Data processing method and device, equipment, storage medium and program product

Abstract

The invention discloses a data processing method and device, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence. Obtaining service session data; determining N service links and starting and ending time of each service link through a large language model according to the service session data; according to the starting and ending time of each service link, determining a time-consuming service link in the N service links, the service duration of the time-consuming service link being greater than or equal to a preset service duration; according to stage session data corresponding to the time-consuming service link in the service session data, time-consuming root cause data is determined, and the time-consuming root cause data comprises at least one of the following items: a specialist factor and a non-specialist factor. Therefore, the problems that traditional manual analysis depends on experience judgment and the root cause is fuzzy are effectively solved, and accurate positioning and root cause tracing of the customer service call duration problem can be realized through intelligent analysis of the service session data by the large language model, so that the requirements of improving the service efficiency and the service quality are met.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a data processing method, apparatus, device, storage medium and program product. Background Technology

[0002] In a customer service system, call duration is a key indicator for measuring the efficiency and service quality of customer service specialists. Excessively long call durations often reflect insufficient knowledge of the service provided by the specialist, a lack of communication skills, or cumbersome service processes, which in turn affects customer satisfaction.

[0003] As service scale expands and user needs diversify, the complexity of user issues has increased significantly, leading to a continuous rise in the online service time of customer service specialists. However, traditional manual sampling and analysis methods are inefficient and heavily influenced by subjective human factors, making it difficult to accurately determine the root cause of the increase in customer service call duration, thus failing to meet the needs of improving service efficiency and quality. Summary of the Invention

[0004] This application provides a data processing method, apparatus, device, storage medium, and program product that can accurately determine the root cause of increased customer service call duration, thereby meeting the need to improve service efficiency and quality.

[0005] In a first aspect, embodiments of this application provide a data processing method, including: Retrieve service session data; Using a large language model, based on service session data, determine N service steps and the start and end times of each service step, where N is an integer greater than or equal to 1; Based on the start and end times of each service step, determine the time-consuming service steps among the N service steps. The service duration of the time-consuming service steps is greater than or equal to the preset service duration. Based on the phase session data corresponding to the time-consuming service steps in the service session data, determine the root cause data of the time consumption. The root cause data of the time consumption includes at least one of the following: specialist factors and non-specialist factors.

[0006] Secondly, embodiments of this application provide a data processing apparatus, including: The acquisition module is used to acquire service session data; The determination module is used to determine N service steps and the start and end times of each service step based on service session data using a large language model, where N is an integer greater than or equal to 1; The determination module is also used to determine the time-consuming service links among the N service links based on the start and end times of each service link, wherein the service duration of the time-consuming service links is greater than or equal to the preset service duration. The determination module is also used to determine the root cause data of time consumption based on the stage session data corresponding to the time-consuming service link in the service session data. The root cause data of time consumption includes at least one of the following: specialist factors and non-specialist factors.

[0007] Thirdly, embodiments of this application provide a computer device, which includes: a processor and a memory storing computer program instructions; When the processor executes computer program instructions, it implements the data processing method as described in the first aspect.

[0008] Fourthly, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the data processing method as described in the first aspect.

[0009] Fifthly, embodiments of this application provide a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the data processing method as shown in the first aspect.

[0010] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the data processing method as described in the first aspect.

[0011] The data processing method, apparatus, device, storage medium, and program product of this application embodiment can acquire service session data and, through a large language model, determine N service steps and the start and end times of each service step based on the service session data; determine time-consuming service steps among the N service steps based on the start and end times of each service step, wherein the service duration of the time-consuming service step is greater than or equal to a preset service duration; and determine time-consuming root cause data based on the stage session data corresponding to the time-consuming service steps in the service session data, wherein the time-consuming root cause data includes at least one of the following: specialist factors and non-specialist factors. Thus, based on a large language model combined with prompt word engineering and zero-shot learning technology, N service steps and the start and end times of each step can be automatically identified without manual annotation, avoiding the subjectivity and inefficiency of traditional manual sampling. Furthermore, by calculating the service duration based on the start and end times of each step and automatically filtering time-consuming service steps through a preset service duration, normal and abnormal time consumption can be quantitatively distinguished, overcoming the deficiency of traditional overall statistical analysis in being unable to locate specific bottlenecks. Furthermore, by using a large language model to perform semantic analysis on the phased conversation data of time-consuming processes, the system automatically extracts the root cause data of time consumption and accurately distinguishes between specialist factors such as service unfamiliarity and redundant communication, and non-specialist factors such as slow response and cumbersome processes. This solves the problems of traditional manual analysis relying on experience judgment and vague root causes. It can provide quantitative basis for service process optimization, specialist training and platform iteration through intelligent analysis of service conversation data using a large language model. It realizes the accurate positioning and root cause tracing of customer service call duration issues to meet the needs of improving service efficiency and service quality. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a data processing method according to an embodiment of the data processing method provided in this application; Figure 2 This is a schematic flowchart of a data processing method according to an embodiment of the data processing method provided in this application; Figure 3 This is a schematic diagram of the customer complaint process of a data processing method according to an embodiment of the data processing method provided in this application; Figure 4 This is a schematic diagram of the structure of a data processing apparatus provided in one embodiment of this application; Figure 5 This is a schematic diagram of the structure of a data processing device provided in one embodiment of this application. Detailed Implementation

[0014] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0015] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0016] The acquisition, storage, use, and processing of data (including but not limited to features and information mentioned in this document) in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0017] In related technologies, determining the root cause of increased customer service call duration can be achieved by setting call duration and post-call duration thresholds for each service type handled by customer service specialists at each job level based on historical call data. When the call duration or post-call duration of a customer service specialist at the corresponding job level is detected to be greater than or less than the set threshold, an alarm or notification is triggered, and the alarm or notification work order and corresponding recording are saved. However, this solution only performs simple statistical analysis of basic customer service call duration data, such as mean, maximum, and minimum values, considering only a single dimension of customer service call duration. It cannot delve into the root causes affecting call duration, nor can it effectively combine specialist service performance, service type, frequently asked customer complaints, and service satisfaction data to construct causal relationships and influence paths, or intuitively analyze the dynamic changes in call duration and influencing factors over different time periods. For example, at the specialist level, it cannot analyze the root causes of specialist time consumption and provide targeted training for specialists in service skills, knowledge retrieval, and platform queries; or at the service process level, it cannot analyze the increase in time consumption of different service processes and optimize the service processes accordingly. As a result, with the expansion of service scale and the diversification of user needs, it is difficult to accurately determine the root cause of the increase in customer service call duration, thus failing to meet the needs of improving service efficiency and service quality.

[0018] To address this pain point, this application provides a data processing method, apparatus, device, storage medium, and program product. Specifically, it proposes a zero-sample intelligent analysis method and apparatus for the entire customer service call process based on a large language model. This addresses the shortcomings of existing customer service call duration analysis methods, and the analysis results provide a basis for optimizing user service processes and providing targeted training for customer service specialists, effectively improving user service levels and operational efficiency.

[0019] The following will be combined with the appendix Figures 1 to 5 This application describes in detail the data processing methods, apparatus, computer equipment, storage media, and program products of the embodiments thereof. It should be noted that these embodiments are not intended to limit the scope of this application.

[0020] The following is combined with Figure 1 The data processing method provided in the embodiments of this application will be described in detail.

