Customer service intelligent response method and system driven by AI big model

Abstract

The present invention discloses a method and system for intelligent customer service response driven by an AI big model, and relates to the field of data processing technology. The method comprises: obtaining a response record set, taking K business items as indexes, performing aggregation analysis on the response record set, and generating K business item response record sets; performing business item service analysis, and obtaining K business item first response coefficients; collecting L consultation work orders, extracting historical consultation dialogue records, and obtaining L historical consultation dialogue record sets; performing dialogue correlation feature comparison, and generating L customer consultation coefficients; performing response response coefficient identification, and obtaining L response response coefficients; performing resource balanced allocation on the target response platform, obtaining L response processing units, and using L response processing units to process L consultation work orders. The technical problem of low efficiency of customer service response in the prior art is solved, and the technical effect of improving response efficiency is achieved through balanced resource allocation.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a customer service intelligent response method and system driven by an AI big model. Background Art

[0002] In today's digital age, customer service, as a bridge of communication between enterprises and customers, is becoming increasingly important. With the continuous expansion of enterprise scale and the intensification of market competition, customer service departments are facing unprecedented challenges: the influx of massive consulting tickets, the diversification and personalization of customer needs, and customers' high expectations for service response speed and quality. Traditional customer service response systems often find it difficult to cope with these challenges, with problems such as slow response speed, low processing efficiency, and insufficient personalized services, which seriously restrict the improvement of enterprise service levels and customer satisfaction. Summary of the invention

[0003] This application provides a customer service intelligent response method and system driven by an AI large model, which solves the technical problem of low customer service response efficiency in the prior art.

[0004] In view of the above problems, this application provides an AI big model driven customer service intelligent response method and system.

[0005] In a first aspect of the present application, a customer service intelligent response method driven by an AI big model is provided, the method comprising:

[0006] Acquire a set of response records of the target response platform within a preset window, perform aggregation analysis on the response record set with K business items as indexes, and generate K business item response record sets; traverse the K business item response record sets to perform business item service analysis, and obtain K business item first response coefficients; collect L consulting work orders of the target response platform at the current moment, extract historical consulting conversation records with L consulting customers of the L consulting work orders as indexes, and obtain L historical consulting conversation record sets, wherein the L consulting work orders include L consulting business items; traverse the L historical consulting conversation record sets to perform conversation correlation feature comparison, and generate L customer consulting coefficients; identify response response coefficients based on the L consulting business items, the L customer consulting coefficients, and the K business item first response coefficients, and obtain L response response coefficients; perform resource balancing allocation on the target response platform based on the L response response coefficients, obtain L fully configured response processing units, and use the L response processing units to process the L consulting work orders.

[0007] The second aspect of the present application provides a customer service intelligent response system driven by an AI big model, the system comprising:

[0008] An analysis module, the analysis module is used to obtain a set of response records of a target response platform within a preset window, and to perform aggregation analysis on the response record set with K business items as indexes to generate K business item response record sets; a parsing module, the parsing module is used to traverse the K business item response record sets to perform business item service parsing to obtain K business item first response coefficients; an extraction module, the extraction module is used to collect L consulting work orders of the target response platform at the current moment, and to extract historical consulting dialogue records with L consulting customers of the L consulting work orders as indexes to obtain L historical consulting dialogue record sets, wherein the L consulting work orders include The invention comprises L consulting business items; a feature comparison module, wherein the feature comparison module is used to traverse the L historical consulting conversation record sets to perform conversation-related feature comparison and generate L customer consulting coefficients; an identification module, wherein the identification module is used to identify answer response coefficients according to the L consulting business items, the L customer consulting coefficients and the first answer coefficients of the K business items and obtain L answer response coefficients; a processing module, wherein the processing module is used to perform resource balancing allocation on the target answer platform based on the L answer response coefficients, obtain L fully configured answer processing units, and use the L answer processing units to process the L consulting work orders.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, obtain the response record set of the target response platform within the preset window, use K business items as indexes, perform aggregate analysis on the response record set, and generate K business item response record sets. Next, traverse the K business item response record sets to perform business item service analysis and obtain the first response coefficients of K business items. By collecting L consultation work orders of the target response platform at the current moment, extract historical consultation dialogue records using L consultation customers of L consultation work orders as indexes to obtain L historical consultation dialogue record sets, where L consultation work orders include L consultation business items. Next, traverse the L historical consultation dialogue record sets to perform dialogue correlation feature comparison and generate L customer consultation coefficients. Then, identify the response response coefficient based on the L consultation business items, L customer consultation coefficients and K business item first response coefficients to obtain L response response coefficients. Finally, balance the resources of the target response platform based on the L response response coefficients, obtain L fully configured response processing units, and use the L response processing units to process the L consultation work orders. The technical problem of low customer service response efficiency in the prior art is solved, and the technical effect of improving response efficiency is achieved through balanced resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A flowchart of a customer service intelligent response method driven by an AI large model provided in an embodiment of the present application;

