Customer emotional fluctuation analysis method, system and device and storage medium

Abstract

The invention provides a customer emotion fluctuation analysis method, system and device and a storage medium, which can be applied to the field of medical health insurance and financial claim service, and the method comprises the following steps: obtaining historical call audio, and separating customer audio from the historical call audio; converting the client audio into a text, performing semantic analysis on the text, and outputting an emotion state parameter of the client in a historical call process; constructing an emotion state parameter array according to the sequential relationship, marking emotion stable segments in the emotion state parameter array, and constructing a jump relationship table based on the emotion stable segments; in response to a query demand, determining the longest continuous stable interval of the customer emotion after a specified node through the jump relation table; generating customer emotion grading early warning information in combination with the longest continuous stable interval and the corresponding emotion state parameter range; the emotion of the customer can be objectively and uniformly analyzed and evaluated, and the early warning information is generated in time to intervene and reduce the complaint risk in advance.

Description

Technical Field

[0001] The present application belongs to the field of emotion monitoring technology, and specifically relates to a method, system, device and storage medium for analyzing customer emotion fluctuations. Background Art

[0002] In the healthcare and financial claims services sectors, the insured's emotional state and psychological state during the claims process significantly impact the effectiveness of business processing. During the claims negotiation process, policyholders are prone to persistent negative emotions due to information asymmetry, poor communication, complex procedures, or claims outcomes that don't meet expectations. If these emotional fluctuations aren't addressed promptly, they can lead to increased risk of complaints from insurance companies.

[0003] Traditional customer emotion recognition methods mainly rely on manual analysis of customer emotional states through telephone communication. This method is greatly influenced by customer service subjective factors. Different customer service staff may have different emotional judgments on the same conversation, making it difficult to establish a unified risk assessment standard.

[0004] At the same time, traditional identification methods are unable to predict the continuous trend of changes in customer emotions and lack long-term and effective monitoring of customer emotional stability, resulting in delayed warnings of high-value customer loss and slow responses to sudden complaints, causing insurance institutions to miss key intervention opportunities and, to a certain extent, increasing complaint risks and customer retention costs. Summary of the Invention

[0005] This application provides a customer emotion fluctuation analysis method, system, device and storage medium, which can objectively and uniformly analyze and evaluate customer emotions, and generate early warning information in a timely manner to intervene in advance and reduce the risk of complaints.

[0006] In order to solve the above technical problems, in a first aspect, the present application provides a method for analyzing customer emotion fluctuations, comprising the following steps:

[0007] Acquire historical call audio, and separate the customer audio from the historical call audio;

[0008] Convert the customer audio into text, perform semantic analysis on the text, and output emotional state parameters of the customer during the historical call;

[0009] constructing an emotional state parameter array according to a time sequence relationship, marking emotionally stable segments in the emotional state parameter array, and constructing a jump relationship table based on the emotionally stable segments;

[0010] In response to the query requirement, determining the longest continuous stable interval of customer emotions after the designated node through the jump relationship table;

[0011] The customer emotion classification warning information is generated by combining the longest continuous stable interval and the corresponding emotional state parameter range.

[0012] As a further improvement of the present application, obtaining historical call audio and separating the customer audio from the historical call audio includes:

[0013] Acquire historical call audio, perform voiceprint feature extraction on the historical call audio, and separate the customer audio from the historical call audio.

[0014] As a further improvement of the present application, converting the customer audio into text, performing semantic analysis on the text, and outputting the customer's emotional state parameters during historical calls include:

[0015] converting the customer audio into text via automatic speech recognition technology;

[0016] Analyze and process the text using preset semantic analysis rules to output emotional state parameters of the customer during historical calls;

[0017] The emotional state parameter is used to characterize the intensity of the customer's emotions during historical calls.

[0018] As a further improvement of the present application, the step of marking the emotionally stable segment in the emotional state parameter array includes:

[0019] The difference between two adjacent emotional state parameters in the marked emotional state parameter array is calculated. If the difference is less than a preset fluctuation threshold, the current time node and the adjacent time node are marked as emotional stable segments.

[0020] As a further improvement of the present application, the step of constructing the jump relationship table includes:

[0021] Constructing the jump relationship table by combining the emotionally stable segment, the customer name, and the time node corresponding to the emotionally stable segment;

[0022] Calculate the farthest stable node at each time node under different jump step sizes;

[0023] The jump step size is a power of 2, and the farthest stable node is used to locate the boundary of the longest continuous stable interval.

[0024] As a further improvement of the present application, the method of determining the longest continuous stable interval of customer emotions after a designated node through the jump relationship table includes:

[0025] Starting from the designated node, jumping step by step starting from the maximum jump step based on the jump step sizes of each level pre-stored in the jump relationship table;

[0026] If the current designated node is still in the pre-marked emotional stability segment after the jump, the current node is updated to the node after the jump, and the current jump step is accumulated;

[0027] If the current designated node is not within the emotional stability segment after the jump, the current time node is kept unchanged, and the jump step is reduced to continue the step-by-step jump;

[0028] Repeat the step-by-step jump until all jump steps cannot continue to jump, forming the longest continuous stable interval with the specified node as the starting point and the farthest stable node reached by the accumulated jump steps as the end point.

[0029] As a further improvement of the present application, the generation of customer emotion classification warning information by combining the longest continuous stable interval and the corresponding emotional state parameter range includes:

[0030] When the emotional state parameters of the longest continuous stable interval are all lower than a first preset threshold, generating an emotional stability warning;

[0031] When the emotional state parameter of the longest continuous stable interval is between the first preset threshold and the second preset threshold, or the difference between adjacent emotional state parameters in the interval exceeds the preset stability threshold, an emotional fluctuation warning is generated;

[0032] When the emotional state parameters of the longest continuous stable interval are all higher than the second preset threshold and show a monotonically increasing trend, or the mutation amplitude of the emotional state parameters at adjacent time nodes exceeds the preset mutation threshold, a mood deterioration warning is generated;

[0033] The first preset threshold is smaller than the second preset threshold.