[0021] Figure 1 This is a flowchart of a data processing method provided in an embodiment of this application.

[0022] like Figure 1 As shown, this data processing method can be applied to computer equipment, and may specifically include: Step 110: Obtain service session data; Step 120: Using a large language model, determine N service steps and the start and end times of each service step based on the service session data, where N is an integer greater than or equal to 1; Step 130: Based on the start and end times of each service step, determine the time-consuming service steps among the N service steps, where the service duration of the time-consuming service steps is greater than or equal to the preset service duration; Step 140: Based on the stage session data corresponding to the time-consuming service steps in the service session data, determine the root cause data of the time consumption, which includes at least one of the following: specialist factors and non-specialist factors.

[0023] In this embodiment of the application, taking the old-for-new trade-in as an example, steps 110 to 140 are illustrated as follows.

[0024] Step 110: In the trade-in service, service session data comes from the call records between customer service specialists and users. For example, if a user inquires about the "application process for participating in the home appliance trade-in subsidy," this call record is stored in the database, containing the session ID, timestamp, speaker identity (specialist / user), and dialogue content, such as "User: I bought a refrigerator and want to apply for the trade-in subsidy, how do I do it?" "Specialist: You need to provide photos of your ID card, bank card, and purchase invoice..." etc. It also includes the time each message was sent, for example, if the user asked the question at 10:00:00, the specialist replied at 10:00:15, etc.

[0025] Step 120: Analyze the above conversation data using a Large Language Model (LLM). Combined with the trade-in service process, abstract four service stages and determine the start and end times of each stage based on timestamps: P1: Identifying and confirming user requests (10:00:00 - 10:01:30); The user states they want to purchase a refrigerator and apply for a trade-in subsidy. The specialist further inquires about the refrigerator model, purchase time, etc., to clarify the user's core needs. P2: Collecting service elements (10:01:31 - 10:05:20); The specialist guides the user to provide information such as their ID card, bank card photos, purchase invoice, and sales company name, corresponding to the requirements in Reference 8 that "purchase information includes sales company, invoice, and price." P3: Customer service system query (10:05:21 - 10:09:10); The specialist logs into the subsidy application system, enters user information to query subsidy eligibility and amount. During this process, system lag occurs, requiring repeated verification of invoice information. P4: Q&A Session (10:09:11 - 10:11:00) The specialist informed the user that the subsidy amount was 300 yuan and explained how to check the review progress. The user could click "View Information" to check the review progress of all current forms.

[0026] Step 130: Obtain the preset service duration for each service stage. This preset service duration can be based on the average time consumption statistics of each service stage of the trade-in service, such as P1≤2 minutes, P2≤4 minutes, P3≤3 minutes, P4≤2 minutes. Then, calculate the actual duration of each service stage based on its start and end times. Finally, determine whether the actual duration exceeds the preset service duration to identify time-consuming service nodes, for example: P1: 1 minute 30 seconds (normal), P2: 3 minutes 49 seconds (normal), P3: 3 minutes 49 seconds (exceeds the threshold of 3 minutes, determined to be a time-consuming service stage), P4: 1 minute 49 seconds (normal).

[0027] Step 140: Analyze the session data corresponding to the P3 service stage, such as the session excerpt: "Specialist: The system is a bit slow, please wait... (2-minute pause) Sorry, invoice verification needs to be retried, the system displays 'Deduplication failed'...". Non-specialist factors: The subsidy application system is experiencing lag, and the invoice verification / deduplication function is malfunctioning, consistent with the technical process issue of "uploading invoices automatically verifying applicant information, invoice verification, and deduplication." Specialist factors: There is no obvious lack of service familiarity (the specialist accurately guided the information filling requirements). Therefore, the root cause of the time consumption data is non-specialist factors, especially service system factors.

[0028] In this way, based on a large language model combined with prompt word engineering and zero-shot learning technology, N service steps and their start and end times can be automatically identified without manual annotation. This avoids the subjectivity and inefficiency of traditional manual sampling. Furthermore, service duration is calculated based on the start and end times of each step, and time-consuming service steps are automatically filtered through preset service durations, quantitatively distinguishing between normal and abnormal time consumption. This overcomes the shortcomings of traditional overall statistical analysis, which cannot pinpoint specific bottlenecks. Moreover, semantic analysis of the stage conversation data of time-consuming steps using the large language model automatically extracts the root cause data of time consumption, accurately distinguishing between specialist factors such as service unfamiliarity and redundant communication, and non-specialist factors such as slow response and cumbersome processes. This solves the problems of traditional manual analysis relying on experience judgment and unclear root causes. Intelligent analysis of service conversation data through the large language model provides quantitative basis for service process optimization, specialist training, and platform iteration, achieving accurate positioning and root cause tracing of customer service call duration issues to meet the needs of improving service efficiency and quality.

[0029] The steps described above are explained in detail below.

[0030] First, regarding step 110, the analysis process first requires effective processing of the massive amount of call data in the service session database. This call data is stored in the service session database in the form of session text, with the daily data volume reaching hundreds of thousands of records. To ensure the efficiency and accuracy of data processing, this embodiment of the application adopts a two-stage query processing step, that is, firstly, a pre-query is used to obtain the unique identifier of the session text, namely the session ID, and then the corresponding complete session text is extracted based on these identifiers.

[0031] Based on this, in some embodiments of this application, a pre-query method can be used to filter service session data from the massive amount of data in the service session database. That is, in the massive call data, the unique identifier of the session, such as the session ID, is quickly located and obtained through filtering conditions, rather than directly extracting the complete session record. Based on this, before step 110, the data processing method may also include steps 1501 to 1503, as shown below.

[0032] Step 1501: Obtain service session sampling requirements.

[0033] For example, if it is necessary to analyze customer service calls for the trade-in service, the service session sampling requirement can be "service type = trade-in" and "call duration = the last 7 days".

[0034] Step 1502: Based on the service session sampling requirements, retrieve M service session identifiers related to the service session sampling requirements from the service session database, where M is an integer greater than or equal to 1.

[0035] For example, based on "service type = trade-in" and "call duration = last 7 days", a list of session IDs that meet the service session sampling requirements is queried from the service session database. The list of session IDs includes M service session identifiers.

[0036] Step 1503: The candidate service session data corresponding to each service session identifier is determined as the service session data.

[0037] For example, the corresponding complete session text can be extracted based on these IDs, and these session texts can be identified as service session data.

[0038] Therefore, the aforementioned pre-query is an efficient data filtering mechanism that first locates the ID and then extracts the conversation text. Its core objective is to accurately and quickly identify the analysis target from massive amounts of data, reducing subsequent processing costs. In this case, daily call data can reach hundreds of thousands of records. Directly extracting all complete conversation records would consume significant computing resources and time. Pre-query narrows the scope, obtaining only the conversation IDs that match the analysis, and then extracting the corresponding complete records based on the IDs, reducing the processing of invalid data and improving data processing efficiency. Furthermore, pre-query rules can initially filter out valid conversation IDs, avoiding noise during subsequent complete data extraction, laying the foundation for subsequent cleaning and analysis, and effectively ensuring data accuracy.

[0039] In the embodiments of this application, after obtaining the conversation text as described above, the conversation data can be cleaned using a rule engine to remove noise data such as invalid dialogues and abnormal records. The call records are labeled with the speaker's identity, such as specialist or user, to differentiate roles in subsequent analysis. Simultaneously, the conversation content is timestamped to record the sending time of each message, providing a basis for subsequent time-consuming analysis. Finally, the scattered conversation records are sorted chronologically to generate complete conversation text for subsequent analysis.