[0013] Figure 2 A schematic diagram of the structure of the customer service intelligent response system driven by an AI big model provided in an embodiment of the present application.

[0014] Explanation of the reference numerals: analysis module 11 , parsing module 12 , extraction module 13 , feature comparison module 14 , recognition module 15 , processing module 16 . DETAILED DESCRIPTION

[0015] This application solves the technical problem of low customer service response efficiency in the prior art by providing a customer service intelligent response method and system driven by an AI big model.

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0017] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0018] Embodiment 1, as Figure 1 As shown, the present application provides a customer service intelligent response method driven by an AI large model, wherein the method includes:

[0019] A response record set of a target response platform within a preset window is obtained, and K business items are used as indexes to perform aggregation analysis on the response record set to generate K business item response record sets.

[0020] The answer records of the target answer platform are collected within a preset window (a preset time window, such as one hour), including the answer time, answer content, related customer consultations, business item identifiers, etc., to form an answer record set; K specific business items are used as indexes to perform aggregation analysis on the answer record set, that is, each answer record is assigned to the corresponding business item answer record set according to its associated business item identifier, thereby generating K answer record sets for different business items.

[0021] The K business item response record sets are traversed to perform business item service analysis to obtain the K business item first response coefficients.

[0022] By traversing K business item response record sets, in-depth service analysis is performed on each business item response record set to obtain K business item first response coefficients. The business item first response coefficient is used to reflect the business item response performance. The larger the response coefficient, the more complicated the business item itself involves, which requires more easy to understand and more detailed analysis.

[0023] Furthermore, traversing the K business item response record sets to perform business item service analysis and obtain K business item first response coefficients includes:

[0024] Based on the K business item answer record sets, single conversation information entropies of the K business items are extracted to obtain K business item single conversation information entropy sets; repeated question and answer frequencies are extracted from the K business item answer record sets respectively to obtain K business repeated question and answer frequencies; data is extracted from the K business item answer record sets using the repeated question and answer interval time period as an index to obtain K repeated question and answer interval time period sets; the K business item single conversation information entropy sets and the K repeated question and answer interval time period sets are respectively subjected to flood screening analysis to determine K focused conversation information entropies and K focused interval time periods; the service resolution network layer is used to analyze the K business repeated question and answer frequencies, the K focused conversation information entropies and the K focused interval time periods to obtain the first response coefficients of the K business items.