[0034] In a second aspect, the present application provides a customer emotion fluctuation analysis system, which includes:

[0035] an acquisition unit, configured to acquire historical call audio and separate the customer audio from the historical call audio;

[0036] An output unit, configured to convert the customer audio into text, perform semantic analysis on the text, and output emotional state parameters of the customer during historical calls;

[0037] a construction unit, configured to construct an emotional state parameter array according to a time sequence relationship, mark emotionally stable segments in the emotional state parameter array, and construct a jump relationship table based on the emotionally stable segments;

[0038] A query unit, configured to respond to a query demand and determine the longest continuous stable interval of customer emotions after a specified node through the jump relationship table;

[0039] The early warning unit is used to generate customer emotion classification early warning information based on the longest continuous stable interval and the corresponding emotional state parameter range.

[0040] In a third aspect, the present application provides a computer device comprising a processor and a memory coupled to the processor, wherein a computing program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of any one of the above-mentioned customer emotion fluctuation analysis methods.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement any of the above-mentioned customer emotion fluctuation analysis methods.

[0042] Compared with the existing technology, the customer emotion fluctuation analysis method, system, device and storage medium provided by this application obtain historical call audio to separate the customer audio therein, convert the customer audio into text and perform semantic analysis, and output time-series emotion state parameters to provide a data basis for subsequent emotion fluctuation analysis; then construct an emotion state parameter array, mark the emotional stability segments therein and construct a jump relationship table, and use the principle of the doubling algorithm to achieve fast query of the longest continuous stable interval, which significantly improves the processing efficiency of large-scale data; finally, combining the longest continuous stable interval and the emotion parameter range, different warning information is generated through graded threshold comparison to accurately identify customer emotion risks. This application can objectively and uniformly analyze and evaluate customer emotions, reduce the number of queries and improve query efficiency through the jump relationship table, so as to generate warning information in a timely manner, ensuring that customer service personnel in scenarios such as medical health insurance and financial claims can intervene in advance, thereby reducing the risk of complaints. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A flowchart of a method for analyzing customer emotional fluctuations provided in an embodiment of the present application;

[0045] Figure 2 A flowchart of outputting emotional state parameters in the customer emotional fluctuation analysis method provided in an embodiment of the present application;

[0046] Figure 3 A flowchart of constructing a jump relationship table in the customer emotion fluctuation analysis method provided in an embodiment of the present application;

[0047] Figure 4 A flowchart for determining the longest continuous stable interval in the customer emotion fluctuation analysis method provided in an embodiment of the present application;

[0048] Figure 5 A flowchart of generating customer emotion classification warning information in the customer emotion fluctuation analysis method provided in an embodiment of the present application;

[0049] Figure 6 A schematic diagram of the structure of a customer emotion fluctuation analysis system provided in an embodiment of the present application;

[0050] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present application;

[0051] Figure 8 A schematic diagram of the structure of the storage medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clearly understood, the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present application and are not intended to limit the embodiments of the present application.

[0053] In the description of the embodiments of the present application, the meaning of "plurality" is at least two, for example, two, three, etc., unless otherwise clearly and specifically limited. In the embodiments of the present application, all directional indications (such as up, down, left, right, front, back...) are only used to explain the relative position relationship, movement, etc. between the components under a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0054] To provide a more detailed and complete description of the present disclosure, the following provides illustrative descriptions of implementation methods and specific embodiments of the present disclosure; however, these descriptions are not intended to be the only way to implement or use the embodiments of the present disclosure. The implementation methods cover features of various specific embodiments, as well as the method steps and sequences for constructing and operating these specific embodiments. However, other specific embodiments may also be used to achieve the same or equivalent functionality and step sequences.

[0055] In the embodiments of this application, the terms "exemplary," "in some embodiments," and "in another embodiment" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" should not be construed as preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner.

[0056] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0057] In the healthcare and financial claims services sectors, the insured's emotional state and psychological state during the claims process significantly impact the effectiveness of business processing. During the claims negotiation process, policyholders are prone to persistent negative emotions due to information asymmetry, poor communication, complex procedures, or claims outcomes that don't meet expectations. If these emotional fluctuations aren't addressed promptly, they can lead to increased risk of complaints from insurance companies.

[0058] Traditional customer emotion recognition methods mainly rely on manual analysis of customer emotional states through telephone communication. This method is greatly influenced by customer service subjective factors. Different customer service staff may have different emotional judgments on the same conversation, making it difficult to establish a unified risk assessment standard.

[0059] At the same time, traditional identification methods are unable to predict the continuous trend of changes in customer emotions and lack long-term and effective monitoring of customer emotional stability, resulting in delayed warnings of high-value customer loss and slow responses to sudden complaints, causing insurance institutions to miss key intervention opportunities and, to a certain extent, increasing complaint risks and customer retention costs.

[0060] In view of this, please refer to Figures 1-8 The embodiments of the present application provide a method, system, device and storage medium for analyzing customer emotion fluctuations, which can objectively and uniformly analyze and evaluate customer emotions, and generate early warning information in a timely manner to intervene in advance and reduce the risk of complaints.

[0061] It is understandable that in the field of medical health insurance and financial claims services, policyholders may experience emotional changes during the claims process due to information asymmetry, poor communication, complicated procedures, or claims results that do not meet expectations. If these are not promptly addressed or handled, they may face a higher risk of complaints. Therefore, it is necessary to promptly analyze and intervene in the customer's emotional fluctuations to reduce the risk of complaints. Next, taking medical health insurance and financial claims services as an example, the customer emotional fluctuation analysis method provided in this application will be explained in detail.

[0062] Please refer to Figure 1 , is a flow chart of a method for analyzing customer emotional fluctuations provided in an embodiment of the present application, the detection method comprising the following steps:

[0063] Step S1: Acquire historical call audio and separate the customer audio from the historical call audio;

[0064] As an optional implementation manner, obtaining historical call audio and separating customer audio from the historical call audio includes:

[0065] Acquire historical call audio, perform voiceprint feature extraction on the historical call audio, and separate the customer audio from the historical call audio.

[0066] In an embodiment of the present application, the insurance company's business system can be used to first batch obtain historical call audio of customers communicating with customer service through the customer service hotline or online voice communication mini-program.

[0067] Furthermore, it is necessary to process historical call audio in scenarios such as medical and health insurance claims and financial policy services, and use voiceprint recognition technologies such as HMM (Hidden Markov Model) to extract features from historical call audio in order to separate the customer audio from the mixed historical call audio of customers and customer service.

