Customer complaint processing method and system based on emotion intensity quantification, electronic equipment and storage medium

Abstract

The invention discloses a customer complaint processing method and system based on emotion intensity quantification, electronic equipment and a storage medium. The method comprises the following steps: acquiring customer complaint information of a target object; inputting the customer complaint information into an emotion perception model, and perceiving context emotion to obtain target emotion; the emotion perception model is established based on a space-time emotion gating memory module; based on an industry emotion dictionary and the target emotion, quantifying the emotion intensity of the target object to obtain a target score of the target emotion; and determining a service priority of the target object according to the target score, and processing a target service related to the customer complaint information according to the service priority. According to the invention, through the emotion perception model based on the space-time emotion gating memory module, the context emotion is perceived, the emotion intensity is further quantified, the accuracy of emotion recognition is improved, the customer complaint processing efficiency is improved, and the customer satisfaction is improved. The method can be widely applied to the technical field of computers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and particularly relates to a complaint processing method and system based on emotion intensity quantification, an electronic device and a storage medium. BACKGROUND

[0002] With the rise of the service industry, there are complaint problems in various industries. For example, for a communication service provider, a large number of customer complaints are faced every day, involving network failures, billing problems, service quality and other aspects. The customer service department needs to handle a large amount of telephone and online interaction information to ensure that customer problems are properly solved. However, the current complaint processing process highly depends on manual judgment, and customer service personnel need to understand customer emotions and assess the problem urgency in a short time, which leads to high work pressure and low efficiency, and further affects customer satisfaction.

[0003] In related technologies, sentiment analysis technology is used to assist complaint processing, but the traditional model cannot realize dynamic recognition and quantification of emotions in the processing process, which leads to poor emotion recognition effect; and further affects the efficiency of complaint processing and customer satisfaction. SUMMARY

[0004] The main purpose of the embodiments of the present application is to provide an efficient complaint processing method and system based on emotion intensity quantification, an electronic device and a storage medium.

[0005] To achieve the above purpose, one aspect of the embodiments of the present application provides a complaint processing method based on emotion intensity quantification, which comprises: obtaining complaint information of a target object; inputting the complaint information into an emotion perception model to perceive contextual emotion and obtain target emotion; the emotion perception model is established based on a spatio-temporal emotion gated memory module; quantifying the emotion intensity of the target object based on an industry emotion dictionary and the target emotion to obtain a target score of the target emotion, determining the business priority of the target object according to the target score, and processing the target business related to the complaint information according to the business priority. The emotion perception model based on the spatio-temporal emotion gated memory module is used to perceive the contextual emotion, and further quantifies the emotion intensity, which is beneficial to improve the accuracy of emotion recognition; and further improves the efficiency of complaint processing and customer satisfaction.

[0006] In some embodiments, the emotion perception model provided by the embodiments of the present application comprises a BART model, and the BART model comprises a plurality of encoder layers, each encoder layer is provided with a spatio-temporal emotion gated memory module, and the inputting of the complaint information into the emotion perception model to perceive the contextual emotion and obtain the target emotion comprises:

[0007] extracting timestamp information in the complaint information;

[0008] According to the dialogue turn of the target object and the timestamp information, a time encoder is generated to add a time position code to the complaint information.

[0009] Through each encoder layer, the complaint information is subjected to emotion perception to obtain a target emotion.

[0010] In some embodiments, the method provided by the embodiments of the present application comprises:

[0011] The output feature of the previous layer is obtained as the first input feature of the current layer, and the first input feature is subjected to multi-head attention processing to obtain a first intermediate feature;

[0012] The first intermediate feature, the previous turn emotion memory and the context emotion information of the previous several turns are subjected to gate processing to obtain a first emotion memory of the current layer;

[0013] The first emotion memory and the first intermediate feature are subjected to splicing processing to obtain a first output feature of the current layer;

[0014] The first output feature is taken as the first input feature of the next layer, the next layer is taken as a new current layer, and the first input feature is subjected to multi-head attention processing to obtain a first intermediate feature, until the first output features of all layers are obtained, and the target emotion is determined according to the first output features of all layers.

[0015] In some embodiments, the method provided by the embodiments of the present application comprises:

[0016] The intermediate feature is subjected to input gate processing to obtain a first feature; wherein the first feature is used to represent new information added;

[0017] The intermediate feature is subjected to forgetting gate processing to obtain a second feature; wherein the second feature is used to represent the retention and forgetting of historical emotion memory;

[0018] The intermediate feature is subjected to time decay gate or spatial weight gate processing to obtain a third feature; wherein the third feature is used to represent local context emotion information;

[0019] The first feature, the second feature and the third feature are subjected to fusion processing to obtain an emotion memory.

[0020] In some embodiments, the method provided by the embodiments of the present application comprises:

[0021] construct an industry sentiment dictionary according to industry corpus information;

[0022] fuse the industry sentiment dictionary with a general sentiment dictionary to obtain a target dictionary;

[0023] weight adjust the original score in the general sentiment dictionary according to the target sentiment to obtain a target score of the target sentiment in the target dictionary.

