Complaint event processing timeliness demand prediction method, device and medium

Abstract

The invention relates to a complaint event processing timeliness demand prediction method and device, and a medium. The method comprises the following steps: collecting complaint event text data, carrying out word segmentation, and extracting emotional feature words and situational features in the complaint event text data; constructing an emotion feature word bank and setting a three-dimensional score of each emotion feature word in the emotion feature word bank; constructing a corresponding three-dimensional emotion coordinate and an emotion interaction matrix for each emotion feature word in the complaint event text data based on an emotion feature word library; calculating an overall emotional state trend of the complaint event based on the three-dimensional emotional coordinates and an emotional interaction matrix; calculating the narrative tension of the complaint event according to the overall emotional state trend and the situation characteristics of the complaint event; constructing a user emotion prediction model, and predicting the future emotion trend of the user corresponding to the complaint event; and based on the narrative tension of the complaint event and the emotional future trend of the user, calculating the processing timeliness demand of the current complaint event.

Description

Technical Field

[0001] The present invention relates to a method, device and medium for predicting timeliness demand for complaint event processing, and belongs to the technical field of complaint management. Background Art

[0002] With the rapid development of Internet technology and customer service business, users' expectations for service quality are constantly increasing. How to quickly and accurately determine the urgency and priority of complaint incidents has become one of the important challenges for enterprises to improve customer satisfaction and service levels.

[0003] Traditional methods for predicting the timeliness of complaint handling generally rely on manual experience or simple rule judgments, which lack objectivity and accuracy. At the same time, although some existing technologies try to use simple sentiment analysis models, they usually ignore the multidimensionality of complaint text emotions, the dynamic characteristics of sentiment changes, and the interactive effects between emotions, resulting in the prediction results being difficult to fully reflect the user's true emotional state and its trend changes. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes a method, device and medium for predicting the timeliness demand for complaint event processing.

[0005] The technical solution of the present invention is as follows: In one aspect, the present invention provides a method for predicting the timeliness demand for complaint event processing, comprising the following steps: Collect complaint event text data, segment the complaint event text data through a pre-trained large model, and extract sentiment feature words and situational features; Construct an emotional feature word library and set a three-dimensional score for each emotional feature word in the emotional feature word library, including an emotional intensity score, an emotional valence score, and an emotional control score; Based on the emotional feature word library, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed for each emotional feature word in the complaint event text data; Based on each emotional feature word, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed to calculate the overall emotional state trend of the complaint event; Calculate the narrative tension of the complaint event based on the overall emotional state trend and situational characteristics of the complaint event; A user emotion prediction model is constructed based on an improved bidirectional long short-term memory network, and the future trend of the emotion of the user corresponding to the complaint event is predicted through the overall emotional state trend of the complaint event. The improved bidirectional long short-term memory network includes a positive and negative long short-term memory network layer and an attention fusion layer; The time efficiency requirement for handling the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions.

[0006] Preferably, the corresponding three-dimensional emotional coordinates are constructed for each emotional feature word in the complaint event text data based on the emotional feature word library as shown in the following formula: ; in: Indicates the first The three-dimensional emotional coordinates corresponding to the emotional feature words; Indicates The emotional intensity score of each emotional feature word; Indicates The emotional valence score of the emotional feature words; Indicates The emotional control score of the emotional characteristic words; Based on the emotional feature word library, the corresponding emotional interaction matrix is ​​constructed for each emotional feature word in the complaint event text data as shown in the following formula: ; in: Indicates The emotional interaction matrix of emotional feature words; Indicates the natural attenuation coefficient of emotion intensity; represents the natural attenuation coefficient of emotional valence; It represents the natural attenuation coefficient of emotional control; Indicates sentiment feature words Dimension ratings and The interaction parameters between dimension ratings, , , .