[0068] It is understandable that historical call audio usually contains emotional clues accumulated by customers in long-term communication, such as the emotional change process of customers from patiently asking questions to gradually getting angry. Therefore, it is necessary to analyze and process the customer's historical call audio to avoid relying solely on a single call or the current call to analyze the customer's emotions, which may lead to one-sided judgments.

[0069] In addition, customer service call audio is usually a standardized response such as "I understand your feelings very well", which has little correlation with customer sentiment analysis. Some soothing words may also reduce the accuracy of customer sentiment analysis. Therefore, it is necessary to separate the customer audio from the historical call audio to avoid misleading the analysis of customer emotions due to customer service response words.

[0070] In an optional embodiment, the insured Mr. Zhang called the customer service hotline several times due to issues related to reimbursement of surgical expenses. After the system batch-acquired historical call audios of Mr. Zhang communicating with customer service through the customer service hotline, it used voiceprint recognition technology to separate Mr. Zhang's call audio from multiple historical call audios.

[0071] Step S2: converting the customer audio into text, performing semantic analysis on the text, and outputting the customer's emotional state parameters during the historical call process;

[0072] As an optional implementation, please refer to Figure 2 , which is a flow chart for outputting emotional state parameters in the customer emotional fluctuation analysis method provided in an embodiment of the present application. The above-mentioned method converts customer audio into text, performs semantic analysis on the text, and outputs the customer's emotional state parameters during historical calls, including:

[0073] Step S20: Converting the customer audio into text using automatic speech recognition technology;

[0074] Step S21: Analyze and process the text using preset semantic analysis rules, and output emotional state parameters of the customer during the historical call process;

[0075] The emotional state parameter is used to characterize the intensity of the customer's emotions during historical calls.

[0076] For example, the separated customer audio may be converted into text using ASR (Automatic Speech Recognition) technology.

[0077] Furthermore, the present application inputs the converted customer text into a pre-trained language model, analyzes and processes the text through preset semantic analysis rules, such as customized Prompt instructions, and outputs the customer's emotional state parameters during historical calls, thereby quantifying the customer's emotional state.

[0078] In an optional embodiment, the customer's emotional state parameter can be controlled to be output between 0 and 80 points. The higher the emotional state parameter value, the stronger the customer's emotion, thereby converting the customer's subjective emotions into objective and quantifiable values, and providing an objective and unified analysis and evaluation basis for subsequent analysis of the customer's emotional fluctuations.

[0079] For example, the insured Mr. Zhang called the customer service hotline several times due to issues related to reimbursement of surgical expenses. After obtaining three historical call audios of Mr. Zhang communicating with customer service through the customer service hotline on different dates, the system used voiceprint recognition technology to separate Mr. Zhang's call audio from multiple historical call audios.

[0080] Furthermore, Mr. Zhang's call audio was converted into text through automatic speech recognition technology. The text after the first conversion was "What information do I need to submit?", the text after the second conversion was "I submitted the materials several times but they still said they were incomplete. Why is your process so confusing?", and the text after the third conversion was "Are you kidding? Forget it, just handle it however you want." The three converted customer texts were input into the pre-trained language model, and the text was analyzed and processed through customized Prompt instructions to output the customer's emotional state parameters during the historical calls.

[0081] For example, the text after the first conversion is a normal inquiry with a calm tone, and the output parameter of Mr. Zhang's emotional state in the first historical call is 20 points, which is a normal consultation. The text after the second conversion includes "a few more times", "confusion" and rhetorical questions, which is a complaint and dissatisfaction. Therefore, the output parameter of Mr. Zhang's emotional state in the second historical call is 60 points. The text after the third conversion includes the ironic expression of "joking" and the passive resistance of "deal with it however you want". Although there is no direct complaint or insult, the semantics include complete disappointment with the current service. Therefore, the output parameter of Mr. Zhang's emotional state in the third historical call is 78 points, which is a situation of strong dissatisfaction but not yet erupting.

[0082] As an optional implementation, a word library can be established that can be used to evaluate and output emotional state parameters. The word library contains positive words that indicate normal and calm emotions, and negative words that indicate large emotional fluctuations or dissatisfaction. Each word and tone can be assigned a corresponding score. For example, the positive words "understanding", "good", and "clear" correspond to scores of 0-10 points, the negative words "delay" and "slow" correspond to scores of 10-30 points, and the negative words "dragging", "cheating", and "complaint" correspond to scores of 50-70 points. Of course, the emotional state parameters can also be output in combination with tone symbols such as exclamation marks and question marks.

[0083] It should be noted that this application does not impose any further restrictions on the specific composition of the above-mentioned vocabulary library, and those skilled in the art can adjust it according to actual application scenarios and analysis and evaluation requirements.

[0084] Of course, other methods of outputting the emotional state parameters of the customer during historical calls are also feasible, as long as the intensity of the customer's emotions during historical calls can be quantified. This application does not impose any further restrictions on this.

[0085] Step S3: constructing an emotional state parameter array according to the time sequence relationship, marking the emotionally stable segments in the emotional state parameter array, and constructing a jump relationship table based on the emotionally stable segments;

[0086] As an optional implementation, please refer to Figure 3 , which is a flowchart of constructing a jump relationship table in the customer emotion fluctuation analysis method provided by an embodiment of the present application. The above-mentioned emotionally stable segment in the emotional state parameter array includes:

[0087] Step S30: Calculate the difference between two adjacent emotional state parameters in the marked emotional state parameter array. If the difference is less than a preset fluctuation threshold, mark the current time node and the adjacent time node as emotionally stable segments.

[0088] As an optional implementation manner, the above-mentioned construction of the jump relationship table includes:

[0089] Step S31: constructing the jump relationship table by combining the emotionally stable segment, the customer name, and the time node corresponding to the emotionally stable segment;

[0090] Step S32: Calculate the farthest stable node at each time node under different jump step sizes;

[0091] The jump step size is a power of 2, and the farthest stable node is used to locate the boundary of the longest continuous stable interval.

[0092] In an embodiment of the present application, the emotional state parameters output in step S2 are preferably sorted according to the time nodes corresponding to the calls to form an emotional state parameter array, wherein each element in the emotional state parameter array includes the time node corresponding to the call and the emotional state parameter corresponding to the call, ensuring that there is temporal continuity between the elements in the emotional state parameter array.