[0024] In some embodiments, the method provided by the embodiments of the present application comprises:

[0025] obtain a co-occurrence coefficient, an influence coefficient, a first fusion coefficient and a second fusion coefficient; the co-occurrence coefficient is related to a target sentiment in an industry sentiment dictionary, and the influence coefficient is related to overall context sentiment information;

[0026] determine a first weight item according to the co-occurrence coefficient and the first fusion coefficient;

[0027] determine a second weight item according to the influence coefficient and the second fusion coefficient;

[0028] determine a target score according to the first weight item, the second weight item and the original score.

[0029] In some embodiments, the method provided by the embodiments of the present application further comprises:

[0030] determine a text score and a sentiment score according to comprehensive information; the comprehensive information comprises sentence perplexity, sarcasm detection confidence and context sentiment offset;

[0031] determine the first fusion coefficient and the second fusion coefficient according to the text score and the sentiment score.

[0032] To achieve the above-mentioned purpose, another aspect of the embodiments of the present application proposes a complaint processing system based on sentiment intensity quantification, which comprises:

[0033] an acquisition module configured to acquire complaint information of a target object;

[0034] a sentiment module configured to input the complaint information into a sentiment perception model to perceive a context sentiment and obtain a target sentiment; the sentiment perception model is established based on a space-time sentiment gate memory module;

[0035] a quantification module configured to quantize sentiment intensity of the target object based on an industry sentiment dictionary and the target sentiment to obtain a target score of the target sentiment;

[0036] The business processing module is configured to determine a business priority of the target object according to the target score, and process a target business related to the complaint information according to the business priority.

[0037] To achieve the above object, another aspect of the embodiments of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0038] To achieve the above object, another aspect of the embodiments of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above method.

[0039] The embodiments of the present application at least have the following beneficial effects: the method provided by the embodiments of the present application comprises: obtaining complaint information of a target object; inputting the complaint information into an emotion perception model to perceive a context emotion and obtain a target emotion; the emotion perception model is established based on a spatio-temporal emotion gated memory module; quantifying an emotion intensity of the target object based on an industry emotion dictionary and the target emotion to obtain a target score of the target emotion; determining a business priority of the target object according to the target score, and processing a target business related to the complaint information according to the business priority. The embodiments of the present application perceive the context emotion through the emotion perception model based on the spatio-temporal emotion gated memory module, further quantify the emotion intensity, which is beneficial to improving the accuracy of emotion recognition, and further improves the complaint processing efficiency and customer satisfaction. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of an embodiment of the complaint processing method based on emotion intensity quantification provided by the present application;

[0041] Figure 2 is a flowchart of an embodiment of the emotion perception process provided by the present application;

[0042] Figure 3 is a flowchart of another embodiment of the emotion perception process provided by the present application;

[0043] Figure 4 is a flowchart of an embodiment of the emotion quantification process provided by the present application;

[0044] Figure 5 is a flowchart of another embodiment of the complaint processing process based on emotion intensity quantification provided by the present application;

[0045] Figure 6 is a flowchart of an embodiment of the processing process of the emotion perception module provided by the present application;

[0046] Figure 7 is a flow chart of another embodiment of the processing procedure of the emotion perception module provided by the present application;

[0047] Figure 8 is a flow chart of another embodiment of the emotion quantification procedure provided by the present application;

[0048] Figure 9 is a structural schematic diagram of the complaint handling system based on emotion intensity quantification provided by the embodiments of the present application;

[0049] Figure 10 is a hardware structural schematic diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to make the objects, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. When the following description relates to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementation described in the following exemplary embodiments does not represent all the implementations consistent with the embodiments of the present application, but is only an example of devices and methods consistent with some aspects of the embodiments of the present application as described in the appended claims.

[0051] It can be understood that the terms "first", "second", and the like used in the present application can be used herein to describe various concepts, but unless specifically stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as the first information. Depending on the context, the word "if" as used herein can be interpreted as "when" or "when" or "in response to determining".

[0052] The terms "at least one", "multiple", "each", "any" and the like used in the present application include one, two or more than two, multiple includes two or more than two, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0054] Before the embodiments of the present application are described in detail, first, some of the terms and terminology involved in the embodiments of the present application are explained, and the terms and terminology involved in the embodiments of the present application are applicable to the following explanations.

[0055] Sentiment Analysis: Also known as Opinion Mining, is an important task in the field of natural language processing, which focuses on whether the sentiment in the text is positive, negative or neutral, or more specific sentiment categories.

[0056] Position Encoding: A technique in natural language processing that assigns unique positional information to words or tokens in a sequence. Usually uses a combination of sine and cosine functions to help the model better understand and process the word position relationship in sequence data.

[0057] Feature Extraction: Extract representative information from raw data for training and building machine learning models.

[0058] BART: A variant of the Transformer architecture designed to solve sequence-to-sequence tasks such as machine translation, text summarization, etc.; BART's pre-training tasks include Monte Carlo autoencoders and other auxiliary tasks.

[0059] VADER: A natural language processing tool for sentiment analysis. It is a rule-based sentiment analysis tool designed to determine the sentiment polarity (positive, negative or neutral) and sentiment intensity of text by scoring the sentiment words in the text.

[0060] Trainable Memory Unit: A dynamically parameterized storage module that automatically learns how to retain, forget or update emotional information in historical conversations through machine learning, and its core is a recurrent neural network with gate control (for example: LSTM).