[0007] Preferably, the specific steps of constructing the corresponding three-dimensional emotional coordinates and emotional interaction matrix based on each emotional feature word and calculating the emotional intensity curve, emotional valence curve and emotional control curve of the complaint event are as follows: According to the order of occurrence of emotional feature words in the complaint event text data, the emotional state vector is constructed based on the three-dimensional emotional coordinates of each emotional feature word. ,in, Indicates the first The three-dimensional emotional coordinates of the emotional feature words that appear, The three-dimensional sentiment coordinates representing the last sentiment feature word that appears in the complaint event text data; At the same time, according to the order of appearance of emotional feature words in the complaint event text data, the emotional interaction matrix vector is constructed based on the emotional interaction matrix of each emotional feature word. ,in, Indicates the first The emotional interaction matrix of the emotional feature words that appear, The sentiment interaction matrix representing the last sentiment feature word that appears in the complaint event text data; The emotional dynamics equation is constructed based on the emotional state vector and the emotional interaction matrix vector, as shown in the following formula: ; ; in: Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; Indicates The starting time of the occurrence of the sentiment feature word, that is, the position of the sentiment feature word in the complaint event text data; Indicates The ending moment of the first appearing emotional feature word, that is, The position of the emotional feature words in the complaint event text data; Indicates The external emotional impact of the emotional feature words that appear; Indicates The weight of the sentiment feature words that appear; represents a natural constant; For each time interval corresponding to the occurrence of the emotional feature word, the final state of the previous time interval is used as the initial state of the next interval. The above formula is independently solved using the RK4 method to obtain the emotional state vector trend of each time interval. The emotional state vector trends of each time interval are then spliced ​​together to obtain the overall emotional state trend.

[0008] Preferably, the calculation formula of the narrative tension of the complaint event is specifically shown as follows: ; ; ; in: Indicates the narrative tension of the complaint incident; Indicates the first The emotional feature words that appear in the corresponding time interval Weighted probability distribution of emotional states within; Indicates the first The emotional feature words that appear in the corresponding time interval The emotional entropy within; represents the smoothing coefficient; represents the emotional self-sensitivity coefficient; Indicates the sensitivity of the situational features of the complaint event; Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; , , Respectively represent The emotional feature words that appear in the corresponding time interval The trends of emotion intensity scores, emotion valence scores, and emotion control scores.

[0009] Preferably, the forward and reverse long short-term memory network layer is specifically shown in the following formula: ; ; ; in: Indicates the forward and reverse long short-term memory network layers in Output results at the moment; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; Indicates the forward and reverse long short-term memory network layers in Input of time; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; represents the long short-term memory network; Express and To splice; Each unit of the forward and reverse long short-term memory networks is provided with a dynamic emotion enhancement gate, as shown in the following formula: ; ; ; ; ; ; ; in: , , Represents the forget gate, input gate, and output gate of the forward and reverse long short-term memory network units respectively. Output at the moment; , , Represent the weight matrices of the forget gate, input gate, and output gate respectively; Indicates the forward and reverse long short-term memory network layers in Output results at the moment; , , Represent the bias vectors of the forget gate, input gate, and output gate respectively; Dynamic emotion enhancement gates representing forward and reverse LSTM network units Output at the moment; represents the weight matrix of the dynamic emotion enhancement gate; Represents the attention fusion layer Multi-scale attention fusion features output at every moment; represents the bias vector of the dynamic emotion enhancement gate; Represents the forward and reverse long short-term memory network units in moment-to-moment enhanced emotional candidate states; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; The weight matrix representing the cell states of the forward and reverse LSTM networks; Bias vectors representing the cell states of forward and reverse LSTM networks; Represents the forward and reverse long short-term memory networks in The hidden state of the moment; represents the Hadamard product operation; Represents the sigmoid function.

[0010] Preferably, the attention fusion layer includes a multi-scale convolution extraction unit, a scale interaction unit, a scale selection unit and a multi-scale feature fusion unit; The multi-scale convolution extraction unit extracts features of different scales from the outputs of the forward and reverse long short-term memory network layers through convolution layers with convolution kernels of different sizes, as shown in the following formula: ; in: express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale is Convolution operation; express Time to The output sequence of the forward and reverse long short-term memory network layer at the moment; Define the interaction weight matrix, and set the size of the interaction weight matrix according to the number of scales. The interaction weight matrix is ​​specifically shown in the following formula: ; ; ; in: express The interaction weight matrix at the moment; express function; express The query matrix at the moment; express The key matrix at the moment; a matrix of learnable parameters representing the query matrix; a learnable parameter matrix representing the bond matrix; Indicates the preset feature dimension; Represents a transpose operation; The scale interaction unit updates the output features of the forward and reverse long short-term memory network layers at each scale based on the interaction weight matrix, as shown in the following formula: ; in: Indicates updated The time scale is The output features of the forward and reverse long short-term memory network layers; express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale selection unit is used to calculate the contribution of different scales, as shown in the following formula: ; in: Indicates that the scale selection unit is Time output scale Contribution of represents the learnable parameter matrix of the scale selection unit; represents the bias vector of the scale selection unit; The final output features of the forward and reverse long short-term memory network layers at different scales are calculated based on the contribution of different scales, as shown in the following formula: ; The multi-scale feature fusion unit is used to calculate the final output features of the forward and reverse long short-term memory network layers at different scales by the contribution of different scales, as described in the following formula: ; ; in: Representation scale exist The position adaptive weight at the moment; represents the attention query vector; a learnable parameter matrix representing the position adaptation weights; Bias vector representing the position adaptation weights.

