since the initial knowledge base is empty, the classification results are mainly determined by the LLM's own reasoning ability. Professional personnel are needed to correct the system's classifications, storing incorrect or missing corrected classification results in the knowledge base. The LLM's classification results can be enhanced using RAG (Reference Object Group), i.e., added to the prompt through RAG retrieval. Thus, based on the prompt information, the LLM can improve its classification performance and reduce the error rate. Furthermore, as data accumulates, the number of useful records in the knowledge base gradually increases, providing the model with more effective reference information. Since customer feedback often exhibits similarity in content, after a period of human intervention and the establishment of a relatively complete knowledge base, the need for human intervention will be significantly reduced. At this point, the LLM's answer accuracy has significantly improved, enabling more efficient automated responses. It is worth mentioning that the Prompt template is used to organize prompts, context, and customer complaint information when inputting large models. One of the templates is described below.
[0060] "You are an after-sales expert who excels at categorizing customer feedback to help businesses improve product and service quality. Here is a customer feedback item, {feedback}. Some similar examples are {sample}. Your previous category was {category}. Please refer to these examples, but they may not be exactly the same."
[0061] Now, please select the most relevant customer complaint category from [{categories}] and output it, without outputting any extra characters. Here, {feedback} is filled with the customer complaint information, {sample} is filled with categories found from the knowledge base that best match the customer complaint information, {category} is filled with the previously selected parent categories, and {categories} is filled with the current level category selection.
[0062] Specifically, in the technical solution of this application, a Large Language Model (LLM) is introduced into the system, avoiding the need for pre-trained classification models. By using a multilingual Large Language Model (ChatGPT4, Qwen), the system can understand and respond to instructions in different languages, providing users with a flexible interactive experience. These Large Language Models possess excellent instruction compliance and reasoning capabilities, and can handle situations where the input and output languages are inconsistent.
[0063] In this way, by using the customer feedback text data and the aforementioned records as contextual information to populate the prompts in the large language model, the resulting prompts can include suggested solutions, detailed explanations of the customer's problem, or improvement measures based on customer feedback. This allows the model to respond more effectively to user needs and improve customer satisfaction.
[0064] In the aforementioned model-driven customer complaint attribution system 100, the customer complaint attribution classification module 150 is used to determine the customer complaint classification result of the customer feedback text data based on the prompt information. It should be understood that customer feedback text often contains various emotions and opinions, potentially involving multiple aspects such as product quality, service attitude, and functional requirements. To accurately classify this feedback, contextual information must be relied upon to help the model understand the customer's specific intent. For example, if the customer feedback mentions "slow after-sales service response," combining information such as service processes and historical response times in relevant records can help the model better determine the specific classification of this feedback. By combining customer feedback with relevant records, the generated prompt information provides the model with rich context, which not only helps in understanding the customer's specific needs but also improves the accuracy and efficiency of classification.
[0065] It is worth mentioning that the technical solution of this application designs a method that supports multi-level classification. Specifically, the system input is customer complaint information, and the output is multiple categories from top to bottom. In each level of customer complaint classification, after selecting a category, the system automatically reads subcategories from a JSON file as candidates for the next level of classification, until no subcategories remain at the next level, at which point all results are output. The customer complaint table is organized into JSON, and the program automatically determines the category at each level, using a loop to delve deeper layer by layer. This way, when the customer complaint classification table is modified, no program changes or model retraining are required; only adding or removing fields in the JSON file is needed to update the system. Here, the JSON-formatted customer complaint classification table represents the specific classifications of customer complaints by the merchants. Organizing it into JSON format facilitates program reading and adaptively populates the Prompt.
[0066] In summary, the large model-driven customer complaint attribution system 100 based on the embodiments of this application is explained. It encodes each record in the knowledge base using the same encoding method as the customer feedback text data, and then performs significant guided semantic feature interaction processing between the customer feedback semantics and the semantics of each record. This process evaluates the matching degree between the customer feedback semantics and the semantics of each record in the knowledge base, returning a number of records with a high degree of matching with the customer feedback semantics as the matching result. In this way, records related to the customer feedback semantics can be automatically matched from the knowledge base and provided as contextual information to the LLM to ensure the accuracy of customer complaint classification. This enhances the flexibility and adaptability of the customer complaint attribution system, providing a foundation for merchants to handle customer feedback information promptly and provide better services.