[0061] Emotion Strength Quantification: A sentiment score calculation method that combines domain-specific dictionaries and contextual information. This method integrates VADER dictionaries and industry-specific sentiment word libraries, and introduces a context attention mechanism to dynamically adjust the weight of emotional words. By considering factors such as contextual ambiguity, sentiment shift, sarcasm detection, and emoji consistency, ESI can generate more contextually appropriate emotion intensity scores for multi-round conversation sentiment analysis and industry-sensitive sentiment recognition.

[0062] For communication service providers, daily face a large number of customer complaints involving network failures, billing problems, service quality, and other aspects. The customer service department needs to handle a large amount of telephone and online interaction information to ensure that customer problems are properly resolved. However, the current complaint handling process relies heavily on manual judgment, and customer service personnel need to understand customer emotions and assess the urgency of the problem within a short period of time, which leads to high work pressure, low efficiency, and ultimately affects customer satisfaction.

[0063] With the rapid development of 5G technology, the complexity of communication services has increased significantly, and the diversity and quantity of customer complaints have surged. Traditional manual classification and priority determination methods are difficult to cope with, especially when customer emotions fluctuate greatly or complaints are lengthy, customer service personnel have difficulty accurately identifying key information and quantifying the intensity of customer emotions. For example, angry or anxious customers may be misjudged as high priority due to strong emotional expression, while some complaints that imply negative emotions (such as sarcasm, disappointment) may be underestimated due to subtle expression. This lack of objective quantitative standards can lead to important complaints being delayed, further exacerbating customer dissatisfaction.

[0064] Currently, some companies are trying to use sentiment analysis technology to assist in complaint handling, but traditional methods (such as VADER based on a dictionary or the general pre-training model BART) have obvious limitations: (1) Lack of context awareness: VADER and other dictionary-based models rely only on static sentiment lexicons and cannot dynamically adjust emotion scores in conjunction with dialogue context, leading to misjudgment of complex expressions such as double negatives and sarcasm. (2) Inadequacy of single-modal analysis: Existing methods usually only analyze a single modality of text or speech, while customer emotions are often expressed through multiple modalities such as tone, word intensity, and contextual coherence, making it difficult for existing technology to fully capture. (3) Coarse quantification of emotion intensity: Most models only output positive / negative binary classification or simple scores, and cannot refine the hierarchy of emotion intensity (such as "anger" can be further divided into "discontent" and "fury"), making it difficult to support accurate priority ranking.

[0065] Therefore, there is an urgent need for a context-aware customer emotion intensity quantification method that combines the deep semantic understanding capabilities of BART with the fine-grained sentiment lexicon of VADER, dynamically adjusts emotion scores in conjunction with dialogue context, and outputs interpretable intensity levels to assist customer service systems in implementing intelligent classification of complaints and optimizing resource allocation.

[0066] Therefore, the embodiments of the present application provide a context-aware customer emotion intensity quantification scheme based on BART+VADER to solve the problem of customer complaint handling faced by communication service providers in the 5G era. Traditional sentiment analysis methods (such as using VADER or general pre-training models independently) cannot accurately quantify the intensity of customer emotions, and lack a deep understanding of contextual semantics, making it difficult for customer service personnel to efficiently identify high-priority complaints.

[0067] The emotion intensity quantification-based complaint processing method provided by the embodiments of the present application relates to the technical field of computers. The emotion intensity quantification-based complaint processing method provided by the embodiments of the present application can be applied to a terminal, can be applied to a server, and can also be software running in the terminal or the server. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a vehicle-mounted terminal, and the like, but is not limited thereto; the server end can be configured as a stand-alone physical server, can be configured as a server cluster or a distributed system formed by multiple physical servers, can be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and big data and artificial intelligence platforms, and the server can also be a node server in a blockchain network; and the software can be an application that implements the emotion intensity quantification-based complaint processing method, and the like, but is not limited to the above forms.

[0068] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0069] It should be noted that in each specific embodiment of the present application, when relevant processing needs to be performed according to user information, user behavior data, user historical data, and user location information, and the like related to the identity or characteristics of the user, the user's permission or consent will be obtained first, and the collection, use, and processing of the data will comply with relevant laws, regulations, and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for enabling the embodiments of the present application to normally operate will be obtained.

[0070] Figure 1 is an optional flowchart of the complaint processing method based on emotion intensity quantification provided by the embodiments of the present application; Figure 1 The method in the above embodiment can include, but is not limited to, steps S100 to S300.

[0071] Step S100, obtaining complaint information of a target object;

[0072] Step S200, inputting the complaint information into an emotion perception model to perceive a context emotion and obtain a target emotion; the emotion perception model is established based on a spatiotemporal emotion gating memory module;

[0073] Step S300, quantifying an emotion intensity of the target object based on an industry emotion dictionary and the target emotion to obtain a target score of the target emotion;

[0074] Step S400, determining a business priority of the target object according to the target score and processing a target business related to the complaint information according to the business priority.