[0011] Preferably, the time efficiency requirement for handling the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions, as shown in the following formula: ; in: Indicates the time limit required for handling the current complaint event; It represents the timeliness demand sensitivity adjustment coefficient; Represents the future prediction of the current complaint event by the user sentiment prediction model The future trend of the user's emotions at the moment; Indicates the critical threshold of situational characteristics.

[0012] On the other hand, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the present invention when executing the program.

[0013] In yet another aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the method of the present invention when executed by a processor.

[0014] The present invention has the following beneficial effects: 1. The present invention constructs an emotional feature vocabulary and systematically scores the three dimensions of emotional intensity, emotional valence, and emotional control, effectively reflecting the multidimensional characteristics of emotional states and improving the comprehensiveness and meticulousness of emotional analysis.

[0015] 2. The present invention constructs three-dimensional emotional coordinates and an emotional interaction matrix to accurately capture the dynamic changes of emotions in complaint texts and the interactive influences between emotional dimensions, thereby achieving accurate prediction of emotional state trends.

[0016] 3. The present invention proposes a model based on the fusion of an improved bidirectional long short-term memory network and an attention mechanism to enhance the model's ability to capture the trend of emotional changes and achieve accurate prediction of future trends in user emotions, thereby more effectively inferring the actual processing needs of complaint events. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0020] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0021] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0022] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0023] Embodiment 1: See also Figure 1 , a method for predicting the timeliness demand of complaint event processing, comprising the following steps: Collect complaint event text data, segment the complaint event text data through a pre-trained large model, and extract sentiment feature words and situational features; Construct an emotional feature word library and set a three-dimensional score for each emotional feature word in the emotional feature word library, including emotional intensity score (calm to extremely intense), emotional valence score (very negative to very positive), and emotional control score (completely helpless to highly autonomous); Based on the emotional feature word library, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed for each emotional feature word in the complaint event text data; Based on each emotional feature word, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed to calculate the overall emotional state trend of the complaint event; Calculate the narrative tension of the complaint event based on the overall emotional state trend and situational characteristics of the complaint event; A user emotion prediction model is constructed based on an improved bidirectional long short-term memory network, and the future trend of the emotion of the user corresponding to the complaint event is predicted through the overall emotional state trend of the complaint event. The improved bidirectional long short-term memory network includes a positive and negative long short-term memory network layer and an attention fusion layer. The time efficiency requirement for handling the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions.

[0024] As a preferred implementation of this embodiment, the corresponding three-dimensional emotional coordinates are constructed for each emotional feature word in the complaint event text data based on the emotional feature word library, as shown in the following formula: ; in: Indicates the first The three-dimensional emotional coordinates corresponding to the emotional feature words; Indicates The emotional intensity score of each emotional feature word; Indicates The emotional valence score of the emotional feature words; Indicates The emotional control score of the emotional characteristic words; Based on the emotional feature word library, the corresponding emotional interaction matrix is ​​constructed for each emotional feature word in the complaint event text data as shown in the following formula: ; in: Indicates The emotional interaction matrix of emotional feature words; Indicates the natural attenuation coefficient of emotion intensity; represents the natural attenuation coefficient of emotional valence; It represents the natural attenuation coefficient of emotional control; Indicates sentiment feature words Dimension ratings and The interaction parameters between dimension ratings, , , .