[0025] Specifically, the information entropy of a single conversation is extracted from the answer record set of each business item to obtain K sets of information entropy of a single conversation for business items. The K sets of information entropy of a single conversation for business items reflect the information richness and complexity of each business item in a single conversation, wherein the information entropy of a single conversation is used to evaluate the complexity and amount of information in a single conversation; the answer record set of each business item is scanned to count and extract the frequency of repeated questions and answers, that is, the frequency of the same person repeatedly asking questions about the same business item, thereby obtaining K frequencies of repeated business questions and answers; taking the time period between repeated questions and answers as an index, relevant data is extracted from the answer record set of each business item to obtain K sets of time periods between repeated questions and answers; The information entropy set and the repeated question and answer interval time period set are respectively subjected to flood screening analysis. By setting certain conditions to filter out unimportant or abnormal data points, key information is highlighted, and K focus conversation information entropies and K focus interval time periods are determined. These are the information entropies and interval time periods that are considered to be the most important for evaluating the service quality and efficiency of business item responses after screening; the service resolution network layer is a deep learning model used to conduct in-depth analysis of the collected data and generate useful insights. It takes the K business repeated question and answer frequencies, the K focus conversation information entropies, and the K focus interval time periods as input, and uses the service resolution network layer for analysis. The output result is the first response coefficient of the K business items.

[0026] Furthermore, the K business item single conversation information entropy sets and the K repeated question-answer interval time periods are respectively subjected to flood screening analysis to determine K focused conversation information entropies and K focused interval time periods, including:

[0027] Extracting a first business item single conversation information entropy set from the K business item single conversation information entropy sets; extracting a minimum value from the first business item single conversation information entropy set to obtain a first flooding starting point; using the first business item single conversation information entropy set as a simulated image, using each first business item single conversation information entropy as a simulated pixel point, performing flooding screening analysis based on the first flooding starting point to obtain a first focused conversation information entropy; performing flooding screening analysis on the remaining K-1 business item single conversation information entropy sets to obtain K-1 focused conversation information entropies, and combining the first focused conversation information entropies to obtain the K focused conversation information entropies; performing flooding screening analysis on the K repeated question and answer interval time period sets to obtain the K focused interval time periods.

[0028] Specifically, one is selected from the K business item single conversation information entropy sets as the first business item single conversation information entropy set; the minimum value is extracted from the first business item single conversation information entropy set as the starting point of flood screening, namely the first flood starting point; the first business item single conversation information entropy set is regarded as a simulated image, in which each information entropy value is regarded as a simulated pixel point, starting from the first flood starting point, a flood filling algorithm (similar to the region growing algorithm in image processing) is applied to identify pixels (i.e., information entropy values) that are similar to it (i.e., have similar information entropy values), and this process will "diffuse" until it encounters a region that is significantly different from the current region (i.e., the information entropy value is different from the current region). The information entropy value in this area is finally considered as the first focused conversation information entropy. Repeat the above steps to perform flood screening analysis on the single conversation information entropy set of the remaining K-1 business items, determine the focused conversation information entropy for each business item, and thus obtain K focused conversation information entropies. The flood screening analysis can also be applied to the set of repeated question-and-answer interval time periods, but at this time, the focus is on the length and pattern of the time interval. By analyzing the interval time period, the question-and-answer behavior pattern of the user or system in certain specific time periods can be identified, such as peak time periods, trough time periods or periodic patterns, and then K focused interval time periods can be obtained.

[0029] Further, including:

[0030] Taking the first flooding starting point as the central water filling point, multiple simulated pixel points whose information entropy difference value of the first business item single conversation information entropy minimum value corresponding to the first flooding starting point in the first business item single conversation information entropy set is within a preset information entropy difference threshold value are injected with water of the first color to obtain a first color simulated pixel point set, wherein the information entropy difference value is positive; adding the first business item single conversation information entropy minimum value to the preset information entropy difference threshold value to obtain a second flooding starting point; taking the second flooding starting point as the central water filling point, adding the first business item single conversation information entropy minimum value corresponding to the first flooding starting point in the first business item single conversation information entropy set to the preset information entropy difference threshold value to obtain a second flooding starting point; taking the second flooding starting point as the central water filling point, Water of the second color is injected into multiple simulated pixel points whose information entropy difference at the starting point of the second flood is within a preset information entropy difference threshold to obtain a second color simulated pixel point set, wherein the second color is different from the first color; after multiple water injections, multiple color simulated pixel point sets of the single conversation information entropy set of the first business item are obtained, and according to the mapping relationship between the simulated pixel points and the single conversation information entropy of the first business item, multiple color first business item single conversation information entropy sets are matched to obtain; based on the multiple color first business item single conversation information entropy sets, screening is performed to obtain the first focus conversation information entropy.