[0075] The complaint information in the embodiments of the present application can be conversation information or complaint information; in addition, the complaint information can be text information or voice information, and the present application does not limit the form and path of the complaint information. The target object can be a customer, and the complaint information can be conversation information of the customer and a customer service. The spatiotemporal emotion gating memory module can dynamically perceive emotion changes at different time points and in different context spaces. The present application quantifies the emotion of the target object through the target emotion and the target score. In some embodiments, if it is necessary to determine the priority and process the complaint of a large number of customers, each customer is taken as a target object, and the target score of each customer is obtained through the method provided by the present application, and then all scores are obtained. After sorting the scores, the priority of the customer is determined according to the sorting result, and the complaint processing is performed according to the priority. In some embodiments, referring to Figure 2 The method provided by the embodiments of the present application includes a BART model, the BART model includes a plurality of encoder layers, each encoder layer is provided with a spatiotemporal emotion gating memory module, the complaint information is input into the emotion perception model to perceive the context emotion and obtain the target emotion, which includes:

[0076] Step S210, extracting timestamp information in the complaint information;

[0077] Step S220, generating a time encoder according to the conversation round and the timestamp information of the target object, and adding time position encoding to the complaint information;

[0078] Step S230, perceiving the emotion of the complaint information through each encoder layer to obtain the target emotion.

[0079] Of course, it can be understood that the emotion perception model in the present application can be any artificial intelligence model, and the present application does not limit the specific algorithm of the emotion perception model. For example, the emotion perception model can be implemented by an LSTM algorithm or a BERT algorithm. The present application will be described in detail by taking a BART model as an example. Each encoder layer of the BART model is provided with a spatio-temporal emotion gated memory module to perceive the spatio-temporal context.

[0080] In some embodiments, with reference to Figure 3 The method provided by the embodiments of the present application performs emotion perception on the complaint information through each encoder layer to obtain a target emotion, including:

[0081] In step S231, the output feature of the previous layer is obtained as the first input feature of the current layer, and the first input feature is subjected to multi-head attention processing to obtain a first intermediate feature;

[0082] In step S232, the first intermediate feature, the previous round emotion memory, and the context emotion information of the previous several rounds are subjected to gating processing to obtain a first emotion memory of the current layer.

[0083] In step S233, the first emotion memory and the first intermediate feature are subjected to concatenation processing to obtain a first output feature of the current layer.

[0084] In step S234, the first output feature is taken as the first input feature of the next layer, the next layer is taken as a new current layer, and the first input feature is subjected to multi-head attention processing to obtain a first intermediate feature, until the first output features of all layers are obtained, and the target emotion is determined according to the first output features of all layers.

[0085] The number of rounds of the context emotion information of the previous several rounds in the present application can be adjusted according to actual needs.

[0086] In some embodiments, the method provided by the embodiments of the present application, the previous round emotion memory is determined by the following steps:

[0087] The intermediate feature is subjected to input gate processing to obtain a first feature; wherein the first feature is used to represent new information added;

[0088] The intermediate feature is subjected to forgetting gate processing to obtain a second feature; wherein the second feature is used to represent the retention and forgetting of historical emotion memory.

[0089] The intermediate feature is subjected to time decay gate or spatial weight gate processing to obtain a third feature; wherein the third feature is used to represent local context emotion information.

[0090] The first feature, the second feature, and the third feature are subjected to fusion processing to obtain an emotion memory.

[0091] The spatio-temporal emotion gate memory module in the application includes an input gate, a forgetting gate, an output gate, a time decay gate, and a spatial weight gate.

[0092] In some embodiments, the method provided by the application provides Figure 4 , quantifying the emotion intensity of the target object based on the industry emotion dictionary and the target emotion to obtain a target score of the target emotion, including:

[0093] Step S310, constructing an industry emotion dictionary according to industry corpus information;

[0094] Step S320, fusing the industry emotion dictionary and the general emotion dictionary to obtain a target dictionary;

[0095] Step S330, adjusting the weight according to the original score of the target emotion in the general emotion dictionary to obtain a target score of the target emotion in the target dictionary.

[0096] The target emotion in the application may exist in the industry emotion dictionary, may exist in the general emotion dictionary, or may exist in the industry emotion dictionary and the general emotion dictionary.

[0097] In some embodiments, the method provided by the application adjusts the weight of the original score in the general emotion dictionary according to the target emotion to obtain a target score of the target emotion in the target dictionary, including:

[0098] Obtaining a co-occurrence coefficient, an influence coefficient, a first fusion coefficient, and a second fusion coefficient; wherein the co-occurrence coefficient is related to the target emotion in the industry emotion dictionary, and the influence coefficient is related to the overall context emotion information;

[0099] Determining a first weight term according to the co-occurrence coefficient and the first fusion coefficient;

[0100] Determining a second weight term according to the influence coefficient and the second fusion coefficient;

[0101] Determining the target score according to the first weight term, the second weight term, and the original score.

[0102] In some embodiments, the method provided by the application further includes:

[0103] Determining a text score and an emotion score according to the comprehensive information; wherein the comprehensive information includes sentence perplexity, sarcasm detection confidence, and context emotion offset;

[0104] Determining the first fusion coefficient and the second fusion coefficient according to the text score and the emotion score.

[0105] Below, combined with specific application examples, the scheme of the embodiment of the present application is described and explained in detail:

[0106] Referring to Figure 5 , according to the emotion quantification process provided by the present application, the core purposes of the present application include:

[0107] 1: Realize the emotion analysis of up and down perception: by fusing the deep semantic understanding ability of BART, the fine-grained sentiment dictionary of VADER and the multi-head attention mechanism, accurately identify:

[0108] (1): Emotion evolution across multiple rounds of dialogue (such as upgrading from "complaint" to "anger").

[0109] (2): Industry-specific expressions (such as "5G signal is poor" has a 40% stronger emotion than "slow network speed").