[0040] The service session data in this embodiment includes session text arranged in chronological order, and the session text is marked with the identity identifier and timestamp of the session dialogue.

[0041] For example, taking the "trade-in service" as an example, the user conversation text in the service conversation data is illustrated below. The conversation involves customer service representative S and user U, as shown in Table 1 below.

[0042] Table 1

[0043] Therefore, the above process can be used to remove noisy data such as invalid dialogues and abnormal records, label the conversation identity and timestamp, and generate complete conversation text in chronological order, providing structured input for subsequent analysis. This overcomes the problems of data dispersion and noise interference in traditional methods and improves the accuracy of analysis.

[0044] Secondly, regarding step 120, the N service steps in this application embodiment include at least one of the following: identifying and confirming user requests, collecting service elements, querying customer service systems, and answering user questions.

[0045] In some embodiments of this application, the service session data includes session text arranged in chronological order, and the session text is marked with a timestamp. Based on this, the service segments can be divided through the following steps, which specifically may include steps 1201 to 1203, as shown below.

[0046] Step 1201: Input the service session data and the prompts from the expandable link template library into the large language model. The large language model then identifies the service session data according to the prompts from the expandable link template library to obtain N service links of the service session data. The expandable link template library is determined by the expandable core links in the service session flow and the link characteristics of the expandable core links.

[0047] In this embodiment, the expandable process template library may include the following four expandable core processes: identifying and confirming user requests, collecting service elements, querying the customer service system, and answering user questions. Based on this, its expandable core processes consist of N service processes. Thus, taking the service processes including identifying and confirming user requests, collecting service elements, querying the customer service system, and answering user questions as an example, the prompts for the expandable process template library could include the following: Please analyze the following service session data, identify the segments in the service session data that belong to the following expandable core processes, and mark the start and end timestamps of each service process.

[0048] Among them, the step of identifying and confirming user requests (P1) involves confirming the user's core needs, the scope of the problem, or the urgency level.

[0049] Service element collection step P2: Collect key information required for the query, such as card number, amount, and date.

[0050] Customer service system query process P3: The process by which a specialist retrieves information from the specialist service system factors.

[0051] P4: Answering user questions: Providing answers or solutions to user requests.

[0052] Step 1202: Based on each service step, obtain the start and end session text of each service step from the service session data.

[0053] For example, the initial conversation text is: the text in which the user first raises their core need, triggering the start of the process. Extraction result: timestamp 09:00:05, User: "Hello, I'd like to inquire about the rules of the trade-in program, especially how to trade in a mobile phone?" The ending conversation text is: the text in which both parties confirm that their needs have been agreed upon, marking the end of the process. Extraction result: timestamp 09:00:35, User: "Yes, that's right."

[0054] Step 1203: Determine the start and end times of each service step based on the timestamps of the start and end timestamps of the session text for each service step.

[0055] For example, the timestamps are obtained as follows: Start timestamp: 09:00:05, which is the timestamp of the start of the session text; End timestamp: 09:00:35, which is the timestamp of the end of the session text. The start and end times of the service step are calculated: Start and end time difference: 09:00:35 - 09:00:05 = 30 seconds. Therefore, the start and end times of the user request confirmation step are 09:00:05-09:00:35, with a duration of 30 seconds.

[0056] In this embodiment, the expandable stage template library and its prompts can be determined through the following steps. The expandable stage template library can be abstracted into a template library L = {Identifying and confirming user requests, collecting service elements, querying customer service systems, answering user questions, ...}. This library supports dynamic addition, deletion, and priority reordering, and can segment stages using zero-shot prompts without manual annotation, eliminating the need for model training with large amounts of labeled data. The prompts for the expandable stage template library are composed of prompts for multiple service stages within the service session scenario. These prompts can include task descriptions, format specifications, and clearly define the model's analysis objectives and output format. Specifically, the task description section requires the large language model to identify key stages in the service session data. This involves using the large language model to determine the bottleneck stages in the four stages: identifying and confirming user requests (P1), collecting service elements (P2), querying customer service systems (P3), and answering user questions (P4), analyzing the time consumption of each stage, and automatically generating targeted optimization suggestions.

[0057] Based on this, the data processing method may further include steps 1601 to 1604, as detailed below.

[0058] Step 1601: Abstract the service session process of the service session scenario into at least two extensible core links.

[0059] For example, the service session process in a service session scenario can be abstracted into four stages: identifying and confirming user requests (P1), collecting service elements (P2), querying the customer service system (P3), and answering user questions (P4).

[0060] Step 1602: Determine the link characteristics of each expandable core link.

[0061] For example, taking the above four expandable core links as examples, they can be the user request identification and confirmation link P1, the service element collection link P2, the customer service system query link P3, and the user question answering link P4.

[0062] Step 1603: Based on each expandable core link and the link characteristics of each expandable core link, construct an expandable link template library; the expandable link template library includes P link data, each link data includes an expandable core link and the link characteristics of the expandable core link, where P is an integer greater than or equal to 1.

[0063] For example, taking the above four expandable core links as an example, the expandable link template library can include four links of data, such as the link data of the user request identification and confirmation link P1, the link data of the service element collection link P2, the link data of the customer service system query link P3, and the link data of the user question answering link P4.

[0064] Specifically, one data point includes the user request identification and confirmation step P1; the characteristics of the core steps can be expanded: confirming the user's core needs, identifying the scope, urgency and expected resolution of the customer complaint, etc., and the start and end times are t1_start and t1_end.

[0065] The two data points include the service element collection step P2; the characteristics of the core step that can be extended are: collecting key elements required for transaction query (such as amount, card number, transaction date, etc.) so as to query transaction information on the platform and locate user problems, with the start and end times recorded as t2_start and t2_end.

[0066] The three data points include the customer service system query step P3; the characteristics of the extended core steps are: retrieving and querying user transaction information in the service system to provide a basis for subsequent answers, with the start and end times recorded as t3_start and t3_end.

[0067] The four data points include the user question answering stage P4; the characteristics of the extended core stage are: providing accurate and actionable answers to the user's core needs, with start and end times recorded as t4_start and t4_end.

[0068] Step 1604: Configure corresponding prompt words for each piece of data, and use the prompt words corresponding to each piece of data as prompt words for the expandable process template library.

[0069] It should be noted that each piece of data in this application embodiment can also be fused from multiple sources. For example, the process data M = {service type, caller region, satisfaction rating, service activity ID, service specialist ID, total communication duration} can be integrated to form a comprehensive service quality assessment.

[0070] Secondly, step 130 allows for drill-down analysis of the service session data, identifying efficiency differences among different specialists and service stages when serving users. This reveals individual specialist issues causing performance deviations and the root causes of time-consuming processes in different service stages. For example, some specialists may be less efficient in handling specific service stages or experience excessive time consumption in certain stages. This analysis effectively identifies employees requiring enhanced training, enabling targeted instruction and ultimately improving the overall service level of the team.

[0071] Based on this, in some embodiments of this application, a service step corresponds to a preset service duration. Based on this, step 130 may specifically include steps 1301 to 1303, as shown below.

[0072] Step 1301: Determine the service duration of each service step based on the start and end times of each service step.