[0093] For example, Mr. Zhang made his first call on June 1, and the emotional state parameter corresponding to the first call was 20 points. Mr. Zhang made his second call on June 8, and the emotional state parameter corresponding to the second call was 60 points. Mr. Zhang then made his third call on June 15, and the emotional state parameter corresponding to the third call was 78 points.

[0094] For the sake of convenience, this application marks June 1 as t1, June 8 as t2, and June 15 as t3, and the emotional state parameters are referred to as emotional values, thus forming an emotional state parameter array moods = [(t1, 20), (t2, 60), (t3, 78)]; among them, (t1, 20), (t2, 60) and (t3, 78) are the components of the emotional state parameter array moods, thereby constructing the required emotional state parameter array according to the time series relationship.

[0095] As an optional implementation, after the required emotional state parameter array is constructed, it is also necessary to mark the emotionally stable segments therein and construct a corresponding jump relationship table.

[0096] Specifically, the absolute difference between two adjacent emotion values ​​in the time series relationship in the emotion state parameter array is calculated. If the difference is less than the preset fluctuation threshold, the current node and the adjacent time node are marked as emotion stable segments, that is, the fluctuation of the customer's emotion value between these two time nodes does not exceed the preset fluctuation threshold.

[0097] In the embodiment of the present application, the absolute difference between Mr. Zhang's t1 and t2 is diff1 = |60-20| = 40, and the absolute difference between t2 and t3 is diff2 = |78-60| = 18. The preset fluctuation threshold is limited to 20 points. It can be observed that diff1 is greater than 20 points and diff2 is less than 20 points. Therefore, t2 and t3 can be marked as emotionally stable segments. Since t1 and t3 are not adjacent, the corresponding absolute difference is not calculated.

[0098] Similarly, after marking all the emotionally stable segments in the emotional state parameter array, it is necessary to further construct a jump relationship table by combining the customer name and the time node corresponding to the emotionally stable segment.

[0099] This application uses the customer as the dimension to establish an association between the marked emotionally stable segment and the corresponding time node, forming a jump relationship table containing the customer name, time node and the emotionally stable segment corresponding to the time node, to ensure that Mr. Zhang's emotionally stable segment will not be mistakenly counted as Ms. Li's emotionally stable segment.

[0100] As an optional implementation, after generating the emotional state parameter array moods, this application uses the recursive idea of ​​the doubling algorithm to pre-calculate the farthest stable node of each time node under different jump steps to form a jump relationship table that can achieve fast response and query.

[0101] In an optional embodiment, the present application calculates the farthest stable node that can be reached by jumping at each time node under different powers of 2 steps, and locates the boundary of the longest continuous stable interval required in the end through the farthest stable node.

[0102] Specifically, this application defines a two-dimensional array pa[x][i], where x represents the current time node and i represents the number of jump steps, which is 2. iThe premise for jumping between two time nodes is that the current time node and the adjacent time node have been marked as emotionally stable segments in step S30, that is, the difference between the two emotional state parameters of the current time node and the adjacent time node is less than the preset fluctuation threshold; it can be seen that pa[x][i] means starting from the current time node x, jumping 2 i The time point that can be reached later.

[0103] Furthermore, by taking different values ​​for i, such as 0, 1, 2, etc., we can calculate the current time node in 2 0 =1 step, 2 1 =2 steps, 2 2 = The farthest stable node that can be reached under different jump step sizes such as pa[x][0], for example, pa[x][0] means jumping 2 from the current time node x. 0 = the time node reached after 1 step.

[0104] In order to construct a higher order jump relationship, this application adopts a doubling algorithm to jump 2 i+1 Step is decomposed into two jumps 2 i step, such as pa[x][i+1]=pa[pa[x][i]][i].

[0105] It can be understood that the current time node x jumps 2 i After steps, it reaches pa[x][i], and then jumps 2 from pa[x][i] i Steps, a total of 2 jumps i +2 i =2 i+1 step, that is, pa[pa[x][i]][i] in the corresponding formula, it can be inferred that pa[x][i+1] is x jump 2 i +1 The time node reached after the step.

[0106] In related technologies, a traversal comparison method is usually used to query continuous stable intervals. For example, for an array of emotional state parameters moods containing N time nodes, it is necessary to compare the difference between two emotional state parameters between adjacent time nodes one by one to further determine whether it belongs to an emotionally stable segment.

[0107] For the mood state parameter array moods containing 10 time nodes, 9 difference calculations are required. For the mood state parameter array moods containing 80 time nodes, 99 difference calculations are required, indicating that the query time complexity corresponding to the traversal comparison method is O(N).

[0108] Where N is the number of time nodes, and O() is a standard way to measure time complexity in the field of algorithms, which describes that when the input scale increases, the algorithm running time shows a gradual growth trend.

[0109] In some implementation scenarios, multiple queries may be required, such as M queries. Since the time complexity of a single query is O(N), the time complexity of M queries is O(MN). Large data volumes or a high number of queries can lead to low computational efficiency, especially in real-time analysis or batch processing scenarios. This can affect the timeliness of customer sentiment interventions due to delays.

[0110] In the embodiment of the present application, it is necessary to perform pre-processing calculation to build a jump relationship table, specifically including calculating the current time node in 2 i The farthest stable node that can be reached under different jump step sizes is obtained, thereby reducing the corresponding query time complexity from O(MN) to O(MlogN) through the doubling algorithm. For the sake of convenience, MlogN will be marked as k in the following.

[0111] For example, suppose there are currently 8 time nodes, corresponding to the 1st to the 8th day, and a single query is required in a certain business scenario. At this time, N=8, M=1, log28=3, and the corresponding k values ​​are 0, 1, 2, and 3, and the corresponding jump steps are 2 0 =1,2 1 =2,2 2 =4,2 3 =8.

[0112] Furthermore, it is necessary to calculate the farthest stable node that each time node can reach under four jump steps of 1, 2, 4, and 8. That is to say, 4 jump steps need to be calculated for 8 time nodes, and a total of 32 calculations of the farthest stable node are performed to split the query interval into several power-of-2 steps.

[0113] That is to say, since there are a total of N time nodes, and the jump step length corresponding to each time node is log2N, the total computational amount of the preprocessing process when constructing the jump relationship table is N×log2N. In this way, the construction of the jump relationship table is completed. In traditional technology, no preprocessing process is required, but direct query is performed directly according to demand.