[0110] (3): Implicit emotional expression recognition (irony / double negative, etc.).

[0111] 2: Improve customer service efficiency: by introducing BART and VADER technology, aiming to achieve more accurate speech emotion analysis to help customer service team more effectively locate problems and improve problem handling speed.

[0112] 3: Build a quantifiable emotion strength index system: create Emotion Strength Index (ESI) quantification standard to solve the rough classification problem of traditional method "non-negative is positive".

[0113] 4: Integrate BART+VADER technology: aims to make full use of BART technology to achieve more advanced natural language processing and understanding, thereby enhancing the system's ability to analyze and handle customer complaints; use VADER sentiment analysis technology to perform more accurate sentiment analysis on customer speech. This helps the customer service team better understand the emotional state of customers and improve service quality.

[0114] The method uses VADER emotion analysis, BART (Seq2Seq), and embeds trainable memory cells (Gated Memory Cell) in the BART Encoder layer. Build a context-aware emotion strength quantification system, the specific core technology is as follows:

[0115] Contextual spatiotemporal emotion gate memory cell embedding BART Encoder layer process:

[0116] The application proposes a spatio-temporal emotion gating memory module (ST-EGM), which is embedded into each layer of the BART encoder to realize deep modeling of the dynamic evolution of emotions in multi-turn dialogues. The module structure is inspired by the long short-term memory network (LSTM) and integrates the multi-head self-attention mechanism of the Transformer, with the ability to model both time and space.

[0117] After each encoding layer, a set of trainable spatio-temporal emotion gating memory module units is inserted. Each spatio-temporal emotion gating memory module includes an input gate, a forget gate, an output gate, a newly added temporal decay gate, and a spatial attention gate to dynamically perceive the changes in emotions at different time points and contextual spaces. The model receives the current dialogue content and its timestamp sequence, and the spatio-temporal emotion gating memory unit synchronously maintains the historical emotion state vector and the spatial context vector. The temporal decay gate dynamically adjusts the importance of historical emotions based on time intervals, while the spatial weight gate dynamically filters effective memories based on the emotional relevance strength of local dialogue context.

[0118] In the memory update part, the formula m_t = (forget_gate * m t-1 )+(input_gate * new_info) + (spatial_gate * spatial_context) is used. Where: forget_gate * m t-1 (Second feature): control the retention and forgetting of historical emotion memory; input_gate * new_info (first feature): input gate determines how much new information to add; spatial_gate * spatial_context (third feature): dynamic fusion of local context emotional information. In this unit, the introduction of the spatio-temporal emotion gating memory module enables the BART encoder to dynamically model the evolution of emotions in the time and space dimensions, significantly improving the accuracy and robustness of complex emotional trajectory modeling in multi-turn dialogues, thereby providing a foundation for context-aware emotion intensity quantification.

[0119] Referring to Figure 6 and Figure 7 , the implementation process of BART+ spatio-temporal emotion gating memory module (ST-EGM) for processing complaint text sequences is as follows:

[0120] 1. Text input and timestamp labeling

[0121] Receive the complaint text X t of the current turn and extract its corresponding timestamp tst The text is first converted into initial word vector representation E t ∈R n×d where n is the length of the text and d is the embedding dimension.

[0122] 2. Introducing Temporal Positional Embedding

[0123] To enhance the modeling ability of the time interval of the dialogue turn, based on the time difference between the current and the last turn Δt = ts t -ts t-1 Add time-driven position embedding to the word vector:

[0124] E new t = E t + PositionEmbed(Δt)

[0125] This mechanism can effectively guide the model to pay attention to the characteristics of long-time silence or delayed emotional outburst.

[0126] 3. Layer-by-layer modeling and embedding ST-EGM units in the encoder layer

[0127] In each layer of the BART encoder, the text representation will be processed in the following order:

[0128] A: Standard Transformer calculation: the current layer first performs the conventional multi-head self-attention calculation and feed-forward network processing to generate the intermediate representation

[0129] B: ST-EGM processing module:

[0130] For each layer output Jointly input the last turn emotional memory And the context representation of the last k turns Emotion gate processing:

[0131] Time decay gate controls the influence of historical memory over time to decay;

[0132] The spatial attention gate identifies the emotional clues of the highly correlated context;

[0133] Calculate and update the emotional memory state of the current layer

[0134] Concatenate With To form the next layer input.

[0135] 4. Cross-turn memory state maintenance

[0136] Emotion memory state of the top layer of the encoder The storage in the global emotion memory pool provides historical support for subsequent rounds. When processing the next round of dialogue X t+1 , the model can recall the memory state of the last k rounds for emotion association modeling, improving the ability to perceive the continuity and progressive change of emotions.

[0137] Referring Figure 8 to the drawings, the present application provides emotion strength quantification (Emotion Strength Index, ESI)

[0138] 1. Construction of industry field emotion dictionary

[0139] This module mines and constructs emotion words based on industry-specific corpus (such as communication customer service dialogue data). First, TF-IDF, emotion label weak supervision labeling is used to automatically identify high-frequency emotion-related words (such as “network interruption”, “line drop”, “poor customer service”), and assign each word item with an initial emotion polarity label (positive, negative, neutral) and an initial strength score (range between -1 and +1).

