Intelligent customer service abnormal conversation detection and intervention method based on deep learning

Abstract

The application relates to the field of artificial intelligence, and provides an intelligent customer service abnormal conversation detection and intervention method based on deep learning, conversation samples are collected, the conversation samples are classified and labeled, and training materials including labeling results are obtained; an initial detection model is constructed, the initial detection model is trained through the training materials, and a target detection model is obtained; user conversation content is collected, the target detection model detects the user conversation content, and abnormal scores including semantic abnormal scores, emotion abnormal scores, behavior abnormal scores and response abnormal scores are obtained; abnormal score weights are determined, total abnormal scores are obtained according to the abnormal score weights and the abnormal scores, an auxiliary adjustment factor is constructed, the total abnormal scores are corrected by using the auxiliary adjustment factor, and corrected total scores are obtained; a dynamic adjustment coefficient is calculated, an intervention trigger condition is determined, an intervention strategy is obtained according to the dynamic adjustment coefficient and the intervention trigger condition, and the detection and intervention of abnormal conversations are completed.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning. Background Technology

[0002] With the widespread application of AI-powered intelligent customer service in e-commerce, finance, government affairs, and other fields, its advantages of 24 / 7 response and low-cost coverage are significant. However, core pain points still exist in complex scenarios: 1. Insufficient accuracy in abnormal dialogue detection: Existing technologies mostly rely on keyword matching or simple rules to determine anomalies, which cannot identify hidden anomalies such as "the user's tone is stable but the question is repeated and unresolved" or "the AI ​​response deviates from the requirements but does not trigger the keyword", resulting in a high false negative rate; 2. Rigid human intervention mechanism: Most systems adopt a passive mode of triggering manual transfer when an anomaly is detected, without considering factors such as the urgency of the anomaly and whether AI can still solve it. For example, in scenarios where the user is slightly dissatisfied but the AI ​​can resolve the issue with an upgraded response, blindly transferring leads to a waste of human customer service resources. 3. Lack of tiered intervention methods: All abnormal scenarios are handled with a full takeover approach, without designing differentiated processes for low-urgency inquiries and high-urgency complaints. For example, simple rights inquiries require full manual takeover, which leads to longer waiting times for high-urgency issues and affects user experience. 4. Lack of iterative capability: The triggering strategy is not optimized based on feedback from human customer service. As business scenarios change, the triggering mechanism gradually becomes ineffective, requiring frequent manual adjustments to the rules, resulting in high maintenance costs. Summary of the Invention

[0003] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a deep learning-based method for detecting and intervening in abnormal dialogues of intelligent customer service, achieving highly accurate and timely intervention in such dialogues by AI customer service representatives.

[0004] This invention provides a deep learning-based intelligent customer service abnormal dialogue detection and intervention method, comprising: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

[0005] According to the present invention, a method for detecting and intervening in abnormal dialogues of intelligent customer service based on deep learning is provided. In step S1, the dialogue samples are classified and labeled according to the abnormality type, abnormality score and optimal intervention method to obtain the training material.

[0006] According to the deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention, step S2 further includes: S21: Construct an initial detection model, wherein the initial detection model extracts user behavior features from the training material through a sliding window, extracts emotion features through a convolutional neural network, and obtains the response standard score by calculating cosine similarity; S22: Train the initial detection model based on the user behavior features, emotional features, response standardization and the annotation results to obtain the target detection model.

[0007] According to the deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention, step S3 further includes: S31: Collect multiple rounds of user dialogue to obtain the user dialogue content. The target detection model uses a convolutional neural network to determine the basic sentiment score of the user dialogue content and judges the dialogue sentiment trend to obtain the sentiment change rate score. The sentiment abnormality score is obtained by combining the basic sentiment score and the sentiment change rate score. S32: Determine the cumulative dialogue duration and the number of repeated questions, and obtain the behavior anomaly score based on the cumulative dialogue duration and the number of repeated questions; S33: Perform semantic detection on the user dialogue content to obtain the semantic anomaly score and the response anomaly score.

[0008] According to the deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention, step S4 further includes: S41: Determine the anomaly score weights, and use the anomaly score weights to perform a weighted summation of the semantic anomaly score, emotional anomaly score, behavioral anomaly score, and response anomaly score to obtain the total anomaly score; S42: Determine the nature of the user dialogue content and the user account status, construct the auxiliary adjustment factor based on the nature of the dialogue and the user account status, and use the auxiliary adjustment factor to correct the total abnormal score to obtain the corrected total score.