[0031] Specifically, starting from the first flooding starting point, check the information entropy difference between other elements (i.e., simulated pixel points) in the set and the first flooding starting point; if the difference is within the preset information entropy difference threshold (and the difference is positive, which means that the information entropy of other elements is higher than but close to the minimum value), mark these elements (simulated pixel points) with the first color, indicating that they are similar to the starting point, and obtain the first color simulated pixel point set; add the minimum value of the information entropy of a single conversation of the first business item to the preset information entropy difference threshold to obtain a new threshold as the second flooding starting point; with the second flooding starting point as the center, check the information entropy difference of other elements in the set again. If the difference is within the preset information entropy difference threshold (but this time it is compared with the second flooding starting point), then mark this element (simulated pixel point) with the first color, indicating that they are similar to the starting point, and obtain the first color simulated pixel point set; add the minimum value of the information entropy of a single conversation of the first business item to the preset information entropy difference threshold to obtain a new threshold as the second flooding starting point; with the second flooding starting point as the center, check the information entropy difference of other elements in the set again. If the difference is within the preset information entropy difference threshold (but this time it is compared with the second flooding starting point), some elements are marked as a second color to obtain a second color simulated pixel point set; repeat the above steps, possibly using more different colors to represent different ranges of information entropy values, until the entire set is fully explored or a certain stopping condition is reached (such as no new elements are marked); associate each marked simulated pixel point (i.e., information entropy value) with its corresponding color (or label) to form multiple color simulated pixel point sets, and through a mapping relationship, convert these color sets back to information entropy sets, i.e., multiple color first business item single conversation information entropy sets; screen based on the multiple color first business item single conversation information entropy sets to obtain the first focused conversation information entropy, for example, the information entropy value in a specific color set can be selected as the focused conversation information entropy.

[0032] Further, including:

[0033] The information entropy quantities of the single conversation information entropy sets of the multiple color first business items are respectively counted to obtain multiple first information entropy quantities; the multiple first information entropy quantities are sorted in order from large to small, and the N color first business item single conversation information entropy sets corresponding to the first information entropy quantities located in the first n digits are processed by information entropy mean value to obtain the first focused conversation information entropy.

[0034] Specifically, for each color-marked first business item single conversation information entropy set, the amount of information entropy contained therein is counted to obtain multiple first information entropy amounts; these first information entropy amounts are sorted in descending order, and according to the sorted list, N color first business item single conversation information entropy sets corresponding to the first information entropy amounts located in the first n positions are selected; information entropy mean processing is performed on these N color first business item single conversation information entropy sets, that is, all information entropy values ​​in each set are added up, and then divided by the information entropy amount in the set, so as to obtain the first focus conversation information entropy.

[0035] Collect L consulting work orders of the target answering platform at the current moment, extract historical consulting conversation records using the L consulting customers of the L consulting work orders as indexes, and obtain L historical consulting conversation record sets, wherein the L consulting work orders include L consulting business items.

[0036] Collect L consultation tickets at the current moment from the target answering platform, where the consultation tickets contain the customer's basic information (such as ID, name or nickname), consultation time, consultation business item (i.e., the specific business or service content that the customer wishes to consult), etc. For the L consultation tickets collected, use the consulting customer in each ticket as an index, query and extract its historical consultation conversation records in the historical database or record storage, including past consultation time, consultation content, consultation results, etc., thereby obtaining L historical consultation conversation record sets, each set containing all past consultation conversation records of the customer.

[0037] The L historical consultation dialogue record sets are traversed to perform dialogue correlation feature comparison to generate L customer consultation coefficients.