[0140] 2. Dictionary fusion module

[0141] This module is responsible for fusing the industry emotion dictionary constructed above with the original general emotion dictionary of VADER. The fusion strategy adopts an incremental embedding method, which retains the original structure of VADER while adding industry-specific emotion entries and covering their default emotion scores. At the same time, for overlapping words, the strength is adjusted through a context co-occurrence scoring mechanism to improve context adaptability.

[0142] 3. Emotion weight adjustment

[0143] This module dynamically calibrates the score of each emotion word in the fused dictionary. Specifically, the following correction formula is used:

[0144] S final = S vader + α * f domain (w) + β * f context (w)

[0145] Where:

[0146] S final represents the final corrected emotion strength (target score);

[0147] S vader represents the original VADER score (original score);

[0148] f domain (w) is the negative co-occurrence coefficient of the word item in the industry corpus (co-occurrence coefficient);

[0149] f context (w) is the context influence factor module output (influence coefficient);

[0150] α,β are experience setting or trainable weight parameters, when setting the emotion fusion parameters α (i.e. the first fusion coefficient in this application) and β (the second fusion coefficient), the application introduces a context-aware attention weighting network (CAW) based on attention mechanism; A_text (text score) and A_emoji (emotion score) are obtained by scoring according to the context features through the attention weighting network, and then the fusion coefficients α and β are calculated by using the softmax function:

[0151]

[0152] where A text ,A emoji are scored based on context features respectively, and the features include but are not limited to current sentence perplexity, sarcasm detection confidence, context emotion offset ΔS, emoji emotion consistency matching degree, etc.

[0153] The new emotion dictionary output by this module will be used to replace or enhance the VADER original word library, so that the system can adaptively adjust the emotion recognition result according to the context and industry attributes, and can be used in subsequent multi-round dialogue analysis, user emotion evolution tracking and industry sensitive emotion monitoring tasks, etc., further improving the adaptability of the system in cross-domain scenarios.

[0154] Referring to the specific embodiments shown in Figure 5 The method provided by the application comprises the following steps:

[0155] 1. Data collection and preprocessing: First, the customer complaint text data is extracted from the A City B Department- Customer Service Digitalization Application Development Project. The data needs to be preprocessed, which aims to optimize the data quality and ensure the integrity and availability of the data. Data cleaning, handling of missing values, and supplementing operations are performed to ensure the accuracy and reliability of the data set.

[0156] 2. Introduction of BART model: The BART model is introduced, which uses Monte Carlo autoencoder and autoregressive decoder to process sequence data. Specifically, the Monte Carlo autoencoder is trained by introducing random noise and Monte Carlo sampling to improve the robustness and generalization ability of the model. The autoregressive decoder is used to generate output sequences related to the input sequences. This can effectively handle text generation tasks.

[0157] 3. BART introduces spatio-temporal emotion gate memory unit + VADER analysis: the application innovatively embeds a spatio-temporal emotion gate memory module in the Encoder layer of BART, which combines the gating mechanism of LSTM with the multi-head attention structure of Transformer, and has the ability of temporal dynamic modeling and context spatial perception. The spatio-temporal emotion gate memory unit is composed of input gate, forget gate, output gate, time decay gate and spatial attention gate, which can adjust the emotion memory state according to the dialogue time interval and semantic dependence, and realize the adaptive update of historical emotion vector and context features. This module significantly improves the model's ability to perceive complex emotional evolution in cross-turn dialogues (such as irony, emotional turning point, double negative, etc.), providing context-enhanced support for subsequent VADER analysis, thereby improving the accuracy and robustness of emotion recognition.

[0158] 4. Emotion intensity quantification

[0159] Emotion intensity quantification includes three parts: industry domain emotion vocabulary construction, dictionary fusion and weight adjustment. First, the industry emotion dictionary construction adopts TF-IDF and weakly supervised emotion labeling to extract high-frequency emotion words from industry corpus and assign initial emotion polarity and intensity values. Second, the dictionary fusion module combines industry-specific dictionaries and VADER original dictionaries to dynamically correct the emotion scores of the words through semantic co-occurrence and context relationship, enhancing the domain adaptability. Finally, the emotion weight adjustment module introduces a context weight adjustment network (CAW) based on attention mechanism, which considers features such as irony confidence, confusion, emotion shift and emoji consistency, and uses softmax to calculate the fusion weights α and β to dynamically correct the emotion items in the dictionary. The final output emotion intensity quantification score interval is [-1, 1], and is subdivided into multiple emotion levels (such as anger, anger, dissatisfaction, neutral, satisfaction, joy, etc.), achieving precise quantification and dynamic modeling of user emotion intensity.

[0160] Traditional text processing methods rely on static rules and keyword matching to identify customer emotions, such as searching for fixed words like "problem", "urgent", "complaint" to infer the emotional tendency of the text. Such methods are easily affected by expression changes, lack of context information, etc., resulting in low accuracy of emotion recognition, and are difficult to handle complex scenarios such as irony, double negative, multi-turn dialogue, etc.

[0161] The application realizes deep understanding of the evolution of text emotions by introducing the BART model and its context perception ability. The BART model uses a joint pre-training strategy of Monte Carlo autoencoder and autoregressive decoder, which not only has superior semantic encoding ability, but also can generate natural and fluent reconstructed text according to the context, thereby revealing hidden emotional information (such as implicit complaints, implicit dissatisfaction, etc.).