[0025] As a preferred implementation of this embodiment, the specific steps of constructing the corresponding three-dimensional emotional coordinates and emotional interaction matrix based on each emotional feature word and calculating the emotional intensity curve, emotional valence curve and emotional control curve of the complaint event are as follows: According to the order of occurrence of emotional feature words in the complaint event text data, the emotional state vector is constructed based on the three-dimensional emotional coordinates of each emotional feature word. ,in, Indicates the first The three-dimensional emotional coordinates of the emotional feature words that appear, The three-dimensional sentiment coordinates representing the last sentiment feature word that appears in the complaint event text data; At the same time, according to the order of appearance of emotional feature words in the complaint event text data, the emotional interaction matrix vector is constructed based on the emotional interaction matrix of each emotional feature word. ,in, Indicates the first The emotional interaction matrix of the emotional feature words that appear, The sentiment interaction matrix representing the last sentiment feature word that appears in the complaint event text data; The emotional dynamics equation is constructed based on the emotional state vector and the emotional interaction matrix vector, as shown in the following formula: ; ; in: Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; Indicates The starting time of the occurrence of the sentiment feature word, that is, the position of the sentiment feature word in the complaint event text data; Indicates The ending moment of the first appearing emotional feature word, that is, The position of the emotional feature words in the complaint event text data; Indicates The external emotional impact of the emotional feature words that appear; Indicates The weight of the sentiment feature words that appear; represents a natural constant; For each time interval corresponding to the occurrence of the emotional feature word, the final state of the previous time interval is used as the initial state of the next interval. The above formula is independently solved using the RK4 method to obtain the emotional state vector trend of each time interval. The emotional state vector trends of each time interval are then spliced ​​together to obtain the overall emotional state trend.

[0026] As a preferred implementation of this embodiment, the calculation formula of the narrative tension of the complaint event is specifically shown as follows: ; ; ; in: Indicates the narrative tension of the complaint incident; Indicates the first The emotional feature words that appear in the corresponding time interval Weighted probability distribution of emotional states within; Indicates the first The emotional feature words that appear in the corresponding time interval The emotional entropy within; represents the smoothing coefficient; represents the emotional self-sensitivity coefficient; Indicates the sensitivity of the situational features of the complaint event; Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; , , Respectively represent The emotional feature words that appear in the corresponding time interval The trends of emotion intensity scores, emotion valence scores, and emotion control scores; The sensitivity of the situational features of the complaint events is determined based on a pre-designed complaint field scoring system. Taking the catering field as an example, the specifics are shown in the following table:

[0027] The context feature sensitivity of the complaint event is obtained by performing a weighted summation based on the context features and their corresponding sensitivities that appear in the complaint event text data.

[0028] As a preferred implementation of this embodiment, the forward and reverse long short-term memory network layer is specifically shown in the following formula: ; ; ; in: Indicates the forward and reverse long short-term memory network layers in The output result at the moment is the future of the current complaint event The future trend of the user's emotions at the moment; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; Indicates the forward and reverse long short-term memory network layers in Input of time; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; represents the long short-term memory network; Express and To splice; Each unit of the forward and reverse long short-term memory networks is provided with a dynamic emotion enhancement gate, as shown in the following formula: ; ; ; ; ; ; ; in: , , Represents the forget gate, input gate, and output gate of the forward and reverse long short-term memory network units respectively. Output at the moment; , , Represent the weight matrices of the forget gate, input gate, and output gate respectively; Indicates the forward and reverse long short-term memory network layers in Output results at the moment; , , Represent the bias vectors of the forget gate, input gate, and output gate respectively; Dynamic emotion enhancement gates representing forward and reverse LSTM network units Output at the moment; represents the weight matrix of the dynamic emotion enhancement gate; Represents the attention fusion layer Multi-scale attention fusion features output at every moment; represents the bias vector of the dynamic emotion enhancement gate; Represents the forward and reverse long short-term memory network units in moment-to-moment enhanced emotional candidate states; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; The weight matrix representing the cell states of the forward and reverse LSTM networks; Bias vectors representing the cell states of forward and reverse LSTM networks; Represents the forward and reverse long short-term memory networks in The hidden state of the moment; represents the Hadamard product operation; Represents the sigmoid function.