[0009] According to the deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention, step S5 further includes: S51: Obtain historical intervention data and calculate the dynamic adjustment coefficient based on the historical intervention data; S52: Determine the intervention trigger conditions including the condition threshold, adjust the condition threshold using the dynamic adjustment coefficient, and obtain the intervention strategy using the adjusted intervention trigger conditions and the corrected total score to complete the detection and intervention of abnormal dialogue.

[0010] This invention also provides a deep learning-based intelligent customer service abnormal dialogue detection and intervention system, comprising: Sample annotation module: used to collect dialogue samples, classify and annotate the dialogue samples, and obtain training materials including the annotation results; Model training module: Used to build an initial detection model, and train the initial detection model with training materials to obtain the target detection model; Anomaly score module: Used to collect user dialogue content. The object detection model detects the user dialogue content and obtains anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. The scoring correction module is used to determine the weight of the abnormal score, obtain the total abnormal score based on the abnormal score weight and the abnormal score, construct an auxiliary adjustment factor, and use the auxiliary adjustment factor to correct the total abnormal score to obtain the corrected total score. Intervention Strategy Module: Used to calculate dynamic adjustment coefficients, determine intervention trigger conditions, and obtain intervention strategies based on the corrected total score, dynamic adjustment coefficients, and intervention trigger conditions to complete the detection and intervention of abnormal dialogues.

[0011] The present invention also 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 executes the computer program to implement the steps of the deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described above.

[0012] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described above.

[0013] The present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, and when the program instructions are executed by a computer, the computer is able to perform the steps of a deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described above.

[0014] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: This invention provides a deep learning-based intelligent customer service abnormal dialogue detection and intervention method. By training the detection model in a targeted manner, a target detection model is obtained, which then obtains anomaly scores from the perspectives of semantic anomalies, user behavior and emotional anomalies, and response anomalies. The anomaly scores are then adjusted to a certain extent to obtain a corrected total score, and finally a graded intervention strategy is obtained to complete the timely detection and appropriate intervention of abnormal dialogues.

[0015] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of an intelligent customer service abnormal dialogue detection and intervention system based on deep learning provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the structure of an intelligent customer service abnormal dialogue detection and intervention device based on deep learning provided by the present invention.

[0020] Figure label: 100. Sample labeling module; 200. Model training module; 300. Anomaly score module; 400. Scoring correction module; 500. Intervention strategy module; 810. Processor; 820. Communication interface; 830. Memory; 840. Communication bus. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0022] In the description of the embodiments of the present invention, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0023] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention based on the specific circumstances.

[0024] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0025] The following is combined Figures 1 to 3 Specific embodiments of the present invention are described below. Figure 1 This is a flowchart illustrating a deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the present invention, including: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; Furthermore, the purpose of this stage is to classify and label the dialogue samples to obtain training materials including the labeling results. Specifically, in step S1, the dialogue samples are classified and labeled according to the anomaly type, anomaly score and optimal intervention method to obtain the training materials.

[0026] The specific implementation method for the above steps in this embodiment is as follows: First, historical dialogue samples from the intelligent customer service system need to be collected as dialogue samples, including normal and abnormal dialogue samples, as well as special cases such as gradual changes in user emotions such as anger. The dialogue samples are then manually categorized according to the type of abnormality and labeled with an anomalousness score based on the degree of abnormality. User emotions and behaviors also need to be labeled, and the optimal intervention method is assigned based on the specific circumstances of the dialogue. This process yields the labeled results. The labeled dialogue samples are then used as training material.

[0027] S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; Furthermore, the objective of this stage is to train the initial detection model to obtain the object detection model. Specifically, step S2 further includes: S21: Construct an initial detection model, wherein the initial detection model extracts user behavior features from the training material through a sliding window, extracts emotion features through a convolutional neural network, and obtains the response standard score by calculating cosine similarity; S22: Train the initial detection model based on the user behavior features, emotional features, response standardization and the annotation results to obtain the target detection model.

[0028] The specific implementation method for the above steps in this embodiment is as follows: First, an initial detection model needs to be built. Then, the initial detection model is trained using training materials. During the training process, the initial detection model learns the content of the training materials and the annotation results. In this way, the convolutional neural network can be used to identify and extract the customer's emotions in the training materials. The user's behavioral features are extracted using a sliding window with a size of 5 rounds of dialogue. The cosine similarity between the answers generated by the intelligent customer service and the written standard answers is calculated to obtain the answer standardization.