[0162] On this basis, the application further innovatively embeds a context memorability unit (Gated Memory Cell) in the BART encoder layer, specifically adopts a spatiotemporal emotion gated memory unit, dynamically models the evolution of the dialogue history emotion through a time decay mechanism and a space attention mechanism, enables the system to have the modeling capability of long-time dependence on emotion and context association, and significantly improves the detection accuracy of cross-turn dialogue and implicit emotion changes. Meanwhile, combined with an industry-specific emotion lexicon and a context influence factor (CSM), the emotion polarity and intensity score are dynamically optimized, and the recognition capability of the system for complex emotional expressions (such as irony, exaggeration, and escalation of dissatisfaction) is further enhanced.

[0163] Finally, by constructing an emotion intensity quantification system, multi-level quantification of customer emotional expression is realized, not only the emotion polarity (positive / negative / neutral) can be judged, but also different intensity levels (such as slight dissatisfaction, strong anger, and extreme rage) can be distinguished in detail, which provides accurate basis for subsequent complaint priority sorting and resource scheduling.

[0164] The application provides a BART text generation scheme based on context perception: the scheme introduces a trainable spatiotemporal emotion gated memory unit in the BART model, realizes the time sequence modeling and context perception of the evolution of emotion in multi-turn dialogue, significantly enhances the understanding capability of the model for complex dialogue context, effectively improves the emotion consistency and expression accuracy of the generated text, and provides key support for emotion reconstruction and accurate context response.

[0165] The application provides an application of specific industry VADER sentiment analysis: the context influence factor (CSM) is introduced to dynamically correct the original emotion score of VADER, and the accurate identification of industry-specific negative emotions and implicit emotions is realized.

[0166] The application provides an emotion intensity quantification and complaint priority sorting scheme: an ESI quantification system is designed, an industry emotion dictionary is constructed, emotion expressions are refined and graded according to intensity, and the urgency of complaint content is sorted according to the quantification results, which effectively improves the intelligent level of customer problem handling and the efficiency of resource scheduling.

[0167] Please refer to Figure 9 The application embodiment further provides a complaint handling system based on emotion intensity quantification, which can realize the above-mentioned complaint handling method based on emotion intensity quantification. The system comprises:

[0168] The acquisition module 510 is configured to acquire complaint information of a target object.

[0169] The emotion module 520 is configured to input the complaint information into an emotion perception model, perceive a context emotion, and obtain a target emotion. The emotion perception model is established based on a spatiotemporal emotion gated memory module.

[0170] quantify, according to an industry sentiment dictionary and a target sentiment, an emotional intensity of a target object, to obtain a target score of the target sentiment;

[0171] a business processing module 540, configured to determine a business priority of the target object according to the target score, and process a target business related to the complaint information according to the business priority.

[0172] It can be understood that the content in the above method embodiments is applicable to the present system embodiments, the present system embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0173] The present application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above complaint processing method based on emotional intensity quantification when executing the computer program. The electronic device can be any intelligent terminal, such as a tablet computer or a vehicle-mounted computer.

[0174] It can be understood that the content in the above method embodiments is applicable to the present device embodiments, the present device embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0175] Please refer to Figure 10 , Figure 10 a hardware structure of an electronic device of another embodiment is shown, which includes:

[0176] The processor 901 can be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is used to execute related programs to implement the technical solutions provided by the present application;

[0177] The memory 902 can be implemented in the form of a ROM (ReadOnly Memory), a static storage device, a dynamic storage device, or a RAM (Random Access Memory). The memory 902 can store an operating system and other application programs. When the technical solutions provided by the present application are implemented by software or firmware, the related program codes are stored in the memory 902 and are called and executed by the processor 901 to implement the complaint processing method based on emotional intensity quantification of the present application;

[0178] The input / output interface 903 is configured to realize information input and output.

[0179] The communication interface 904 is configured to realize communication interaction between the device and other devices, and the communication can be realized in a wired manner (for example, a USB, a network cable, and the like) or in a wireless manner (for example, a mobile network, WIFI, Bluetooth, and the like).

[0180] The bus 905 is configured to transmit information between various components (for example, the processor 901, the memory 902, the input / output interface 903, and the communication interface 904) of the device.

[0181] The processor 901, the memory 902, the input / output interface 903, and the communication interface 904 are communicatively connected to each other in the device through the bus 905.

[0182] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the above-mentioned complaint processing method based on emotion intensity quantification.

[0183] It can be understood that the contents in the above-mentioned method embodiments are all applicable to the present storage medium embodiment, the function specifically realized by the present storage medium embodiment is the same as that of the above-mentioned method embodiments, and the beneficial effects achieved by the present storage medium embodiment are also the same as those achieved by the above-mentioned method embodiments.

[0184] The memory is a non-transitory computer readable storage medium, and can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory, and can also include a non-transitory memory, for example, at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and the remote memory can be connected to the processor through a network. Examples of the network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0185] The embodiments described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0186] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than those shown in the figures, or combine certain steps or different steps.

[0187] The apparatus embodiments described above are merely exemplary, and the units described as separate units can or can not be physically separate, i.e., can be located in one place, or can be distributed over multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purposes of the embodiments.

[0188] Those skilled in the art can understand that all or some of the steps in the method disclosed above, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and appropriate combinations thereof.