[0114] Step S4: In response to the query requirement, determining the longest continuous stable interval of customer emotions after the designated node through the jump relationship table;

[0115] As an optional implementation, please refer to Figure 4, which is a flow chart for determining the longest continuous stable interval in the customer emotion fluctuation analysis method provided in an embodiment of the present application. The above-mentioned determination of the longest continuous stable interval of customer emotion after a specified node through the jump relationship table includes:

[0116] Step S40: Starting from the designated node, jumping step by step starting from the maximum jump step based on the jump step pre-stored in the jump relationship table;

[0117] Step S41: If the current designated node is still in the pre-marked emotionally stable segment after the jump, the current node is updated to the node after the jump, and the current jump step length is accumulated;

[0118] Step S42: If the current designated node is not within the emotional stability segment after the jump, the current time node is kept unchanged, and the jump step is reduced to continue the step-by-step jump;

[0119] Step S43: Repeat the step-by-step jump until all jump steps cannot be continued, forming the longest continuous stable interval with the designated node as the starting point and the farthest stable node reached by the accumulated jump steps as the end point.

[0120] In an embodiment of the present application, after the jump relationship table is obtained through preprocessing, the following operation steps are required, specifically including responding to a query request, starting from a specified node, and jumping step by step from the maximum jump step based on the jump steps of each order pre-stored in the jump relationship table, thereby ensuring that each jump covers as long a stable interval as possible according to the binary splitting idea, thereby reducing the number of jumps required.

[0121] Furthermore, if the current designated node is still in the pre-marked emotionally stable segment after the jump, the current node is updated to the node after the jump, and the current jump step is accumulated; if the current designated node is not in the emotionally stable segment after the jump, it means that the current jump step is too large, and it is necessary to keep the current time node unchanged, reduce the jump step and continue to jump step by step, and so on, continue to try smaller jump steps until all jump steps can no longer jump, thus forming the longest continuous stable interval with the designated node as the starting point and the farthest stable node reached by the accumulated jump step as the end point.

[0122] In an optional embodiment, 8 time nodes are also provided, corresponding to the 1st to the 8th day, and a single query is required. In this case, N=8, M=1, log28=3, and the values ​​of k are 0, 1, 2, and 3, and the corresponding jump steps are 2 0 =1,2 1 =2,2 2 =4,2 3 =8, and the 3rd to 7th days have been marked as the emotionally stable period in step S30.

[0123] In this embodiment, it is assumed that the specified time node is the 3rd day. Since the maximum jump step corresponding to k=3 is 8, and 3+8=11 is greater than the preset 8 days, it is necessary to jump from the jump step of 4 when k=2, and cannot jump from the jump step of 8.

[0124] Furthermore, if the jump starts from the 3rd day with a jump step number of 4, it will come to the 7th day after the jump, and the 3rd to 7th days are all marked as emotionally stable segments, so the current time node can be updated to the 7th day, and the jump step length of 4 can be accumulated.

[0125] Then, based on the 7th day, a jump is made. When the jump step is 8, 4, and 2, the corresponding time nodes are 15, 11, and 9, respectively, which exceed the preset 8 days. Therefore, it is necessary to jump from k=0 with a jump step of 1. After jumping with a jump step of 1 starting from the 7th day, since the 3rd day to the 7th day are all marked as the emotionally stable segment, the 7th day to the 8th day do not belong to the emotionally stable segment, the current time node needs to be kept unchanged, and it is currently impossible to continue to try a smaller jump step. All jump steps cannot continue to jump and the jump steps cannot be accumulated, thus forming the longest continuous stable interval with the specified node on the 3rd day as the starting point and the farthest stable node on the 7th day reached after the accumulated jump step as the end point. That is to say, the longest continuous stable interval obtained is the 3rd day to the 7th day.

[0126] It should be noted that although the emotionally stable segment has been marked in step S30, the marking here only marks whether the adjacent time nodes belong to the emotionally stable segment. It cannot guarantee whether the adjacent time nodes after the jump belong to the emotionally stable segment. Therefore, it is also necessary to judge whether the time node after the jump and its adjacent time nodes belong to the pre-marked emotionally stable segment, so as to ensure the stability of the entire longest continuous stable interval.

[0127] For example, suppose the 3rd and 4th days are marked as emotionally stable segments, which meets the prerequisite for jumping. However, when the number of jump steps from the 3rd day is 4, the corresponding 7th day after the jump may not belong to the emotionally stable segment between the 6th day and the 7th day. This large-step jump method may span multiple time nodes. Therefore, it is necessary to further determine whether the time node after the jump and its adjacent time nodes belong to the pre-marked emotionally stable segment.

[0128] When batch sentiment analysis is required on the historical call records of a large number of customers, due to the extremely large total number of customers, the corresponding number of queries M will also be very large, and as time accumulates, the number of historical calls of a single customer and the corresponding number of time nodes N will also increase over time. Therefore, when N is small in the initial stage, the pre-processing method of constructing a jump relationship table provided by this application may not be much different from the traditional linear query; but when M and N continue to increase, the time complexity of a single query can be reduced from O(N) to O(logN), and the time complexity of M queries can be reduced from O(MN) to O(MlogN), greatly reducing the amount of calculation.

[0129] In one provided embodiment, when N=10 and M=80, if a traditional linear query method is used, 80×10=800 query operations are required. If a query is performed through the jump relationship table provided in this application, only 80×log210<80×log216=400 query operations are required. Even if the number of preprocessing calculation operations at each time node is added, the number of query operations will not exceed 440, which is much less than the traditional linear query method.

[0130] It is understandable that in actual applications, it may be necessary to perform high-frequency real-time sentiment analysis on thousands or even tens of thousands of customers. Although the number of historical calls for each customer may be small, the number of customers is extremely large, and the corresponding number of queries is also large. As time accumulates, the number of historical calls for a single customer and the corresponding number of time nodes N will also increase accordingly. Therefore, when N increases, the advantages of the doubling algorithm provided in this application will become more obvious. For example, when N = 80, the linear query requires 80 operations, but under the doubling algorithm, only log280, about 7 operations, is required, which is much less than the number of operations required for the linear query, avoiding the subsequent slow response speed due to the increase in number and the problem of inability to respond in real time.