[0029] The user behavior and emotion features identified by the initial detection model are compared with the relevant content in the labeled results to make them closer to the standard results. At the same time, the standard of the responses generated by the initial detection model should be gradually improved. The initial detection model is trained in multiple rounds with this goal in mind to obtain the target detection model. The anomaly score and accuracy of the target detection model need to meet the requirements.

[0030] S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. Furthermore, the objective of this stage is to detect user dialogue content and obtain anomaly scores, including semantic anomaly scores, sentiment anomaly scores, behavioral anomaly scores, and response anomaly scores. Specifically, step S3 further includes: S31: Collect multiple rounds of user dialogue to obtain the user dialogue content. The target detection model uses a convolutional neural network to determine the basic sentiment score of the user dialogue content and judges the dialogue sentiment trend to obtain the sentiment change rate score. The sentiment abnormality score is obtained by combining the basic sentiment score and the sentiment change rate score. S32: Determine the cumulative dialogue duration and the number of repeated questions, and obtain the behavior anomaly score based on the cumulative dialogue duration and the number of repeated questions; S33: Perform semantic detection on the user dialogue content to obtain the semantic anomaly score and the response anomaly score.

[0031] The specific implementation method for the above steps in this embodiment is as follows: Here, the intelligent customer service needs to engage in multiple rounds of dialogue with the user and collect these dialogues to obtain the user's conversation content. The object detection model uses a convolutional neural network to judge the emotion contained in the user's conversation content and obtain a basic emotion score. Specifically, the base emotion score is 0 for "satisfaction," 3 for "neutral," 7 for "anxiety," and 10 for "anger." Furthermore, considering drastic changes in user emotions, an additional score is added for the emotion variability rate. Here, if a user's emotion rapidly changes from neutral to anxious, or from anxious to angry, or from neutral to angry within 5 rounds of dialogue, the emotion change rate score is increased by 2; otherwise, it is increased by 0. The emotion abnormality score is obtained by adding the base emotion score and the emotion change rate score. .

[0032] After determining the cumulative conversation duration T, it is also necessary to determine the number of times N the user repeatedly asked the same question or core request during the conversation. This allows us to calculate the abnormal behavior score. : Here, min() means taking the minimum of the two values ​​in parentheses.

[0033] It is also necessary to analyze the user dialogue content and determine the cosine similarity between the AI ​​customer service's response and a pre-set standard response template for user questions, thereby obtaining semantic anomaly scores. Specifically, when the cosine similarity is less than 0.4, the semantic anomaly score is 10; when it is greater than 0.8, the semantic anomaly score is 0; and when it falls between these two values, a semantic anomaly score is obtained through linear interpolation. Furthermore, it is necessary to determine the amount of key information included in the AI ​​customer service's response that is required to resolve the user's problem. If the AI ​​customer service's response includes all the key information needed to solve the problem, then the response anomaly score is [not specified]. The value is 0. When it does not contain any key information needed to solve the problem, the response is classified as 10.

[0034] S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. Furthermore, the objective of this stage is to obtain the total anomaly score and correct it to obtain the corrected total score. Specifically, step S4 further includes: S41: Determine the anomaly score weights, and use the anomaly score weights to perform a weighted summation of the semantic anomaly score, emotional anomaly score, behavioral anomaly score, and response anomaly score to obtain the total anomaly score; S42: Determine the nature of the user dialogue content and the user account status, construct the auxiliary adjustment factor based on the nature of the dialogue and the user account status, and use the auxiliary adjustment factor to correct the total abnormal score to obtain the corrected total score.

[0035] The specific implementation method for the above steps in this embodiment is as follows: First, the weights of the anomaly scores need to be determined based on the specific circumstances. These weights include a first anomaly score weight α, a second anomaly score weight β, a third anomaly score weight γ, and a fourth anomaly score weight δ. Then, the semantic anomaly scores, sentiment anomaly scores, behavioral anomaly scores, and response anomaly scores can be weighted and summed to obtain the total anomaly score S. Here, not all user conversations are necessarily valid, and some users may have higher priority or ask more urgent questions. Therefore, it's necessary to determine the nature of the conversation and the user's account status. Specifically, if the user is a high-value customer with a high membership level, or if the user's question is an urgent issue such as an account freeze or order payment failure, the auxiliary adjustment factor will range from 0.5 to 1. When the user is a new user, or when the question is a tentative inquiry about how to use the system or its functions, the auxiliary adjustment factor will range from -0.5 to -1. Adding the auxiliary adjustment factor to the total anomaly score yields the corrected total score.