[0029] As a preferred implementation of this embodiment, the attention fusion layer includes a multi-scale convolution extraction unit, a scale interaction unit, a scale selection unit and a multi-scale feature fusion unit; In this embodiment, the multi-scale convolution extraction unit performs feature extraction of different scales on the outputs of the forward and reverse long short-term memory network layers through three convolution layers with convolution kernels of different sizes, as shown in the following formula: , ; in: express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale is Convolution operation; express Time to The output sequence of the forward and reverse long short-term memory network layer at the moment; Define the interaction weight matrix and set the size of the interaction weight matrix according to the number of scales. In this embodiment, , the interaction weight matrix is ​​specifically shown in the following formula: ; ; ; in: express The interaction weight matrix at the moment; express function; express The query matrix at the moment; express The key matrix at the moment; a matrix of learnable parameters representing the query matrix; a learnable parameter matrix representing the bond matrix; Indicates the preset feature dimension; Represents a transpose operation; The scale interaction unit updates the output features of the forward and reverse long short-term memory network layers at each scale based on the interaction weight matrix, as shown in the following formula: ; in: Indicates updated The time scale is The output features of the forward and reverse long short-term memory network layers; express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale selection unit is used to calculate the contribution of different scales, as shown in the following formula: ; in: Indicates that the scale selection unit is Time output scale Contribution of represents the learnable parameter matrix of the scale selection unit; represents the bias vector of the scale selection unit; The final output features of the forward and reverse long short-term memory network layers at different scales are calculated based on the contribution of different scales, as shown in the following formula: ; The multi-scale feature fusion unit is used to calculate the final output features of the forward and reverse long short-term memory network layers at different scales by the contribution of different scales, as described in the following formula: ; ; in: Representation scale exist The position adaptive weight at the moment; Represents the attention query vector, which can be regarded as a "template" or "standard feature" of emotional features. The features extracted at each scale will be compared with the "standard". The more similar the features are, the more they will be noticed by the model, thus obtaining a higher weight. The features with large differences will be assigned a lower attention weight. a learnable parameter matrix representing the position adaptation weights; Bias vector representing the position adaptation weights.

[0030] As a preferred implementation of this embodiment, the processing time requirement of the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions, as shown in the following formula: ; in: Indicates the time limit required for handling the current complaint event; Indicates the timeliness demand sensitivity adjustment coefficient (determined based on historical complaint processing timeliness data); Represents the future prediction of the current complaint event by the user sentiment prediction model The future trend of the user's emotions at the moment, ; Represents the critical threshold of contextual features, which is set based on the average context sensitivity.

[0031] Embodiment 2: An intelligent analysis and judgment system for complaint events, including a presentation layer, an application layer, a support layer, and a data layer; The presentation layer includes the PC web terminal and the front-end architecture. The PC web terminal is used for the user interface, which is convenient for people to submit complaints and view the processing progress. The front-end architecture uses JS / CSS, vue.js and element-plus to ensure that the interface is beautiful, interactive and smooth, and improve the user experience.

[0032] The application layer is mainly composed of Python and server services, redis services, and message queue services. Python provides the development language foundation. The server service supports customized development of the server side and ensures the system service process. The redis service performs structured caching, and the message queue service builds queue producer and consumer modes to improve system efficiency.

[0033] The support layer includes the Doubao large model, which has intelligent question and answer, proprietary domain knowledge base, proprietary domain workflow, large model fine-tuning, customized development and deployment functions. It also includes a complaint event processing timeliness demand prediction method described in Example 1, which is used to realize the emotion analysis of complaint events, narrative tension assessment and processing timeliness prediction. These functions help to accurately and intelligently judge people's complaints and provide reasonable handling suggestions and legal basis.

[0034] The data layer is divided into public cloud and private cloud. The public cloud stores desensitized large model question and answer data and desensitized knowledge base data, and the private cloud stores user data and sensitive data. This data storage method can not only ensure data security, but also achieve effective data utilization.

[0035] Embodiment three: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present invention when executing the program.

[0036] Embodiment 4: This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present invention is implemented.