[0131] Step S5: generating customer emotion classification warning information based on the longest continuous stable interval and the corresponding emotional state parameter range;

[0132] As an optional implementation, please refer to Figure 5 , which is a flow chart for generating customer emotion graded warning information in the customer emotion fluctuation analysis method provided in an embodiment of the present application. The above-mentioned method combines the longest continuous stable interval and the corresponding emotional state parameter range to generate customer emotion graded warning information, including:

[0133] Step S50: When the emotional state parameters of the longest continuous stable interval are all lower than a first preset threshold, generating an emotional stability warning;

[0134] Step S51: When the emotional state parameter of the longest continuous stable interval is between a first preset threshold and a second preset threshold, or when the difference between adjacent emotional state parameters within the interval exceeds a preset stability threshold, generating an emotional fluctuation warning;

[0135] Step S52: When the emotional state parameters of the longest continuous stable interval are all higher than the second preset threshold and show a monotonically increasing trend, or when the mutation amplitude of the emotional state parameters at adjacent time nodes exceeds the preset mutation threshold, a mood deterioration warning is generated;

[0136] The first preset threshold is smaller than the second preset threshold.

[0137] In an embodiment of the present application, when all emotional state parameters within the longest continuous stable interval are lower than the first preset threshold, the customer's emotion is judged to be in a long-term stable state. By comparing all emotional state parameters, the impact caused by large fluctuations in individual emotional state parameters is eliminated, thereby avoiding potential risks of missed judgments.

[0138] In an optional embodiment, the present application sets the first preset threshold to 30 points. Taking Ms. Li, a medical health insurance customer, as an example, step S4 determines that Ms. Li's longest continuous stable interval is the last 5 calls, and the corresponding emotion values ​​are 20 points, 25 points, 22 points, 28 points, and 23 points, which are all lower than the first preset threshold of 30 points, thus generating an "emotional stability warning".

[0139] At the same time, the longest continuous stable interval of a certain customer is the past 4 calls, and the corresponding emotional values ​​are 20 points, 25 points, 22 points, and 35 points, respectively. Among them, 35 points is greater than the first preset threshold of 30 points, and the "emotional stability warning" will not be triggered, ensuring the reliability of the warning results.

[0140] Furthermore, when all emotional state parameters within the longest continuous stable interval are between the first preset threshold and the second preset threshold, it indicates that the customer is in a state of moderate dissatisfaction but has not yet exploded.

[0141] Optionally, when the difference between adjacent emotional state parameters within the longest continuous stable interval exceeds a preset stability threshold, it indicates that the customer may currently be in a state of sudden emotional change.

[0142] Therefore, when the emotional state parameter of the longest continuous stable interval is between the first preset threshold and the second preset threshold, or the difference between adjacent emotional state parameters in the interval exceeds the preset stability threshold, an emotional fluctuation warning will be generated.

[0143] In an optional embodiment, the present application sets the first preset threshold value to 30 points and the second preset threshold value to 70 points. In a communication scenario of a financial insurance claim for a certain customer, the longest continuous stable interval determined is the last three calls, and the corresponding emotion values ​​are 45 points, 50 points and 60 points respectively. Since 45 points, 50 points and 60 points are all between 30 points and 70 points, it indicates that the customer is moderately dissatisfied but has not yet erupted, and therefore it will trigger the generation of an emotion fluctuation warning.

[0144] In another optional embodiment, the present application sets the preset stability threshold to 15 points. The longest continuous stability interval determined by a certain customer is also the last three calls, and the corresponding emotion values ​​are 32 points, 50 points and 60 points respectively. Since the difference between the first emotion value and the second emotion value is 18 points, which is greater than the preset stability threshold of 15 points, it indicates that the customer currently has certain emotional changes, and therefore it will also trigger the generation of an emotion fluctuation warning.

[0145] It can be understood that compared with only judging the difference between adjacent emotional state parameters, the present application is more comprehensive by comparing the first preset threshold and the second preset threshold, as well as the difference between adjacent emotional state parameters with the preset stability threshold, and can effectively capture risk signals and intervene in time to reduce the risk of complaints.

[0146] As an optional implementation, when all emotional state parameters within the longest continuous stable interval are above a second preset threshold and show a monotonically increasing trend, it indicates that the customer's mood is continuously deteriorating. Alternatively, when the difference between adjacent emotional state parameters within the longest continuous stable interval exceeds a preset mutation threshold, it indicates that the customer is experiencing a sudden deterioration in mood or a sudden burst of anger.

[0147] Therefore, in an embodiment of the present application, when the emotional state parameters of the longest continuous stable interval are all higher than the second preset threshold and show a monotonically increasing trend, or when the mutation amplitude of the emotional state parameters of adjacent time nodes exceeds the preset mutation threshold, an emotional deterioration warning will be generated.

[0148] In an optional embodiment, the present application limits the second preset threshold to 70 points. If the longest continuous stable interval determined by a customer is the last 4 calls, and the corresponding emotional values ​​are 72 points, 78 points, 85 points and 72 points respectively, since 72 points, 78 points, 85 points and 72 points are all greater than the second preset threshold of 70 points and show a trend of increasing successively, it indicates that the customer is in a state of continuous emotional deterioration, and therefore it will trigger the generation of an emotional deterioration warning.

[0149] In another optional embodiment, the preset mutation threshold is set to 30 points. If a customer's emotion value in the previous call was 60 points, and in the next call the emotion value suddenly increases to 95 points because the materials are returned and not approved, an emotion deterioration warning will be immediately triggered.

[0150] It can be understood that the above-mentioned first preset threshold should be smaller than the second preset threshold, the preset stability threshold should be smaller than the preset fluctuation threshold, and the preset mutation threshold should be smaller than the preset threshold.

[0151] Specifically, this application sets the first preset threshold to 30 points, the second preset threshold to 70 points, the preset stability threshold to 15 points, the preset fluctuation threshold to 20 points, and the preset mutation threshold to 30 points; of course, in different scenario applications, the specific parameters of the above preset thresholds can also be adjusted according to actual needs, and this application does not impose further restrictions on this.

[0152] Since sudden changes in customer emotions in medical health insurance and financial claims scenarios may directly lead to complaints or churn, this application compares all emotional state parameters in the longest continuous stable interval one by one to minimize missed judgments.