[0036] S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

[0037] Specifically, the objective of this stage is to derive an intervention strategy based on the revised total score, dynamic adjustment coefficients, and intervention trigger conditions, thereby completing the detection and intervention of abnormal dialogues. Specifically, step S5 further includes: S51: Obtain historical intervention data and calculate the dynamic adjustment coefficient based on the historical intervention data; S52: Determine the intervention trigger conditions including the condition threshold, adjust the condition threshold using the dynamic adjustment coefficient, and obtain the intervention strategy using the adjusted intervention trigger conditions and the corrected total score to complete the detection and intervention of abnormal dialogue.

[0038] The specific implementation method for the above steps in this embodiment is as follows: Here, the system retrieves the intervention records from the previous day, and then calculates the dynamic adjustment coefficient P based on the historical intervention data. Specifically, the calculation method for the dynamic adjustment coefficient is as follows: Where b represents the total number of manual interventions on the previous day in the historical intervention data, and a represents the effective number of manual interventions on the previous day in the historical intervention data.

[0039] Next, determine the intervention trigger conditions, including condition thresholds. Condition thresholds are the specific values ​​of the conditions that trigger different intervention levels. Specifically, if the total score is below 3, no intervention is needed; if the total score is between 3 and 5 and the intelligent customer service has not optimized the given answer, a simple warning is sufficient; if the total score is between 5 and 8 and the intelligent customer service has optimized the given answer once, or if the total score is between 5 and 6 and the intelligent customer service has not optimized the given answer, a semi-takeover intervention is required. This involves sending the user a notification that "We have connected you with a dedicated customer service assistant who will provide more accurate answers based on your questions," and sending the intelligent customer service-generated answer to the backend for adjustment before sending it back to the user; if the total score is greater than 8 and the intelligent customer service has optimized the given answer twice, or if the total score is greater than 9 and the intelligent customer service has optimized the given answer once, a full takeover intervention is required. This involves notifying the user that a human assistant has been assigned to them, and sending the intelligent customer service conversation record to the backend for direct service to the user. Here, for the conditional thresholds related to the overall score correction, the specific value of the conditional threshold needs to be multiplied by the dynamic adjustment coefficient to adjust the intervention triggering conditions. The intervention strategy is then derived from the adjusted intervention triggering conditions, thereby enabling the detection and intervention of abnormal dialogues.

[0040] The following describes a deep learning-based intelligent customer service abnormal dialogue detection and intervention device provided by the present invention. The deep learning-based intelligent customer service abnormal dialogue detection and intervention device described below and the deep learning-based intelligent customer service abnormal dialogue detection and intervention method described above can be referred to and correspond to each other.

[0041] Figure 2 This example illustrates the structural diagram of a deep learning-based intelligent customer service abnormal dialogue detection and intervention system, as shown below. Figure 2 As shown, the method for detecting and intervening in abnormal customer service dialogues based on deep learning, as described above, includes: Sample labeling module 100: Used to collect dialogue samples, classify and label the dialogue samples, and obtain training materials including the labeling results; Model training module 200: Used to build an initial detection model, and train the initial detection model with training materials to obtain the target detection model; Anomaly score module 300: Used to collect user dialogue content. The object detection model detects the user dialogue content and obtains anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score and response anomaly score. Scoring Correction Module 400: Used to determine the weight of abnormal scores, obtain the total abnormal score based on the weight of abnormal scores and the abnormal scores, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. Intervention Strategy Module 500: Used to calculate dynamic adjustment coefficients, determine intervention trigger conditions, obtain intervention strategies based on the corrected total score, dynamic adjustment coefficients, and intervention trigger conditions, and complete the detection and intervention of abnormal dialogues.