[0073] Step 1302: Compare the service duration of each service step with the preset service duration corresponding to each service step.

[0074] Step 1303: If the service duration of each service step is greater than or equal to the preset service duration corresponding to each service step, then each service step is identified as a time-consuming service step.

[0075] For example, the preset service duration can be based on the average time consumption statistics of each service stage of the trade-in service, such as P1≤2 minutes, P2≤4 minutes, P3≤3 minutes, P4≤2 minutes. The actual duration of each service stage is calculated based on its start and end times. Then, based on the preset service duration of each service stage, it is determined whether the actual duration has exceeded the time limit to identify time-consuming service nodes. For example, P1: 1 minute 30 seconds (normal), P2: 3 minutes 49 seconds (normal), P3: 3 minutes 49 seconds (exceeds the threshold of 3 minutes, determined to be a time-consuming service stage), P4: 1 minute 49 seconds (normal).

[0076] In this embodiment of the application, to ensure the scientific validity of subsequent analysis, a normal user service time, i.e., a preset service duration, is set for each service step. Based on this, the data processing method may further include steps 1701 and 1702, which can be used to determine the preset service duration through the following steps, as detailed below.

[0077] Step 1701 is used to statistically analyze the service duration of each scalable core link in the scalable link template library within a preset time period. The statistical value may include at least one of the following: mean or standard deviation.

[0078] In this embodiment, the baseline value can be determined through mean calculation and / or threshold determination. The baseline value calculation involves analyzing the duration of each service stage within the same service scenario to calculate the average time spent on each stage. The threshold determination involves performing adaptive threshold calculation on the stage duration sequence of all calls within the same scenario in each statistical period. Specifically, the standard deviation of the time spent on each service stage can be calculated based on the mean μ of the time spent on each service stage. ; with μ + k· The preset service duration T is used, where k is an adjustable coefficient, with a default value of k=2. If the service process takes longer than the threshold T, it is marked as a time-consuming service process, and a root cause analysis of the service anomaly is performed.

[0079] Step 1702: Use the statistical value of the service duration of the expandable core links as the preset service duration corresponding to each service link.

[0080] In this embodiment of the application, taking the average statistical value as an example, based on the preset service duration T, further stage-level splitting and service type aggregation are performed to construct a matrix of service links-preset service durations. The specific steps are as follows. The set of service links B = {b1,b2,…,bn}, for any service link bj∈B, the average duration of each service link in the last N calls can be calculated using the following formula (1):

[0081] Based on this, the matrix of serviceable links and preset service durations is... .

[0082] It should be noted that the matrix M of service links and preset service durations can be normalized column-wise to obtain M′. If any element in M′ ( If it is a constant, such as 0.5), then the determination stage... For service nodes The bottleneck stage is determined by calling the root cause database R based on the bottleneck stage. If P1 is the bottleneck, it may be a specialist factor, which can be optimized through information pre-collection strategies, such as summarizing the user's previous conversation content using a large language model to reduce the number of times specialists repeatedly ask users questions. If P2 is the bottleneck, it may be a non-specialist factor, specifically a user service system factor, which can be optimized through element pre-collection, such as automatically collecting some information through APP pop-ups or IVR before the next call to reduce manual inquiries. If P3 is the bottleneck, it may be a non-specialist factor, specifically a specialist service system factor within the service system factors, which can be optimized through the query platform, providing specialists with a unified query platform to shorten platform query time. If P4 is the bottleneck, it may be a non-specialist factor, specifically a service product factor within the service system factors, which can be optimized through knowledge enhancement, pushing FAQ summaries or script prompts generated by a large language model to specialists to reduce the time specialists spend explaining to users.

[0083] In step 140, in this embodiment of the application, for cases where the threshold of each stage is exceeded, the stage session data corresponding to the time-consuming service link can be extracted as the object of further analysis. The LLM large language model can be called to perform secondary semantic clustering analysis to identify whether there are unreasonable service process design, poor information transmission, low platform query efficiency or other potential reasons for excessive time consumption in the stage session data. The output time consumption root cause data may include, but is not limited to, at least one of the following: the time consumption root cause label set includes specialist factors, non-specialist factors, specialist service system factors, user service system factors, and service product factors related to specialist service unfamiliarity.

[0084] Based on this, in some embodiments of this application, step 140 may specifically include: The phased conversation data corresponding to the time-consuming service steps and root cause extraction prompts are input into a large language model. The large language model performs semantic clustering analysis on the phased conversation data according to the root cause extraction prompts, and obtains the time-consuming root cause data of the time-consuming service steps output by the large language model. In this embodiment, the root cause extraction prompts can be instructional text that guides the large language model to perform semantic clustering analysis on the phased conversation data. This is used to clarify the analysis objectives, such as extracting the time-consuming root causes of the time-consuming service steps, standardizing the analysis logic such as breaking down the causes from dimensions such as hardware / software / operation habits / external environment, and constraining the output format such as requiring the root causes to be sorted out according to the "cause → result" logic and presented in points, so that the model can accurately focus on the deep-seated causes of the time-consuming steps.

[0085] In this embodiment of the application, the root cause extraction prompt words can be as follows: "Please perform semantic clustering analysis on the session data for the following time-consuming service steps to extract the root causes of the time consumption in each step. The analysis should break down the causes from dimensions such as hardware, software, operating habits, and external environment, and output a clear list of root causes. Each root cause should be associated with key information in the session data."

[0086] Alternatively, "Analyze the root causes of time consumption during the system query phase of customer service calls related to the trade-in program. This requires considering clues such as system response delays, ambiguous user information, and complex processes. Break down the causes from three dimensions: specialist service system factors, user operations, and service product factors, and output segmented root cause data for each stage of the time consumption."

[0087] In this way, after inputting the session data of time-consuming service steps and root cause extraction prompts into the large language model, the model will perform cluster analysis on the semantics and contextual relationships of the session text based on the instruction logic of the prompts, and finally output the root cause data of the time-consuming steps. Therefore, based on the output of the large language model, targeted training plans can be developed for problems caused by service unfamiliarity, improving the professional skills and operational efficiency of customer service personnel; for excessive time consumption caused by technical issues, the platform's query efficiency can be improved, and automated tools or intelligent auxiliary systems can be introduced to reduce redundant manual operations and improve overall service efficiency.

[0088] It should be noted that the data processing method provided in this application embodiment, compared with traditional solutions, utilizes the reasoning and analysis capabilities of a large language model. It does not rely on a large dataset sample or require manual annotation of the dataset. It accurately divides the conversation data into stages, analyzes the duration of each stage, and performs duration anomaly detection and root cause analysis on conversations with longer durations. Furthermore, it combines dimensions such as service type, user caller location, satisfaction level, service activity ID, and frequently asked customer complaints for full-process intelligent analysis and decision support. In traditional solutions, manual analysis of a single conversation takes an average of 4 minutes. In this application embodiment, only the output results of the large language model need to be analyzed, and manual analysis of a single service takes only 30 seconds. The daily analysis volume has increased from 120 conversations to 15,000 conversations across all text services, improving the efficiency of determining customer service call duration by 7 times. Therefore, it is possible to accurately determine the root cause of increased customer service call duration, thus meeting the needs of improving service efficiency and quality.