[0153] As an optional implementation method, when an emotional stability warning is generated, it means that the customer currently has no obvious negative emotions and is at a low complaint risk. Therefore, there is no need to actively call out for intervention, and follow up according to routine processing; when an emotional fluctuation warning is generated, it means that the customer currently has emotional fluctuations or dissatisfaction, and is at a medium complaint risk. At this time, customer service can actively return the customer, and the customer service specialist will call the customer, actively ask whether there are any service problems, provide personalized explanations, and resolve dissatisfaction caused by information asymmetry; when an emotional deterioration warning is generated, it means that the customer's current mood is continuing to deteriorate or suddenly deteriorating, and is at a high complaint risk. The customer needs to respond in a timely manner, give priority to solving the customer's core demands, or transfer the customer to a senior customer service or complaint handling specialist to avoid complaint escalation or regulatory intervention.

[0154] In this way, by setting up different risk warnings and corresponding intervention strategies, the cost of complaint handling is reduced while the refined operational capabilities of medical health insurance and financial claims services are improved.

[0155] It can be understood that the size of the serial numbers of the steps in the above embodiment does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0156] The customer emotion fluctuation analysis method provided in this application obtains historical call audio and separates the customer audio therein, converts the customer audio into text and performs semantic analysis, and outputs time-series emotion state parameters to provide a data basis for subsequent emotion fluctuation analysis; then constructs an array of emotion state parameters, marks the emotionally stable segments and constructs a jump relationship table, and uses the principle of the doubling algorithm to achieve fast query of the longest continuous stable interval, reducing the time complexity of traditional linear query from O(MN) to O(MlogN), significantly improving the processing efficiency of large-scale data; finally, combining the longest continuous stable interval and the emotion parameter range, different warning information is generated through graded threshold comparison to accurately identify customer emotion risks. This application can objectively and uniformly analyze and evaluate customer emotions, and through the jump relationship table, it reduces the number of queries and improves query efficiency, so as to generate warning information in a timely manner to ensure that customer service personnel in scenarios such as medical health insurance and financial claims can intervene in advance and reduce the risk of complaints.

[0157] Based on the above customer emotion fluctuation analysis method, this application provides a customer emotion fluctuation analysis system, please refer to Figure 6 , is a structural diagram of a customer emotion fluctuation analysis system provided in an embodiment of the present application, wherein the teaching recommendation system includes an acquisition unit, an output unit, a construction unit, a query unit, and an early warning unit;

[0158] The acquisition unit is used to acquire historical call audio and separate the customer audio from the historical call audio;

[0159] An output unit, configured to convert the customer audio into text, perform semantic analysis on the text, and output emotional state parameters of the customer during historical calls;

[0160] a construction unit, configured to construct an emotional state parameter array according to a time sequence relationship, mark emotionally stable segments in the emotional state parameter array, and construct a jump relationship table based on the emotionally stable segments;

[0161] A query unit, configured to respond to a query demand and determine the longest continuous stable interval of customer emotions after a specified node through the jump relationship table;

[0162] The early warning unit is used to generate customer emotion classification early warning information based on the longest continuous stable interval and the corresponding emotional state parameter range.

[0163] As an optional implementation manner, the above-mentioned obtaining historical call audio and separating the customer audio from the historical call audio includes:

[0164] Acquire historical call audio, perform voiceprint feature extraction on the historical call audio, and separate the customer audio from the historical call audio.

[0165] As an optional implementation, the above-mentioned conversion of the customer audio into text, performing semantic analysis on the text, and outputting the customer's emotional state parameters during the historical call process include:

[0166] converting the customer audio into text via automatic speech recognition technology;

[0167] Analyze and process the text using preset semantic analysis rules to output emotional state parameters of the customer during historical calls;

[0168] The emotional state parameter is used to characterize the intensity of the customer's emotions during historical calls.

[0169] As an optional implementation, the emotionally stable segment in the emotional state parameter array is marked as follows:

[0170] The difference between two adjacent emotional state parameters in the marked emotional state parameter array is calculated. If the difference is less than a preset fluctuation threshold, the current time node and the adjacent time node are marked as emotional stable segments.

[0171] As an optional implementation manner, the above-mentioned construction of the jump relationship table includes:

[0172] Constructing the jump relationship table by combining the emotionally stable segment, the customer name, and the time node corresponding to the emotionally stable segment;

[0173] Calculate the farthest stable node at each time node under different jump step sizes;

[0174] The jump step size is a power of 2, and the farthest stable node is used to locate the boundary of the longest continuous stable interval.

[0175] As an optional implementation, the above-mentioned determination of the longest continuous stable interval of customer emotions after a designated node through the jump relationship table includes:

[0176] Starting from the designated node, jumping step by step starting from the maximum jump step based on the jump step sizes of each level pre-stored in the jump relationship table;

[0177] If the current designated node is still in the pre-marked emotional stability segment after the jump, the current node is updated to the node after the jump, and the current jump step is accumulated;

[0178] If the current designated node is not within the emotional stability segment after the jump, the current time node is kept unchanged, and the jump step is reduced to continue the step-by-step jump;

[0179] Repeat the step-by-step jump until all jump steps cannot continue to jump, forming the longest continuous stable interval with the specified node as the starting point and the farthest stable node reached by the accumulated jump steps as the end point.

[0180] As an optional implementation, the above-mentioned combination of the longest continuous stable interval and the corresponding emotional state parameter range to generate customer emotion classification warning information includes:

[0181] When the emotional state parameters of the longest continuous stable interval are all lower than a first preset threshold, generating an emotional stability warning;

[0182] When the emotional state parameter of the longest continuous stable interval is between the first preset threshold and the second preset threshold, or the difference between adjacent emotional state parameters in the interval exceeds the preset stability threshold, an emotional fluctuation warning is generated;

[0183] When the emotional state parameters of the longest continuous stable interval are all higher than the second preset threshold and show a monotonically increasing trend, or the mutation amplitude of the emotional state parameters at adjacent time nodes exceeds the preset mutation threshold, a mood deterioration warning is generated;

[0184] The first preset threshold is smaller than the second preset threshold.