[0038] By traversing L historical consultation dialogue record sets and comparing dialogue-related features, L customer consultation coefficients are generated.

[0039] Further, including:

[0040] Keyword groups are extracted from L sets of historical consultation dialogue records respectively to obtain L sets of historical consultation keyword group sequences; conversation-related feature recognition is performed on the L sets of historical consultation keyword group sequences using a semantic similarity identifier to generate L sets of conversation-related feature recognition coefficients; mean calculation is performed on the L sets of conversation-related feature recognition coefficients to obtain the L customer consultation coefficients.

[0041] Specifically, keyword groups are extracted from each set of historical consultation dialogue records to form L sets of historical consultation keyword group sequences, wherein each keyword group sequence includes a phrase in which the dialogue records are sorted in chronological order; for each historical consultation keyword group sequence, a semantic similarity identifier (such as an algorithm based on models such as Word2Vec and BERT) is used to calculate its semantic similarity with a reference point, and this similarity reflects the degree of association between the historical consultation and the current consultation or business standard, and the calculated semantic similarity is converted into a dialogue-related feature recognition coefficient, which is a direct mapping of the similarity, thereby generating L sets of dialogue-related feature recognition coefficients, wherein a larger recognition coefficient indicates that the customer repeatedly asks about a question and needs a more detailed explanation; for each customer's L historical consultation dialogues, the corresponding L dialogue-related feature recognition coefficients can be averaged to obtain a comprehensive customer consultation coefficient, and this coefficient reflects the average degree of association between the customer's historical consultation and the current consultation or business standard.

[0042] Response coefficients are identified based on the L consulting business items, the L customer consulting coefficients and the K business item first response coefficients to obtain L response coefficients.

[0043] The response coefficients are identified based on the L consulting business items, the L customer consulting coefficients and the K business item first response coefficients. The identification can be performed by constructing an evaluation model to obtain the L response coefficients.

[0044] Furthermore, a response coefficient identifier is constructed, and the response coefficient identifier is used to identify the response coefficients of the L consulting business items, the L customer consulting coefficients and the first response coefficients of the K business items to obtain L response coefficients.

[0045] Specifically, historical data is collected as training data, including input data (consulting business items, customer consultation coefficients, business item first response coefficients) and output data (response coefficients), and the training data is used to train the machine learning model. After the training is completed, a response coefficient identifier is obtained; L consulting business items, L customer consultation coefficients and K business item first response coefficients are used as input data and input into the response coefficient identifier, and the response coefficient identifier calculates L response coefficients based on the input data.

[0046] Based on the L response coefficients, resources of the target response platform are evenly allocated to obtain L fully configured response processing units, and the L response processing units are used to process the L consulting work orders.

[0047] Based on L response coefficients, the resources of the target response platform are evenly allocated. Through intelligent scheduling, the system configures a complete response processing unit for each consultation ticket to ensure efficient use of resources and fast and accurate responses. Subsequently, these response processing units are used to process the L consultation tickets one by one to meet the customer's consultation needs.

[0048] In summary, the embodiments of the present application have at least the following technical effects:

[0049] First, obtain the response record set of the target response platform within the preset window, use K business items as indexes, perform aggregate analysis on the response record set, and generate K business item response record sets. Next, traverse the K business item response record sets to perform business item service analysis and obtain the first response coefficients of K business items. By collecting L consultation work orders of the target response platform at the current moment, extract historical consultation dialogue records using L consultation customers of L consultation work orders as indexes to obtain L historical consultation dialogue record sets, where L consultation work orders include L consultation business items. Next, traverse the L historical consultation dialogue record sets to perform dialogue correlation feature comparison and generate L customer consultation coefficients. Then, identify the response response coefficient based on the L consultation business items, L customer consultation coefficients and K business item first response coefficients to obtain L response response coefficients. Finally, balance the resources of the target response platform based on the L response response coefficients, obtain L fully configured response processing units, and use the L response processing units to process the L consultation work orders. The technical problem of low customer service response efficiency in the prior art is solved, and the technical effect of improving response efficiency is achieved through balanced resource allocation.