[0089] Based on the above, in the embodiments of this application, such as Figure 2As shown, the large language model automatically breaks down customer service call session data into four stages: "identifying and confirming user requests → collecting service elements → querying the customer service system → answering user questions," and outputs time nodes. It detects call timeout anomalies, automatically distinguishes between specialist-related and non-specialist-related factors, and optimizes services accordingly. For non-specialist-related factors, semantic clustering using the large language model generates root cause labels such as specialist service system factors, user service system factors, and service product factors. In this embodiment, for service process issues, the LLM large language model performs zero-shot semantic parsing of service session data, constructs the correlation between service session data and historical abnormal calls, assesses information based on the emotional information of both parties and the severity of customer complaints, traces and locates problem stages, and automatically generates high-frequency hot issues and handling suggestions. This achieves full-process automation from call breakdown and duration anomaly detection to intelligent auxiliary decision-making for customer complaints, providing a basis for targeted optimization of user service processes and specialist training.

[0090] Based on this, after step 140, the embodiments of this application can also propose corresponding improvement measures based on the root cause data of time consumption. To better illustrate the data processing method provided by the embodiments of this application, combined with... Figure 3 The data processing system in the embodiments of this application will perform a data processing method in detail.

[0091] In some embodiments of this application, the time-consuming root cause data includes specialist factors, and the service session data includes session text arranged in chronological order. The session text is marked with the identity identifier of the session dialogue. The data method may also include step 210, which is used to generate training information for service specialists corresponding to the identity identifier based on the time-consuming service links and the time-consuming root cause data of the time-consuming service links.

[0092] For example, in a customer service center's "trade-in" service, LLM call analysis revealed that the average time spent by P3 in the customer service system's query process accounted for 45% (exceeding the threshold). =30%), which was determined to be the bottleneck stage. Further service analysis was used to pinpoint the root cause of the time consumption data as staff-related factors, such as "unfamiliarity with system operation" and "time consumption when switching between multiple platforms". Combined with the identity identifiers in the conversation text, such as staff IDs: CS001, CS002, etc., targeted training information was generated.

[0093] Among the reasons cited are: staff member unfamiliarity with system operation (switching time > 2 minutes / time); and process issues stemming from a lack of understanding of how to use the "one-stop query tool." Therefore, training information could include the following: Trainee identification: CS001 and staff members in the same batch whose P3 stage time exceeded the threshold, such as CS003 and CS005. Training objective: To master the operation of the "trade-in one-stop query tool" within 3 days, reducing the average P3 stage time from 5 minutes to within 2 minutes. Training content: 1) System operation skills: Explanation of tool interface functions, shortcut key usage, and error handling; 2) Process optimization techniques: Pre-set query templates, such as automatic order number filling, and multi-task parallel processing, such as recording customer needs while querying. Training methods: 1) Practical exercises: Simulating real order scenarios for system operation assessment; 2) Mentorship: One-on-one coaching by the staff member with the shortest P3 stage time (e.g., CS010).

[0094] In some embodiments of this application, the root cause data of time consumption includes non-dedicated factors, which include service system factors. The service system factors include at least one of the following: dedicated service system factors and user service system factors. Based on this, the data method may also include steps 2201 and 2202.

[0095] Step 2201: When the service system factor is the specialist service system factor, generate the first optimization information of the specialist service system based on the stage session data corresponding to the time-consuming service link, and send the first optimization information to the specialist service system.

[0096] For example, if the time-consuming service step is the customer service system query step in a trade-in customer service call, and the stage conversation data is labeled with the specialist's identity, such as CS001, analysis reveals that the root cause of the time consumption is related to the specialist's service system, such as the customer service backend system. Therefore, the problem is identified as specialist CS001 spending an average of 5 minutes in the specialist service system, far exceeding the average time for similar steps, such as 3 minutes, and the conversation repeatedly shows clues such as system switching time and incomplete information retrieval. At this point, combining system functions and specialist operating habits, the first optimization information for the specialist service system can be generated. This first optimization information can include: Functional optimization: adding a one-stop query tool to the specialist service system, integrating data from multiple platforms such as orders, subsidies, and logistics, reducing the number of times specialists switch systems; Operation guidance: updating the system operation manual, adding quick query paths, such as order query → trade-in zone → subsidy progress, and recording an operation demonstration video.

[0097] Step 2202: When the service system factor is a user service system factor, generate second optimization information for the user service system based on the stage session data corresponding to the time-consuming service step, and provide the second optimization information to the user service system. The user service system includes the application installed on the user's electronic device and the server corresponding to the application.

[0098] For example, if the time-consuming service step is the collection of service elements in a trade-in customer service call, and the stage session data is labeled with the user's identity (e.g., UID001), analysis reveals that the root cause of the time consumption is related to the user service system, such as the user's mobile app and the corresponding server. Therefore, the problem is identified as follows: User UID001 spent an average of 4 minutes checking subsidy progress on their own, far exceeding the average time of 1 minute for similar steps, and the conversation repeatedly revealed slow app loading and cumbersome information entry. Thus, by combining the performance of the user app and server, a second optimization information for the user service system can be generated. This second optimization information can include app-side optimizations, such as performance optimization of the subsidy query page (e.g., reducing redundant loading, optimizing interaction logic, and adding a "one-click query" shortcut); and server-side optimizations, such as upgrading the query interface response speed, reducing the subsidy progress query interface response time from 2 seconds to 0.5 seconds. User feedback: A pop-up notification within the app notifies users like UID001 that the time-consuming query function has been optimized, making the operation smoother, and guides users to update the app version.

[0099] In some embodiments of this application, after identifying root cause data on time consumption (i.e., the root cause data includes non-dedicated personnel factors, which in turn include service product factors), sessions with long service consumption times due to service process issues are prioritized for subsequent customer complaint analysis. Targeted optimization of complex service process issues allows for customer complaint analysis, specifically the steps of analyzing and making decisions based on abnormal calls and frequently asked customer complaints. Therefore, this data method may further include steps 2301 to 2204.

[0100] Step 2301: Input the stage conversation data and communication summary element prompts corresponding to the time-consuming service steps into the large language model. The large language model extracts communication summary information from the stage conversation data according to the communication summary element prompts. The communication summary information includes the service products inquired by the user, user problem information, user request information, specialist feedback information, and emotional information of both parties.

[0101] For example, such as Figure 3As shown, the acquired customer complaint text and selected longer conversations are semantically parsed using a large language model to generate a user conversation summary. This summary covers key interaction nodes such as the user's inquiry, customer service promises, and the time when the user raised objections, as well as conversation outcome information such as whether the problem was resolved and whether the user still has dissatisfaction. Simultaneously, key elements of customer complaint handling are automatically extracted, including service identifiers such as activity IDs, explicitly stated problem descriptions such as "call interrupted 3 times" and "customer service failed to resolve the issue," emotional information from both parties, and user requests such as "refund," "reprocessing," and "apology." The emotional information from both parties can include text sentiment values, which are calculated using a sentiment dictionary and contextual semantic weighting, ranging from 0 to 10, with higher values ​​indicating more intense emotions.

[0102] Step 2302: Based on the communication summary information, establish the association between service session data and historical abnormal calls.

[0103] For example, the abnormal call-customer complaint association engine is invoked to match the extracted key elements of the customer complaint and the user session summary with the previously identified communication summary information to obtain the association between service session data and historical abnormal calls.

[0104] Step 2303: Using the correlation degree of the relationship, assess the severity of the customer complaint event corresponding to the service session data to obtain the severity assessment information of the customer complaint event.