[0037] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0038] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0039] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0040] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0041] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting the timeliness demand for complaint event processing, characterized in that: The following steps are involved: Collect complaint event text data, segment the complaint event text data through a pre-trained large model, and extract sentiment feature words and situational features; Construct an emotional feature word library and set a three-dimensional score for each emotional feature word in the emotional feature word library, including an emotional intensity score, an emotional valence score, and an emotional control score; Based on the emotional feature word library, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed for each emotional feature word in the complaint event text data; Based on each emotional feature word, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed to calculate the overall emotional state trend of the complaint event; Calculate the narrative tension of the complaint event based on the overall emotional state trend and situational characteristics of the complaint event; A user emotion prediction model is constructed based on an improved bidirectional long short-term memory network, and the future trend of the emotion of the user corresponding to the complaint event is predicted through the overall emotional state trend of the complaint event. The improved bidirectional long short-term memory network includes a positive and negative long short-term memory network layer and an attention fusion layer; The time efficiency requirement for handling the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions.

2. A method for predicting the timeliness demand for handling complaint events according to claim 1, characterized in that: Based on the sentiment feature word library, the corresponding three-dimensional sentiment coordinates are constructed for each sentiment feature word in the complaint event text data as shown in the following formula: ； in: Indicates the first The three-dimensional emotional coordinates corresponding to the emotional feature words; Indicates The emotional intensity score of each emotional feature word; Indicates The emotional valence score of the emotional feature words; Indicates The emotional control score of the emotional feature words; Based on the emotional feature word library, the corresponding emotional interaction matrix is ​​constructed for each emotional feature word in the complaint event text data as shown in the following formula: ； in: Indicates The emotional interaction matrix of emotional feature words; Indicates the natural attenuation coefficient of emotion intensity; represents the natural attenuation coefficient of emotional valence; It represents the natural attenuation coefficient of emotional control; Indicates sentiment feature words Dimension ratings and The interaction parameters between dimension ratings, , , , Indicates the emotional intensity dimension, Represents the emotional valence dimension, Represents the dimension of emotional control.

3. A method for predicting the timeliness demand for handling complaint events according to claim 2, characterized in that: Based on each emotional feature word, the corresponding three-dimensional emotional coordinates and emotional interaction matrix are constructed, and the specific steps for calculating the emotional intensity curve, emotional valence curve, and emotional control curve of the complaint event are as follows: Construct the emotional state vector based on the order of occurrence of emotional feature words in the complaint event text data and the three-dimensional emotional coordinates of each emotional feature word ,in, Indicates the first The three-dimensional emotional coordinates of the emotional feature words that appear, The three-dimensional sentiment coordinates representing the last sentiment feature word that appears in the complaint event text data; At the same time, the emotional interaction matrix vector is constructed based on the appearance order of emotional feature words in the complaint event text data and the emotional interaction matrix of each emotional feature word. ,in, Indicates the first The emotional interaction matrix of the emotional feature words that appear, The sentiment interaction matrix representing the last sentiment feature word that appears in the complaint event text data; The emotional dynamics equation is constructed based on the emotional state vector and the emotional interaction matrix vector, as shown in the following formula: ； ； in: Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; Indicates The starting time of the occurrence of the sentiment feature word, that is, the position of the sentiment feature word in the complaint event text data; Indicates The ending moment of the first appearing emotional feature word, that is, The position of the emotional feature words in the complaint event text data; Indicates The external emotional impact of the emotional feature words that appear; Indicates The weight of the sentiment feature words that appear; represents a natural constant; For each time interval corresponding to the occurrence of the emotional feature word, the final state of the previous time interval is used as the initial state of the next interval. The above formula is independently solved using the RK4 method to obtain the emotional state vector trend of each time interval. The emotional state vector trends of each time interval are then spliced ​​together to obtain the overall emotional state trend.

4. A method for predicting the timeliness demand for handling complaint events according to claim 3, characterized in that: The calculation formula of the narrative tension of the complaint event is as follows: ； ； ； in: Indicates the narrative tension of the complaint incident; Indicates the first The emotional feature words that appear in the corresponding time interval Weighted probability distribution of emotional states within; Indicates the first The emotional feature words that appear in the corresponding time interval The emotional entropy within; represents the smoothing coefficient; represents the emotional self-sensitivity coefficient; Indicates the sensitivity of the situational features of the complaint event; Indicates The emotional feature words that appear in the corresponding time interval The emotional state vector trend within; , , Respectively represent The emotional feature words that appear in the corresponding time interval The trends of emotion intensity scores, emotion valence scores, and emotion control scores.