[0050] Embodiment 2 is based on the same inventive concept as the customer service intelligent response method driven by the AI ​​big model in the above embodiment. Figure 2 As shown, the present application provides an AI large model-driven customer service intelligent response system, wherein the system includes:

[0051] An analysis module 11 is used to obtain a set of response records of a target response platform within a preset window, and to perform aggregation analysis on the response record set with K business items as indexes to generate K business item response record sets; a parsing module 12 is used to traverse the K business item response record sets to perform business item service parsing to obtain K business item first response coefficients; an extraction module 13 is used to collect L consulting tickets of the target response platform at the current moment, and to extract historical consulting conversation records with L consulting customers of the L consulting tickets as indexes to obtain L historical consulting conversation record sets, wherein the L consulting tickets include The invention comprises L consulting business items; a feature comparison module 14, wherein the feature comparison module 14 is used to traverse the L historical consulting conversation record sets to perform conversation-related feature comparison and generate L customer consulting coefficients; an identification module 15, wherein the identification module 15 is used to identify answer response coefficients according to the L consulting business items, the L customer consulting coefficients and the first answer coefficients of the K business items to obtain L answer response coefficients; a processing module 16, wherein the processing module 16 is used to perform resource balancing allocation on the target answer platform based on the L answer response coefficients, obtain L fully configured answer processing units, and use the L answer processing units to process the L consulting work orders.

[0052] Furthermore, the parsing module 12 is used to execute the following method:

[0053] Based on the K business item answer record sets, single conversation information entropies of the K business items are extracted to obtain K business item single conversation information entropy sets; repeated question and answer frequencies are extracted from the K business item answer record sets respectively to obtain K business repeated question and answer frequencies; data is extracted from the K business item answer record sets using the repeated question and answer interval time period as an index to obtain K repeated question and answer interval time period sets; the K business item single conversation information entropy sets and the K repeated question and answer interval time period sets are respectively subjected to flood screening analysis to determine K focused conversation information entropies and K focused interval time periods; the service resolution network layer is used to analyze the K business repeated question and answer frequencies, the K focused conversation information entropies and the K focused interval time periods to obtain the first response coefficients of the K business items.

[0054] Furthermore, the parsing module 12 is used to execute the following method:

[0055] Extracting a first business item single conversation information entropy set from the K business item single conversation information entropy sets; extracting a minimum value from the first business item single conversation information entropy set to obtain a first flooding starting point; using the first business item single conversation information entropy set as a simulated image, using each first business item single conversation information entropy as a simulated pixel point, performing flooding screening analysis based on the first flooding starting point to obtain a first focused conversation information entropy; performing flooding screening analysis on the remaining K-1 business item single conversation information entropy sets to obtain K-1 focused conversation information entropies, and combining the first focused conversation information entropies to obtain the K focused conversation information entropies; performing flooding screening analysis on the K repeated question and answer interval time period sets to obtain the K focused interval time periods.

[0056] Furthermore, the parsing module 12 is used to execute the following method:

[0057] Taking the first flooding starting point as the central water filling point, multiple simulated pixel points whose information entropy difference value of the first business item single conversation information entropy minimum value corresponding to the first flooding starting point in the first business item single conversation information entropy set is within a preset information entropy difference threshold value are injected with water of the first color to obtain a first color simulated pixel point set, wherein the information entropy difference value is positive; adding the first business item single conversation information entropy minimum value to the preset information entropy difference threshold value to obtain a second flooding starting point; taking the second flooding starting point as the central water filling point, adding the first business item single conversation information entropy minimum value corresponding to the first flooding starting point in the first business item single conversation information entropy set to the preset information entropy difference threshold value to obtain a second flooding starting point; taking the second flooding starting point as the central water filling point, Water of the second color is injected into multiple simulated pixel points whose information entropy difference at the starting point of the second flood is within a preset information entropy difference threshold to obtain a second color simulated pixel point set, wherein the second color is different from the first color; after multiple water injections, multiple color simulated pixel point sets of the single conversation information entropy set of the first business item are obtained, and according to the mapping relationship between the simulated pixel points and the single conversation information entropy of the first business item, multiple color first business item single conversation information entropy sets are matched to obtain; based on the multiple color first business item single conversation information entropy sets, screening is performed to obtain the first focus conversation information entropy.