[0067] As described above, the large model-driven customer complaint attribution system 100 according to the embodiments of this application can be implemented in various terminal devices. In one example, the large model-driven customer complaint attribution system 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the large model-driven customer complaint attribution system 100 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the large model-driven customer complaint attribution system 100 can also be one of many hardware modules of the terminal device.
[0068] Alternatively, in another example, the large model-driven customer complaint attribution system 100 and the terminal device can also be separate devices, and the large model-driven customer complaint attribution system 100 can be connected to the terminal device via wired and / or wireless networks and transmit interactive information in accordance with an agreed data format.
[0069] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.
[0070] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.
[0071] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.
[0072] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.
[0073] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.
[0074] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. The term "second class" is used to indicate names and does not indicate any specific order.
[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit of the technical solutions of the present invention.
Claims
1. A customer complaint attribution system driven by a large model, characterized in that: include: A customer feedback data collection module is used to obtain customer feedback text data; a customer feedback text semantic encoding module is used to perform semantic embedding encoding on the customer feedback text data to obtain a customer feedback semantic encoding vector; a record matching module is used to match a number of records from a knowledge base based on the customer feedback semantic encoding vector, the records including customer feedback record text data and classification labels; a large model prompt word filling module is used to fill the customer feedback text data and the number of records as context information into the prompt word of the large language model to obtain prompt information; The customer complaint attribution classification module is used to determine the customer complaint classification result of the customer feedback text data based on the prompt information.
2. The large model-driven customer complaint attribution system according to claim 1 is characterized in that: The record matching module includes: a knowledge base record semantic embedding coding unit, which is used to perform semantic embedding coding on each record in the knowledge base to obtain a set of record semantic embedding coding vectors; a semantic feature joint significant fusion unit, which is used to perform semantic feature joint significant fusion on each record semantic embedding coding vector in the set of the customer feedback semantic coding vector and the record semantic embedding coding vector to obtain a customer feedback-record semantic joint significant fusion representation, including: a joint implicit feature analysis activation processing subunit, which is used to perform joint implicit feature analysis and activation processing on the customer feedback semantic coding vector and the record semantic embedding coding vector to obtain a customer feedback-record semantic conditional feature; a feature interaction fusion subunit, which is used to perform feature interaction fusion on the customer feedback semantic coding vector and the record semantic embedding coding vector with the customer feedback-record semantic conditional feature as conditional information to obtain the customer feedback-record semantic joint significant fusion representation; a matching degree estimation value analysis unit, which is used to perform matching degree evaluation based on the customer feedback-record semantic joint significant fusion representation, determine a matching degree estimation value, and determine whether to return a certain record as a matching result.
3. The large model-driven customer complaint attribution system according to claim 2 is characterized in that: The joint implicit feature analysis activation processing subunit includes: a joint implicit feature capture convolutional coding secondary subunit, which is used to input the customer feedback semantic coding vector and the record semantic embedding coding vector into a joint implicit feature capture network to obtain a customer feedback-record semantic joint implicit feature vector; and a customer feedback-record semantic joint implicit feature activation secondary subunit, which is used to perform feature activation based on a Sigmoid function on the customer feedback-record semantic joint implicit feature vector to obtain a customer feedback-record semantic conditional feature vector as the customer feedback-record semantic conditional feature.
4. The large model-driven customer complaint attribution system according to claim 3 is characterized in that: The joint implicit feature capture convolutional coding secondary subunit is used to: after adding the customer feedback semantic coding vector and the record semantic embedding coding vector by position point, multiply the obtained customer feedback-record semantic sum vector with the weight matrix and then add it by position with the bias vector to obtain the customer feedback-record semantic joint interaction vector; use the tanh function to process the customer feedback-record semantic joint interaction vector to obtain the customer feedback-record semantic joint implicit feature vector.