[0042] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call a computer program stored in the memory 830 to execute a deep learning-based intelligent customer service abnormal dialogue detection and intervention method, which includes: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

[0043] Furthermore, when the computer program in the aforementioned memory 830 can be implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute a deep learning-based intelligent customer service abnormal dialogue detection and intervention method provided by the above methods, the method comprising: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

[0045] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for detecting and intervening in abnormal dialogues of intelligent customer service based on deep learning, the method comprising: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

[0046] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0047] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning, characterized in that, include: S1: Collect dialogue samples, classify and label the dialogue samples to obtain training materials including the labeling results; S2: Construct an initial detection model, and train the initial detection model using training materials to obtain the target detection model; S3: Collect user dialogue content, and the target detection model detects the user dialogue content to obtain anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. S4: Determine the weight of the abnormal score, obtain the total abnormal score based on the weight of the abnormal score and the abnormal score, construct an auxiliary adjustment factor, use the auxiliary adjustment factor to correct the total abnormal score, and obtain the corrected total score. S5: Calculate the dynamic adjustment coefficient, determine the intervention trigger condition, obtain the intervention strategy based on the corrected total score, dynamic adjustment coefficient and intervention trigger condition, and complete the detection and intervention of abnormal dialogue.

2. The method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning according to claim 1, characterized in that, In step S1, the dialogue samples are classified and labeled according to anomaly type, anomaly score and optimal intervention method to obtain the training material.

3. The method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning according to claim 1, characterized in that, Step S2 further includes: S21: Construct an initial detection model, wherein the initial detection model extracts user behavior features from the training material through a sliding window, extracts emotion features through a convolutional neural network, and obtains the response standard score by calculating cosine similarity; S22: Train the initial detection model based on the user behavior features, emotional features, response standardization and the annotation results to obtain the target detection model.

4. The method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning according to claim 1, characterized in that, Step S3 further includes: S31: Collect multiple rounds of user dialogue to obtain the user dialogue content. The target detection model uses a convolutional neural network to determine the basic sentiment score of the user dialogue content and judges the dialogue sentiment trend to obtain the sentiment change rate score. The sentiment abnormality score is obtained by combining the basic sentiment score and the sentiment change rate score. S32: Determine the cumulative dialogue duration and the number of repeated questions, and obtain the behavior anomaly score based on the cumulative dialogue duration and the number of repeated questions; S33: Perform semantic detection on the user dialogue content to obtain the semantic anomaly score and the response anomaly score.

5. The method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning according to claim 1, characterized in that, Step S4 further includes: S41: Determine the anomaly score weights, and use the anomaly score weights to perform a weighted summation of the semantic anomaly score, emotional anomaly score, behavioral anomaly score, and response anomaly score to obtain the total anomaly score; S42: Determine the nature of the user dialogue content and the user account status, construct the auxiliary adjustment factor based on the nature of the dialogue and the user account status, and use the auxiliary adjustment factor to correct the total abnormal score to obtain the corrected total score.

6. The method for detecting and intervening in abnormal dialogues in intelligent customer service based on deep learning according to claim 1, characterized in that, Step S5 further includes: S51: Obtain historical intervention data and calculate the dynamic adjustment coefficient based on the historical intervention data; S52: Determine the intervention trigger conditions including the condition threshold, adjust the condition threshold using the dynamic adjustment coefficient, and obtain the intervention strategy using the adjusted intervention trigger conditions and the corrected total score to complete the detection and intervention of abnormal dialogue.

7. A deep learning-based intelligent customer service abnormal dialogue detection and intervention system, used to execute the deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described in any one of claims 1 to 6, characterized in that, include: Sample annotation module: used to collect dialogue samples, classify and annotate the dialogue samples, and obtain training materials including the annotation results; Model training module: Used to build an initial detection model, and train the initial detection model with training materials to obtain the target detection model; Anomaly score module: Used to collect user dialogue content. The object detection model detects the user dialogue content and obtains anomaly scores including semantic anomaly score, sentiment anomaly score, behavioral anomaly score, and response anomaly score. The scoring correction module is used to determine the weight of the abnormal score, obtain the total abnormal score based on the abnormal score weight and the abnormal score, construct an auxiliary adjustment factor, and use the auxiliary adjustment factor to correct the total abnormal score to obtain the corrected total score. Intervention Strategy Module: Used to calculate dynamic adjustment coefficients, determine intervention trigger conditions, and obtain intervention strategies based on the corrected total score, dynamic adjustment coefficients, and intervention trigger conditions to complete the detection and intervention of abnormal dialogues.

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 computer program, it implements the steps of the deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, characterized in that, When the program instructions are executed by the computer, the computer is able to perform the steps of the deep learning-based intelligent customer service abnormal dialogue detection and intervention method as described in any one of claims 1 to 6.

Images