5. A method for predicting the timeliness demand for handling complaint events according to claim 1, characterized in that: The forward and reverse long short-term memory network layer is specifically shown in the following formula: ； ； ； in: Indicates the forward and reverse long short-term memory network layers in Output results at the moment; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; Indicates the forward and reverse long short-term memory network layers in Input of time; Indicates that the forward long short-term memory network is The hidden state of the moment; Reverse long short-term memory network The hidden state of the moment; represents the long short-term memory network; Express and To splice; Each unit of the forward and reverse long short-term memory networks is provided with a dynamic emotion enhancement gate, as shown in the following formula: ； ； ； ； ； ； ； in: , , Represents the forget gate, input gate, and output gate of the forward and reverse long short-term memory network units respectively. Output at the moment; , , Represent the weight matrices of the forget gate, input gate, and output gate respectively; Indicates the forward and reverse long short-term memory network layers in Output results at the moment; , , Represent the bias vectors of the forget gate, input gate, and output gate respectively; Dynamic emotion enhancement gates representing forward and reverse LSTM network units Output at the moment; represents the weight matrix of the dynamic emotion enhancement gate; Represents the attention fusion layer Multi-scale attention fusion features output at every moment; represents the bias vector of the dynamic emotion enhancement gate; Represents the forward and reverse long short-term memory network units in moment-to-moment enhanced emotional candidate states; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; Represents the forward and reverse long short-term memory network units in The cell state at a given moment; The weight matrix representing the cell states of the forward and reverse LSTM networks; Bias vectors representing the cell states of forward and reverse LSTM networks; Represents the forward and reverse long short-term memory networks in The hidden state of the moment; represents the Hadamard product operation; Represents the sigmoid function.

6. A method for predicting the timeliness demand for handling complaint events according to claim 5, characterized in that: The attention fusion layer includes a multi-scale convolution extraction unit, a scale interaction unit, a scale selection unit and a multi-scale feature fusion unit; The multi-scale convolution extraction unit extracts features of different scales from the outputs of the forward and reverse long short-term memory network layers through convolution layers with convolution kernels of different sizes, as shown in the following formula: ； in: express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale is Convolution operation; express Time to The output sequence of the forward and reverse long short-term memory network layer at the moment; Define the interaction weight matrix, and set the size of the interaction weight matrix according to the number of scales. The interaction weight matrix is ​​specifically shown in the following formula: ； ； ； in: express The interaction weight matrix at the moment; express function; express The query matrix at the moment; express The key matrix at the moment; a matrix of learnable parameters representing the query matrix; a learnable parameter matrix representing the bond matrix; Indicates the preset feature dimension; Represents a transpose operation; The scale interaction unit updates the output features of the forward and reverse long short-term memory network layers at each scale based on the interaction weight matrix, as shown in the following formula: ； in: Indicates updated The time scale is The output features of the forward and reverse long short-term memory network layers; express The time scale is The output features of the forward and reverse long short-term memory network layers; The scale selection unit is used to calculate the contribution of different scales, as shown in the following formula: ； in: Indicates that the scale selection unit is Time output scale Contribution of represents the learnable parameter matrix of the scale selection unit; represents the bias vector of the scale selection unit; The final output features of the forward and reverse long short-term memory network layers at different scales are calculated based on the contribution of different scales, as shown in the following formula: ； The multi-scale feature fusion unit is used to calculate the final output features of the forward and reverse long short-term memory network layers at different scales by the contribution of different scales, as described in the following formula: ； ； in: Representation scale exist The position adaptive weight at the moment; represents the attention query vector; a learnable parameter matrix representing the position adaptation weights; Bias vector representing the position adaptation weights.

7. A method for predicting the timeliness demand for handling complaint events according to claim 4, characterized in that: The processing time requirement of the current complaint event is calculated based on the narrative tension of the complaint event and the future trend of the user's emotions, as shown in the following formula: ； in: Indicates the time limit required for handling the current complaint event; It represents the timeliness demand sensitivity adjustment coefficient; Represents the future prediction of the current complaint event by the user sentiment prediction model The future trend of the user's emotions at the moment; Indicates the critical threshold of situational characteristics.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images