5. The large model-driven customer complaint attribution system according to claim 4 is characterized in that: The feature interaction fusion subunit includes: a customer feedback semantic contribution calculation secondary subunit, which is used to calculate the customer feedback semantic contribution of the customer feedback semantic coding vector relative to the customer feedback-record semantic conditional feature vector; a record semantic contribution calculation secondary subunit, which is used to calculate the record semantic contribution of the record semantic embedding coding vector relative to the customer feedback-record semantic conditional feature vector; a weighted modulation secondary subunit, which is used to normalize the customer feedback semantic contribution and the record semantic contribution, and use the normalized customer feedback semantic contribution and the normalized record semantic contribution to normalize the customer feedback semantic coding vector and the record semantic embedding coding vector. Weighted modulation is performed to obtain a modulated customer feedback semantic coding vector and a modulated record semantic embedding coding vector; a feature-to-feature significant guided interaction secondary subunit is used to use the modulated customer feedback semantic coding vector as a query vector, the modulated record semantic embedding coding vector as a key vector and the customer feedback-record semantic conditional feature vector as a value vector, and input the modulated customer feedback semantic coding vector, the modulated record semantic embedding coding vector and the customer feedback-record semantic conditional feature vector into a feature-to-feature significant guided interaction module based on a converter structure to obtain a customer feedback-record semantic joint significant fusion representation vector as the customer feedback-record semantic joint significant fusion representation.
6. The large model-driven customer complaint attribution system according to claim 5, characterized in that: The customer feedback semantic contribution calculation secondary subunit is used to: calculate the positional division of the customer feedback semantic encoding vector and the customer feedback-record semantic conditional feature vector to obtain a customer feedback semantic contribution vector; calculate the logarithmic function value with base two of the absolute value of each eigenvalue of the customer feedback semantic contribution vector to obtain a customer feedback semantic contribution logarithmic vector; calculate the positional dot product of the customer feedback semantic encoding vector and the customer feedback semantic contribution logarithmic vector, and perform positional multiplication on the obtained dot product vector to obtain a customer feedback semantic contribution value; calculate an exponential function with the natural constant e as the base and the customer feedback semantic contribution value as the exponent to obtain the customer feedback semantic contribution.
7. The large model-driven customer complaint attribution system according to claim 6, characterized in that: The weighted modulation secondary subunit is used to: calculate the sum of the customer feedback semantic contribution and the record semantic contribution to obtain a total sum of customer feedback-record semantic contribution; The customer feedback semantic contribution and the record semantic contribution are respectively divided by the total sum of the customer feedback-record semantic contributions to obtain the normalized customer feedback semantic contribution and the normalized record semantic contribution; the customer feedback semantic coding vector is point-multiplied by the normalized customer feedback semantic contribution to obtain the modulated customer feedback semantic coding vector; the record semantic embedding coding vector is point-multiplied by the normalized record semantic contribution to obtain the modulated record semantic embedding coding vector.
8. The large model-driven customer complaint attribution system according to claim 7, characterized in that: The feature-to-feature significant guided interaction secondary subunit is used to: vector-multiply the modulated customer feedback semantic coding vector with the transposed vector of the modulated record semantic embedding coding vector, divide the obtained modulated customer feedback-record semantic association matrix by the square root of the length of the modulated record semantic embedding coding vector by position to obtain a modulated customer feedback-record semantic association scaling matrix; use a softmax function to activate the modulated customer feedback-record semantic association scaling matrix, and multiply the obtained modulated customer feedback-record semantic association scaling activation matrix with the customer feedback-record semantic conditional feature vector to obtain the customer feedback-record semantic joint significant fusion representation vector.
9. The large model-driven customer complaint attribution system according to claim 8, characterized in that: The matching degree estimation value analysis unit is used to: input the customer feedback-record semantic joint significant fusion representation vector into a decoder-based matching degree evaluation module to obtain a matching degree estimation value; Based on the comparison between the match degree estimate and a preset threshold, it is determined whether to return a certain record as a match result.
Citation Information
Patent Citations
Training method for improving accuracy of large language model in answering legal compliance questions
CN118152628A
Dynamic adaptation question answering system and method based on hierarchical structure and retrieval enhancement
CN118193714A
Prompt word generation method and device, equipment, medium and program product
CN118296119A
Medical literature intelligent question answering system and method based on RAG and LLM technologies
CN118364088A
Multi-intention recognition method and system and readable storage medium
CN118861200A
Cited By
Data full life cycle operation and maintenance management system and method
CN119313314A
Intelligent construction hoist control system
CN119429885A
Customer complaint barrier analysis system based on generative artficial intelligence and method thereof
TWI906141B