[0105] For example, the text sentiment value, the number of users involved (e.g., group complaints are given high weight), and the severity level of abnormal calls (e.g., the number of call interruptions) are used as input parameters. A preset weighted algorithm is used to calculate the severity score, which is divided into three levels: "moderate," "relatively severe," and "severe," each corresponding to a different processing priority. Simultaneously, high-frequency and popular issues analysis is conducted at this stage. Cluster analysis is performed on all associated customer complaint issues and core topics in the conversation summary to identify frequently occurring issues such as "unclear activity rules" and "refund not received," which are marked as high-frequency and popular issues, and the service volume, the type of service involved, and the characteristics of associated abnormal calls are recorded.

[0106] Step 2304: Based on the severity assessment information of the customer complaint, send service product usage feedback information to the service product's service system. The service product usage feedback information is used to optimize the service product.

[0107] For example, for customer complaints at the "serious" level or above, root cause tracing is performed. Through hierarchical breakdown, the service link where the abnormal call occurred, such as consultation, processing, and answering, is first located. Then, the service logs and process specifications of that link are retrieved, and the characteristic parameters of the abnormal call, such as the system response time when the call is interrupted and the deviation between the customer service operation record and the standard process, are compared. Combined with the problem description mentioned in the customer complaint, the user conversation summary, and the common characteristics of high-frequency and popular questions, the specific cause of the customer complaint is identified, such as service system failure or process design defects. A comprehensive report is generated, which includes the root cause location results, related evidence chains such as abnormal call data fragments, screenshots of key sentences in the customer complaint, core content of the conversation summary, and statistical analysis of high-frequency and popular questions. This report is then sent to the corresponding service department.

[0108] It should be noted that, taking the "trade-in" car subsidy scenario as an example, the "Abnormal Calls - Popular Customer Complaint Issues" full-process auxiliary analysis device revealed abnormal call issues in the car subsidy program, with customer service calls averaging 889 seconds per call, far exceeding other service scenarios. This device pinpointed the long car subsidy review process and assessed the severity of these complaints as "serious." After the user service report generated by the "Abnormal Calls - Popular Customer Complaint Issues" full-process auxiliary analysis device was sent to the service department for optimization, the customer service call duration decreased to 739 seconds per call, a decrease of 16.9%, and the intelligent robot resolution rate increased from 63.4% to 71.7%, an improvement of 8.3%.

[0109] Through the above steps, the entire process from abnormal call screening and accurate customer complaint analysis to root cause identification and hot issue recognition is automated, improving the efficiency of customer complaint handling, the pertinence of problem solving, and the guiding value for service optimization, thereby improving the efficiency of customer service staff in handling customer service.

[0110] In this embodiment of the application, the association relationship includes at least one of the following: a first association relationship and a second association relationship. Step 2302 may specifically include steps 23021 to 23023, as shown below.

[0111] Step 23021: Match the communication summary information with the call summary information of at least one historical abnormal call.

[0112] Step 23022: If the first call summary information corresponding to the communication summary information is matched, establish the first association relationship between the service session data and the historical abnormal calls corresponding to the first call summary information.

[0113] Step 23023: If no target call summary information corresponding to the communication summary information is matched, determine a second call summary information with a similarity higher than a preset threshold from the call summary information of at least one historical abnormal call; and establish a second association relationship between the service session data and the historical abnormal call corresponding to the second call summary information. It should be noted that when the association relationship is a first association relationship, the association degree is the first association degree; when the association relationship is a second association relationship, the association degree is the second association degree; wherein, the value of the first association degree is greater than the value of the second association degree.

[0114] For example, if a first call summary corresponding to the communication summary information is matched, it is considered a first correlation, and its first correlation is 1. If no target call summary information corresponding to the communication summary information is matched, the similarity between the communication summary information and the call summary information of historical abnormal calls is calculated, the timestamp correlation (i.e., the difference between the customer complaint time and the abnormal call time is within a preset threshold) is considered a valid correlation, the problem description matching degree is calculated to determine the customer complaint problem, and the text similarity between the conversation summary and the abnormal call type is calculated. A correlation score is generated by combining these factors. When the score exceeds the threshold, it is determined that there is a causal relationship between the customer complaint and the corresponding abnormal call, i.e., a second correlation relationship. At this time, the second correlation degree of the second correlation relationship should be less than 1, which can be set to a number between greater than 0 and less than 1.

[0115] Therefore, this application proposes a customer service call duration analysis method and system based on a large language model. This method enables full-process analysis from call duration breakdown and duration anomaly detection to customer complaint-assisted decision-making. The method employs a zero-sample customer service time-consuming step breakdown and time-consuming mapping approach, and an automatic customer service call step breakdown method based on a large language model. Without requiring any manually labeled training data, it performs zero-sample step identification on customer service calls, automatically dividing the call into four steps: identifying and confirming user requests, collecting service elements, querying the customer service system, and answering user questions. Timestamps are output synchronously, and step categories can be dynamically added or deleted using configurable templates to adapt to different service scenarios. This achieves unsupervised structured call flow, solving the pain point of traditional systems requiring extensive labeled training. The multi-stage call duration anomaly detection and root cause extraction steps identify abnormal service time-consuming stages based on the time-consuming anomaly thresholds for each service type, and perform service anomaly detection. Text calls with abnormal service latency are used to perform secondary semantic clustering on the LLM large language model, automatically generating root cause data such as staff inexperience, service process issues, query platform problems, and system anomalies. This enables automated extraction of root cause data analysis, such as providing personalized training to staff on service skills and knowledge retrieval based on staff factors, thereby improving the quality and efficiency of user services.

[0116] Thus, a full-process intelligent auxiliary analysis and decision-making method based on abnormal calls and frequently asked customer complaints is developed. For service process issues, zero-sample semantic analysis of calls with abnormal durations is performed using a large language model, automatically generating conversation summaries and constructing an abnormal call-customer complaint association matrix. The severity of customer complaints is assessed based on multi-dimensional factors such as sentiment and anomaly level. For more severe or higher-level complaints, high-frequency frequently asked questions and handling suggestions are intelligently generated and sent to the service department. This achieves full-process intelligent assistance from customer complaint association to hotspot identification to handling suggestions, solving the pain points of traditional systems such as difficulty in discovering frequently asked customer complaints and delays in handling suggestions.

[0117] Based on the same inventive concept, this application also provides a data processing apparatus. (Specifically combined with...) Figure 4 Please provide a detailed explanation.

[0118] Figure 4 This is a schematic diagram of the structure of a data processing apparatus provided in one embodiment of this application.

[0119] In some embodiments of this application, Figure 4 The data processing device shown can be installed in a computer device.

[0120] like Figure 4 As shown, the data processing device 40 may specifically include: Module 401 is used to obtain service session data; The determination module 402 is used to determine N service steps and the start and end times of each service step based on the service session data using a large language model, where N is an integer greater than or equal to 1. The determining module 402 is also used to determine the time-consuming service link among the N service links based on the start and end time of each service link, wherein the service duration of the time-consuming service link is greater than or equal to the preset service duration. The determination module 402 is also used to determine the root cause data of time consumption based on the stage session data corresponding to the time-consuming service link in the service session data. The root cause data of time consumption includes at least one of the following: specialist factors and non-specialist factors.

[0121] The data processing device 40 in the embodiments of this application will be described in detail below.