[0185] For other details about the technical solutions for implementing each unit in the customer emotion fluctuation analysis system provided in the above embodiment, please refer to the description of the customer emotion fluctuation analysis method in the above embodiment, which will not be repeated here.

[0186] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similarities between the various embodiments can be referred to in conjunction with each other. For system-related embodiments, since they are generally similar to method-related embodiments, their description is relatively simple. For relevant details, refer to the description of the method-related embodiments.

[0187] Please refer to Figure 7 , is a structural diagram of a computer device provided in an embodiment of the present application, wherein the computer device 70 includes a processor 71 and a memory 72 coupled to the processor 71.

[0188] The memory 72 stores a computer program. When the computer program is executed by the processor 71, the processor 71 executes the steps of the customer emotion fluctuation analysis method in the above embodiment.

[0189] The processor 71 may also be referred to as a CPU (Central Processing Unit). The processor 71 may be an integrated circuit chip having signal processing capabilities. The processor 71 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor.

[0190] Please refer to Figure 8 , a structural diagram of the storage medium provided in an embodiment of the present application, the computer-readable storage medium of the embodiment of the present application stores a computer program 80, and the computer program 80 is executed by a processor to implement the artificial intelligence-based actuarial classification method in the above embodiment, wherein the computer program 80 can be stored in the above storage medium in the form of a software product, including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a computer device such as a computer, a server, a mobile phone, or a tablet. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDNs), and big data and artificial intelligence platforms.

[0191] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0192] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit. The above is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

[0193] The above embodiments are merely exemplary embodiments for illustrating the principles of the embodiments of the present application, but the embodiments of the present application are not limited thereto. Those skilled in the art may make various modifications and improvements without departing from the spirit and essence of the embodiments of the present application, and such modifications and improvements are also considered to be within the scope of protection of the embodiments of the present application.

Claims

1. A method for analyzing customer emotional fluctuations, characterized in that: The following steps are involved: Acquire historical call audio, and separate the customer audio from the historical call audio; Convert the customer audio into text, perform semantic analysis on the text, and output emotional state parameters of the customer during the historical call; constructing an emotional state parameter array according to a time sequence relationship, marking emotionally stable segments in the emotional state parameter array, and constructing a jump relationship table based on the emotionally stable segments; In response to the query requirement, determining the longest continuous stable interval of customer emotions after the designated node through the jump relationship table; The customer emotion classification warning information is generated by combining the longest continuous stable interval and the corresponding emotional state parameter range.

2. The method for analyzing customer emotional fluctuations according to claim 1, wherein: The acquiring of historical call audio and separating the customer audio from the historical call audio includes: Acquire historical call audio, perform voiceprint feature extraction on the historical call audio, and separate the customer audio from the historical call audio.

3. The method for analyzing customer emotional fluctuations according to claim 1, wherein: The step of converting the customer audio into text, performing semantic analysis on the text, and outputting the customer's emotional state parameters during the historical call process includes: converting the customer audio into text via automatic speech recognition technology; Analyze and process the text using preset semantic analysis rules to output emotional state parameters of the customer during historical calls; The emotional state parameter is used to characterize the intensity of the customer's emotions during historical calls.

4. The method for analyzing customer emotional fluctuations according to claim 1, wherein: The emotionally stable segment in the emotional state parameter array includes: The difference between two adjacent emotional state parameters in the marked emotional state parameter array is calculated. If the difference is less than a preset fluctuation threshold, the current time node and the adjacent time node are marked as emotional stable segments.

5. The method for analyzing customer emotion fluctuations according to claim 4, wherein: The step of constructing the jump relationship table includes: Constructing the jump relationship table by combining the emotionally stable segment, the customer name, and the time node corresponding to the emotionally stable segment; Calculate the farthest stable node at each time node under different jump step sizes; The jump step size is a power of 2, and the farthest stable node is used to locate the boundary of the longest continuous stable interval.

6. The method for analyzing customer emotion fluctuations according to claim 5, wherein: The longest continuous stable interval of customer emotions after the designated node is determined by the jump relationship table, including: Starting from the designated node, jumping step by step starting from the maximum jump step based on the jump step sizes of each level pre-stored in the jump relationship table; If the current designated node is still in the pre-marked emotional stability segment after the jump, the current node is updated to the node after the jump, and the current jump step is accumulated; If the current designated node is not within the emotional stability segment after the jump, the current time node is kept unchanged, and the jump step is reduced to continue the step-by-step jump; Repeat the step-by-step jump until all jump steps cannot continue to jump, forming the longest continuous stable interval with the specified node as the starting point and the farthest stable node reached by the accumulated jump steps as the end point.

7. The method for analyzing customer emotional fluctuations according to claim 1, wherein: The generation of customer emotion classification warning information by combining the longest continuous stable interval and the corresponding emotional state parameter range includes: When the emotional state parameters of the longest continuous stable interval are all lower than a first preset threshold, generating an emotional stability warning; When the emotional state parameter of the longest continuous stable interval is between the first preset threshold and the second preset threshold, or the difference between adjacent emotional state parameters in the interval exceeds the preset stability threshold, an emotional fluctuation warning is generated; When the emotional state parameters of the longest continuous stable interval are all higher than the second preset threshold and show a monotonically increasing trend, or the mutation amplitude of the emotional state parameters at adjacent time nodes exceeds the preset mutation threshold, a mood deterioration warning is generated; The first preset threshold is smaller than the second preset threshold.

8. A customer emotion fluctuation analysis system, characterized in that: include: an acquisition unit, configured to acquire historical call audio and separate the customer audio from the historical call audio; An output unit, configured to convert the customer audio into text, perform semantic analysis on the text, and output emotional state parameters of the customer during historical calls; a construction unit, configured to construct an emotional state parameter array according to a time sequence relationship, mark emotionally stable segments in the emotional state parameter array, and construct a jump relationship table based on the emotionally stable segments; A query unit, configured to respond to a query demand and determine the longest continuous stable interval of customer emotions after a specified node through the jump relationship table; The early warning unit is used to generate customer emotion classification early warning information based on the longest continuous stable interval and the corresponding emotional state parameter range.

9. A computer device, characterized in that: The computer device includes a processor and a memory coupled to the processor, wherein a computing program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the customer emotion fluctuation analysis method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the customer emotion fluctuation analysis method according to any one of claims 1 to 7.

Images