[0058] Furthermore, the parsing module 12 is used to execute the following method:

[0059] The information entropy quantities of the single conversation information entropy sets of the multiple color first business items are respectively counted to obtain multiple first information entropy quantities; the multiple first information entropy quantities are sorted in order from large to small, and the N color first business item single conversation information entropy sets corresponding to the first information entropy quantities located in the first n digits are processed by information entropy mean value to obtain the first focused conversation information entropy.

[0060] Furthermore, the feature comparison module 14 is used to perform the following method:

[0061] Keyword groups are extracted from L sets of historical consultation dialogue records respectively to obtain L sets of historical consultation keyword group sequences; conversation-related feature recognition is performed on the L sets of historical consultation keyword group sequences using a semantic similarity identifier to generate L sets of conversation-related feature recognition coefficients; mean calculation is performed on the L sets of conversation-related feature recognition coefficients to obtain the L customer consultation coefficients.

[0062] Furthermore, the identification module 15 is used to execute the following method:

[0063] A response coefficient identifier is constructed, and the response coefficient identifier is used to identify the response coefficients of the L consulting business items, the L customer consulting coefficients and the first response coefficients of the K business items to obtain L response coefficients.

[0064] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0066] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. An AI large model driven customer service intelligent response method, characterized in that: The method comprises: Acquire a response record set of a target response platform within a preset window, perform aggregation analysis on the response record set with K business items as indexes, and generate K business item response record sets; Traversing the K business item response record sets to perform business item service analysis, and obtaining the first response coefficients of the K business items; Collect L consultation work orders of the target answering platform at the current moment, extract historical consultation dialogue records using L consultation customers of the L consultation work orders as indexes, and obtain L historical consultation dialogue record sets, wherein the L consultation work orders include L consultation business items; Traversing the L historical consultation dialogue record sets to perform dialogue correlation feature comparison and generate L customer consultation coefficients; Performing response coefficient identification according to the L consulting business items, the L customer consulting coefficients and the K business item first response coefficients to obtain L response coefficients; Based on the L response coefficients, resources of the target response platform are evenly allocated to obtain L fully configured response processing units, and the L consultation work orders are processed by using the L response processing units; the K business item response record sets are traversed to perform business item service analysis to obtain K business item first response coefficients, including: Extracting single conversation information entropy of the K business items based on the K business item answer record sets to obtain a single conversation information entropy set of the K business items; Extracting repeated question and answer frequencies from the K business item answer record sets respectively to obtain K business repeated question and answer frequencies; Extracting data from the K business item answer record sets using the repeated question-answer interval time period as an index to obtain K repeated question-answer interval time period sets; Performing flood screening analysis on the K business item single conversation information entropy sets and the K repeated question-answer interval time periods to determine K focused conversation information entropies and K focused interval time periods; Analyze the K business repeated question and answer frequencies, the K attention conversation information entropies and the K attention interval time periods by using the service analysis network layer to obtain the first response coefficients of the K business items; The K business item single conversation information entropy sets and the K repeated question-answer interval time periods are respectively subjected to flood screening analysis to determine K focused conversation information entropies and K focused interval time periods, including: Extracting a first business item single conversation information entropy set from the K business item single conversation information entropy sets; Extracting a minimum value from the single conversation information entropy set of the first business item to obtain a first flooding starting point; Taking the first business item single conversation information entropy set as a simulated image, taking each first business item single conversation information entropy as a simulated pixel point, performing flood screening analysis based on the first flood starting point, and obtaining the first focused conversation information entropy; Performing a flood screening analysis on the remaining K-1 business item single conversation information entropy sets to obtain K-1 focused conversation information entropies, and combining the first focused conversation information entropies to obtain the K focused conversation information entropies; Performing a flood screening analysis on the K repeated question-answer interval time periods to obtain the K focus interval time periods; Taking the first flooding starting point as the central water injection point, multiple simulated pixel points whose information entropy difference of the minimum information entropy value of the first business item single conversation corresponding to the first flooding starting point in the first business item single conversation information entropy set is within a preset information entropy difference threshold are injected into water of a first color to obtain a first color simulated pixel point set, wherein the information entropy difference is positive; Adding the minimum information entropy value of a single conversation of the first business item to the preset information entropy difference threshold to obtain a second flooding starting point; Taking the second flooding starting point as the central water injection point, multiple simulated pixel points whose information entropy difference between the first business item single conversation information entropy set and the second flooding starting point is within a preset information entropy difference threshold are injected with water of a second color to obtain a second color simulated pixel point set, wherein the second color is different from the first color; After multiple water injections, multiple color simulation pixel point sets of the first business item single conversation information entropy set are obtained, and multiple color first business item single conversation information entropy sets are matched according to the mapping relationship between the simulation pixel points and the first business item single conversation information entropy; The first focused conversation information entropy is obtained by filtering based on the multiple color first business item single conversation information entropy sets.