[0122] In one or more optional embodiments, the acquisition module 401 can also be used to acquire service session sampling requirements; The acquisition module 401 can also be used to retrieve M service session identifiers related to the service session sampling requirements from the service session database, where M is an integer greater than or equal to 1, based on the service session sampling requirements. The determination module 402 is further configured to determine the candidate service session data corresponding to each service session identifier as service session data.

[0123] In one or more optional embodiments, the determining module 402 may be specifically used to input the service session data and the prompt words of the extensible link template library into the large language model when the service session data includes session text arranged in chronological order and the session text is marked with a timestamp, and to identify the service session data according to the prompt words of the extensible link template library through the large language model to obtain N service links of the service session data. Based on each service step, obtain the start and end session texts of each service step from the service session data; The start and end times of each service step are determined based on the timestamps of the start and end conversation texts of each service step.

[0124] In one or more optional embodiments, the determining module 402 can also be used to abstract the service session process of the service session scenario into at least two scalable core links. The determination module 402 can also be used to determine the link characteristics of each expandable core link based on each expandable core link. In this embodiment, the data processing device 40 may further include a construction module for constructing an expandable link template library based on each expandable core link and the link features of each expandable core link; the expandable link template library includes P link data, each link data includes an expandable core link and the link features of the expandable core link, where P is an integer greater than or equal to 1. The determination module 402 can also be used to configure corresponding prompt words for each piece of stage data, and use the prompt words corresponding to each piece of stage data as prompt words for the expandable stage template library.

[0125] In one or more optional embodiments, the determining module 402 may be specifically used to determine the service duration of each service step based on the start and end times of each service step when a service step corresponds to a preset service duration. Compare the service duration of each service step with the preset service duration for each service step; If the service duration of each service step is greater than or equal to the preset service duration corresponding to each service step, then each service step is defined as a time-consuming service step.

[0126] In one or more optional embodiments, the data processing device 40 in this application embodiment may further include a statistics module, used to count the service duration of each service link corresponding to the expandable core link in the expandable link template library within a preset time period. The determination module 402 can also be used to use the statistical value of the service duration of the expandable core links as the preset service duration corresponding to each service link.

[0127] In one or more optional embodiments, the data processing device 40 in this application embodiment may further include an input module, which is used to input the stage conversation data and root cause extraction prompt words corresponding to the time-consuming service link into the large language model, and perform semantic clustering analysis on the stage conversation data according to the root cause extraction prompt words through the large language model to obtain the time-consuming root cause data of the time-consuming service link output by the large language model.

[0128] In one or more optional embodiments, the data processing device 40 in this application embodiment may further include a generation module, used to generate training information for service specialists corresponding to the identity identifiers, based on the time-consuming service links and the time-consuming root cause data of the time-consuming service links, when the time-consuming root cause data includes specialist factors, the service session data includes session text arranged in chronological order, and the session text is marked with the identity identifier of the session dialogue.

[0129] In one or more optional embodiments, the time-consuming root cause data includes non-dedicated factors, which include service system factors, and the service system factors include at least one of the following: dedicated service system factors and user service system factors; In this embodiment, the data processing device 40 may further include a generation module, which is used to generate first optimization information of the specialist service system based on the stage session data corresponding to the time-consuming service link when the service system factor is a specialist service system factor. In this embodiment of the application, the data processing device 40 may further include a sending module for sending first optimization information to the specialist service system; The generation module can also be used to generate second optimization information for the user service system based on the stage session data corresponding to the time-consuming service link, when the service system factor is the user service system factor. The sending module can also be used to send second optimization information to the user service system.

[0130] In one or more optional embodiments, the data processing device 40 in this application embodiment may further include an input module, used to input the stage conversation data and communication summary element prompts corresponding to the time-consuming service link into the large language model when the time-consuming root cause data includes non-specialist factors and the non-specialist factors include service product factors, and the large language model extracts communication summary information from the stage conversation data according to the communication summary element prompts. The communication summary information includes the service product inquired by the user, user question information, user request information, specialist feedback information, and emotional information of both parties. In this embodiment, the data processing device 40 may further include an establishment module, used to establish an association between service session data and historical abnormal calls based on communication summary information; The determination module 402 can also be used to assess the severity of customer complaint events corresponding to service session data by utilizing the correlation degree of the association relationship, and obtain severity assessment information of customer complaint events. In this embodiment, the data processing device 40 may further include a sending module, which is used to send service product usage feedback information to the service system of the service product based on the severity assessment information of the customer complaint, and the service product usage feedback information is used to optimize the service product.

[0131] In one or more optional embodiments, the data processing device 40 in this application embodiment may further include a matching module, used to match communication summary information with call summary information of at least one historical abnormal call when the association relationship includes at least one of the following: a first association relationship and a second association relationship; In this embodiment of the application, the data processing device 40 may further include an establishment module, which is used to establish a first association relationship between the service session data and the historical abnormal calls corresponding to the first call summary information when the first call summary information corresponding to the communication summary information is matched; The determining module 402 can also be used to determine, in the case that no target call summary information corresponding to the communication summary information is matched, a second call summary information with a similarity higher than a preset threshold from the call summary information of at least one historical abnormal call; The module can also be used to establish a second association between service session data and historical abnormal calls corresponding to the second call summary information.

[0132] In one or more optional embodiments, when the association relationship is a first association relationship, the degree of association of the association relationship is a first degree of association; When the association relationship is a second-degree association relationship, the degree of association is the second-degree association. Among them, the value of the first degree of association is greater than the value of the second degree of association.

[0133] Therefore, the data processing apparatus provided in one embodiment of this application can acquire service session data and, through a large language model, determine N service steps and the start and end times of each service step based on the service session data; determine time-consuming service steps among the N service steps based on the start and end times of each service step, wherein the service duration of the time-consuming service steps is greater than or equal to a preset service duration; and determine time-consuming root cause data based on the stage session data corresponding to the time-consuming service steps in the service session data, wherein the time-consuming root cause data includes at least one of the following: specialist factors and non-specialist factors. In this way, based on a large language model combined with prompt word engineering and zero-shot learning technology, N service steps and the start and end times of each step can be automatically identified without manual annotation, avoiding the subjectivity and inefficiency of traditional manual sampling. Furthermore, by calculating the service duration based on the start and end times of each step and automatically filtering time-consuming service steps through a preset service duration, normal and abnormal time consumption can be quantitatively distinguished, overcoming the deficiency of traditional overall statistical analysis in being unable to locate specific bottlenecks. Furthermore, by using a large language model to perform semantic analysis on the phased conversation data of time-consuming processes, the system automatically extracts the root cause data of time consumption and accurately distinguishes between specialist factors such as service unfamiliarity and redundant communication, and non-specialist factors such as slow response and cumbersome processes. This solves the problems of traditional manual analysis relying on experience judgment and vague root causes. It can provide quantitative basis for service process optimization, specialist training and platform iteration through intelligent analysis of service conversation data using a large language model. It realizes the accurate positioning and root cause tracing of customer service call duration issues to meet the needs of improving service efficiency and service quality.

[0134] Based on the same inventive concept, this application also provides a computer device. (Specifically combined with...) Figure 5 Please provide a detailed explanation.

[0135] Figure 5 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application.

[0136] like Figure 5 As shown, the computer device may include at least one of the following as described in the embodiments of this application: an electronic device, a server. The computer device may include a processor 501 and a memory 502 storing computer program instructions.