[0189] The terms "first", "second", "third", "fourth" and the like in the description of the application and in the claims of the foregoing drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so construed can be interchanged, such that the embodiments of the application described herein can be carried out in other than the order discussed herein without departing from the scope of the application. Further, the terms "comprise" and "comprising" and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product or apparatus that comprises a list of steps or units does not necessarily comprise only those steps or units but can include other not expressly listed steps or units.

[0190] It should be understood that in the present application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the relationship between associated objects, which means that there can be three relationships, for example, "A and / or B" can mean that there are three cases: only A, only B, and A and B at the same time, where A and B can be singular or plural. The character " / " generally represents that the associated objects before and after are in an "or" relationship. "At least one of the following" or the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can mean a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0191] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented by other manners. For example, the apparatus embodiments described above are merely illustrative, for example, the division of the above units is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, apparatuses or units, and can be electrical, mechanical or other forms.

[0192] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0193] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0194] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that makes a contribution or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0195] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, but this does not limit the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A customer complaint handling method based on the quantification of emotional intensity, characterized in that, The method includes: Obtain customer complaint information from the target group; The customer complaint information is input into the emotion perception model to perceive the contextual emotion and obtain the target emotion; the emotion perception model is established based on the spatiotemporal emotion gating memory module. Based on the industry sentiment dictionary and the target sentiment, the intensity of the target object's sentiment is quantified to obtain a target score for the target sentiment; Based on the target score, the business priority of the target object is determined, and the target business related to the customer complaint information is processed according to the business priority.

2. The method according to claim 1, characterized in that, The emotion perception model includes a BART model, which comprises several encoder layers. Each encoder layer is equipped with a spatiotemporal emotion gating memory module. The process of inputting the customer complaint information into the emotion perception model, perceiving the contextual emotion, and obtaining the target emotion includes: Extract the timestamp information from the customer complaint information; Based on the dialogue turn of the target object and the timestamp information, a time encoder is generated, and the time location code is added to the customer complaint information; By performing emotion perception on the customer complaint information through each encoder layer, the target emotion is obtained.

3. The method according to claim 2, characterized in that, The process of performing emotion perception on the customer complaint information through each encoder layer to obtain the target emotion includes: The output feature of the previous layer is obtained as the first input feature of the current layer, and multi-head attention processing is performed on the first input feature to obtain the first intermediate feature; Gating processing is performed on the first intermediate feature, the emotional memory of the previous round, and the contextual emotional information of several previous rounds to obtain the first emotional memory of the current layer. The first emotional memory and the first intermediate feature are concatenated to obtain the first output feature of the current layer; The first output feature is used as the first input feature of the next layer, and the next layer is used as the new current layer. The process of multi-head attention processing on the first input feature is repeated to obtain the first intermediate feature until the first output features of all layers are obtained. Based on the first output features of all layers, the target emotion is determined.

4. The method according to claim 3, characterized in that, The previous round of emotional memories is determined through the following steps: The intermediate features are processed through an input gate to obtain the first feature; wherein the first feature is used to represent the newly added information. The intermediate features are processed through a forgetting gate to obtain the second feature; the second feature is used to characterize the retention and forgetting of historical emotional memories. The intermediate features are processed by a time decay gate or a spatial weight gate to obtain a third feature; wherein, the third feature is used to characterize local contextual sentiment information; The first feature, the second feature, and the third feature are fused together to obtain emotional memory.

5. The method according to claim 1, characterized in that, The process of quantifying the emotional intensity of the target object based on the industry sentiment dictionary and the target sentiment to obtain a target sentiment score includes: Construct an industry sentiment dictionary based on industry corpus information; The industry-specific sentiment dictionary is then merged with a general sentiment dictionary to obtain the target dictionary; The target score of the target emotion in the target emotion dictionary is obtained by weighting the target emotion according to the original score of the target emotion in the general emotion dictionary.

6. The method according to claim 5, characterized in that, The step of adjusting the weights of the original scores in the general emotion dictionary based on the target emotion to obtain the target score of the target emotion in the target dictionary includes: Obtain the co-occurrence coefficient, influence coefficient, first fusion coefficient, and second fusion coefficient; wherein, the co-occurrence coefficient is related to the target emotion in the industry sentiment dictionary, and the influence coefficient is related to the overall contextual sentiment information; The first weighting term is determined based on the co-occurrence coefficient and the first fusion coefficient; The second weighting term is determined based on the influence coefficient and the second fusion coefficient; The target score is determined based on the first weighting term, the second weighting term, and the original score.

7. The method according to claim 6, characterized in that, The method further includes: Based on comprehensive information, text scores and sentiment scores are determined; wherein, the comprehensive information includes sentence confusion level, irony detection confidence level, and context sentiment shift. Based on the text score and the emotion score, a first fusion coefficient and a second fusion coefficient are determined.

8. A customer complaint handling system based on the quantification of emotional intensity, characterized in that, The system includes: The acquisition module is used to acquire customer complaint information of the target object; The emotion module is used to input the customer complaint information into the emotion perception model, perceive the contextual emotion, and obtain the target emotion; the emotion perception model is established based on the spatiotemporal emotion gating memory module. The quantification module is used to quantify the emotional intensity of the target object based on the industry sentiment dictionary and the target sentiment, and obtain the target score of the target sentiment; The business processing module is used to determine the business priority of the target object based on the target score, and process the target business related to the customer complaint information according to the business priority.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

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

Images