2. The AI ​​big model driven customer service intelligent response method according to claim 1, characterized in that: include: The information entropy quantities of the single-conversation information entropy sets of the plurality of color first business items are respectively counted to obtain a plurality of first information entropy quantities; The multiple first information entropy quantities are sorted in descending order, and the N color first business item single conversation information entropy sets corresponding to the first information entropy quantities located in the first n digits are processed by information entropy mean to obtain the first focused conversation information entropy.

3. The AI ​​big model driven customer service intelligent response method according to claim 1, characterized in that: include: Extracting keyword groups from L sets of historical consultation dialogue records respectively to obtain L sets of historical consultation keyword group sequences; Using a semantic similarity identifier to perform dialogue-related feature recognition on the L historical consultation keyword group sequence sets, and generate L dialogue-related feature recognition coefficient sets; The L conversation-related feature recognition coefficient sets are averaged to obtain the L customer consultation coefficients.

4. The AI ​​big model driven customer service intelligent response method according to claim 1, characterized in that: A response coefficient identifier is constructed, and the response coefficient identifier is used to identify the response coefficients of the L consulting business items, the L customer consulting coefficients and the first response coefficients of the K business items to obtain L response coefficients.

5. An AI large model-driven customer service intelligent response system, characterized in that: The system for implementing the AI ​​large model-driven customer service intelligent response method according to any one of claims 1 to 4 comprises: An analysis module, the analysis module is used to obtain a response record set of a target response platform within a preset window, perform aggregation analysis on the response record set with K business items as indexes, and generate K business item response record sets; A parsing module, the parsing module is used to traverse the K business item response record sets to perform business item service parsing and obtain the first response coefficients of the K business items; An extraction module, the extraction module is used to collect L consultation work orders of the target answering platform at the current moment, extract historical consultation dialogue records with L consultation customers of the L consultation work orders as indexes, and obtain L historical consultation dialogue record sets, wherein the L consultation work orders include L consultation business items; A feature comparison module, the feature comparison module is used to traverse the L historical consultation dialogue record sets to perform dialogue-related feature comparison and generate L customer consultation coefficients; an identification module, the identification module being used to identify the response coefficients according to the L consulting business items, the L customer consulting coefficients and the K business item first response coefficients, to obtain L response coefficients; A processing module, the processing module is used to evenly allocate resources for the target response platform based on the L response coefficients, obtain L fully configured response processing units, and use the L response processing units to process the L consulting work orders.

Images