[0137] Specifically, the processor 501 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0138] Memory 502 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 502 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 502 may include removable or non-removable (or fixed) media. Where appropriate, memory 502 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 502 is non-volatile solid-state memory. In a particular embodiment, memory 502 includes solid-state storage (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0139] The processor 501 implements any of the data processing methods described in the above embodiments by reading and executing computer program instructions stored in the memory 502.

[0140] In one example, the computer device may also include a communication interface 503 and a bus 510. Wherein, as... Figure 5 As shown, the processor 501, memory 502, and communication interface 503 are connected through bus 510 and complete communication with each other.

[0141] The communication interface 503 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0142] Bus 510 includes hardware, software, or both, that couples components of a flow control device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 510 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0143] The computer device can execute the data processing method described in the embodiments of this application, thereby achieving the combination Figures 1 to 5 The data processing methods and apparatus described.

[0144] Furthermore, in conjunction with the data processing methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the data processing methods in the above embodiments.

[0145] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0146] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0147] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0148] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A data processing method, characterized by, The method comprises the following steps: obtaining service session data; determining N service links and the start and end time of each service link according to the service session data by a large language model, wherein N is an integer greater than or equal to 1; determining a time-consuming service link in the N service links according to the start and end time of each service link, wherein the service duration of the time-consuming service link is greater than or equal to a preset service duration; determining time-consuming root cause data according to the stage session data corresponding to the time-consuming service link in the service session data, wherein the time-consuming root cause data comprises at least one of the following: agent factors, non-agent factors.

2. The method of claim 1, wherein, Before the step of obtaining service session data, the method further comprises the following steps: obtaining service session sampling requirements; obtaining M service session identifiers related to the service session sampling requirements from a service session database according to the service session sampling requirements, wherein M is an integer greater than or equal to 1; determining candidate service session data corresponding to each service session identifier as the service session data.

3. The method of claim 1, wherein, The service session data comprises session texts arranged in chronological order, and the session texts are marked with timestamps; the step of determining N service links and the start and end time of each service link according to the service session data by a large language model comprises the following steps: inputting the service session data and prompt words of an expandable link template library into the large language model, identifying the service session data according to the prompt words of the expandable link template library through the large language model, and obtaining N service links of the service session data; obtaining session texts starting and ending at each service link from the service session data according to each service link; determining the start and end time of each service link according to the timestamps of the session texts starting and ending at each service link.

4. The method of claim 3, wherein, The method further comprises the following steps: abstracting a service session process of a service session scenario into at least two expandable core links; determining link characteristics of each expandable core link according to each expandable core link; constructing the expandable link template library based on each expandable core link and the link characteristics of each expandable core link; the expandable link template library comprises P link data, each link data comprises an expandable core link and the link characteristics of the expandable core link, and P is an integer greater than or equal to 1; configuring corresponding prompt words for each link data, and taking the prompt words corresponding to each link data as the prompt words of the expandable link template library.

5. The method of claim 1, wherein, One service link corresponds to one preset service duration; The step of determining a time-consuming service link in the N service links according to the start and end time of each service link comprises the following steps: determining the service duration of each service link according to the start and end time of each service link; comparing the service duration of each service link with the preset service duration corresponding to each service link; In a case where a service duration of each of the service links is greater than or equal to a preset service duration corresponding to each of the service links, each of the service links is determined as the time-consuming service link.

6. The method of claim 5, wherein, The method further includes: statistically counting a statistical value of the service duration of the expandable core link corresponding to each of the service links in the expandable link template library within a preset time period; the statistical value of the service duration of the expandable core link is used as the preset service duration corresponding to each of the service links.

7. The method of claim 1, wherein, The method further includes: inputting the stage session data corresponding to the time-consuming service link and the root cause extraction prompt word into the large language model, performing semantic clustering analysis on the stage session data according to the root cause extraction prompt word through the large language model, and obtaining the time-consuming root cause data of the time-consuming service link output by the large language model.

8. The method according to claim 1 or 7, characterized in that, The time-consuming root cause data includes the non-agent factor, and the non-agent factor includes a service product factor; the method further includes: inputting the stage session data corresponding to the time-consuming service link and a communication summary element prompt word into the large language model, extracting communication summary information from the stage session data according to the communication summary element prompt word through the large language model, and the communication summary information includes a user inquiry service product, user problem information, user demand information, agent feedback information and both parties' emotional information; 9. The method of claim 1, wherein, establishing an association relationship between the service session data and a historical abnormal call according to the communication summary information; using an association degree of the association relationship to evaluate a severity of a complaint event corresponding to the service session data, and obtaining severity evaluation information of the complaint event; sending service product use feedback information to a service system of the service product according to the severity evaluation information of the complaint event, and the service product use feedback information is used to optimize the service product.

10. The method of claim 1, wherein, The time-consuming root cause data includes the non-agent factor, and the non-agent factor includes a service product factor; the method further includes: inputting the stage session data corresponding to the time-consuming service link and a communication summary element prompt word into the large language model, extracting communication summary information from the stage session data according to the communication summary element prompt word through the large language model, and the communication summary information includes a user inquiry service product, user problem information, user demand information, agent feedback information and both parties' emotional information; establishing an association relationship between the service session data and a historical abnormal call according to the communication summary information; using an association degree of the association relationship to evaluate a severity of a complaint event corresponding to the service session data, and obtaining severity evaluation information of the complaint event; sending service product use feedback information to a service system of the service product according to the severity evaluation information of the complaint event, and the service product use feedback information is used to optimize the service product.

11. The method of claim 10, wherein, The association relationship includes at least one of a first association relationship and a second association relationship. The method comprises: matching the communication summary information with call summary information of at least one historical abnormal call; in a case where the first call summary information corresponding to the communication summary information is matched, establishing a first association relationship between the service session data and a historical abnormal call corresponding to the first call summary information; 12. The method of claim 11, wherein, in a case where the target call summary information corresponding to the communication summary information is not matched, determining second call summary information with a high similarity to the communication summary information from the call summary information of the at least one historical abnormal call; and establishing a second association relationship between the service session data and a historical abnormal call corresponding to the second call summary information. In a case where the association relationship is the first association relationship, the association degree of the association relationship is a first association degree. In a case where the association relationship is the second association relationship, the association degree of the association relationship is a second association degree. The value of the first association degree is greater than the value of the second association degree.

13. A data processing apparatus, comprising: an acquisition module configured to acquire service session data; a determination module configured to determine, by a large language model, N service links and start and end times of each service link according to the service session data, N being an integer greater than or equal to 1; the determination module is further configured to determine a time-consuming service link in the N service links according to the start and end times of each service link, the service duration of the time-consuming service link being greater than or equal to a preset service duration; 14. A computer device, comprising: the determination module is further configured to determine time-consuming root cause data according to stage session data corresponding to the time-consuming service link in the service session data, the time-consuming root cause data including at least one of an agent factor and a non-agent factor. The computer device comprises a processor and a memory storing computer program instructions; 15. A computer-readable storage medium, characterized in that, the processor executes the computer program instructions to implement the data processing method of any one of claims 1-12.

16. A computer program product, characterised in that, The computer readable storage medium stores computer program instructions, and the computer program instructions are executed by the processor to implement the data processing method of any one of claims 1-12. The computer program is executed by the processor to implement the data processing method of any one of claims 1-12.

Images