Customer service answering intelligent prompt method and system based on speech data

Abstract

The present invention discloses a method and system for intelligent prompting of customer service responses based on speech data, which relate to the technical field of customer service. A data module is set for each speech data, a summary node is set based on a classification label, and a speech call expression graph is constructed; excellent customer service response records are screened out, the speech data and order of the calls are analyzed and determined, and an excellent response template is formed; according to the call characteristics of the speech data of the excellent response template, a call logic on the graph is constructed, and expressed on the speech call expression graph; real-time customer service responses are obtained, the speech data of the real-time calls are analyzed and determined, and mapped on the speech call expression graph, and based on the distribution characteristics of the excellent response templates, subsequent speech data is predicted and called for prompting. The above technical scheme realizes efficient management and utilization of speech data, provides customer service personnel with real-time and accurate speech prompts, and helps to improve the quality and efficiency of customer service responses.

Description

Technical Field

[0001] The present invention relates to the technical field of customer service management, and in particular to a method and system for intelligent prompting of customer service responses based on speech data. Background Art

[0002] In the field of customer service, efficient response is one of the key factors to improve customer satisfaction and loyalty. However, in the face of increasingly complex customer needs and the ever-changing market environment, traditional manual response methods have gradually exposed many shortcomings. On the one hand, customer service personnel need to master a large amount of speech data in order to make appropriate responses in different situations, but memorizing and retrieving this information is a huge challenge for humans, which often leads to low response efficiency; on the other hand, due to the lack of unified speech management, the response methods of different customer service personnel may vary greatly, affecting the consistency of customer experience. Summary of the invention

[0003] The purpose of the present invention is to provide a customer service answer prompt method and system that can intelligently prompt speech based on real-time conversation records.

[0004] The present invention discloses a customer service answering intelligent prompting method and system based on speech data, including:

[0005] Step S100, classifying and analyzing the dialogue data to form several levels of dialogue data type sets, and configuring classification labels for the dialogue data type sets based on the classification features of each level of the dialogue data type sets;

[0006] Step S200, a data module is set for each speech data, and based on the relationship between the classification labels, a summary node is set between the data corresponding modules, and the summary nodes are connected, and the corresponding data modules are connected to the summary nodes to form a speech call expression graph;

[0007] Step S300, screening out excellent customer service response records, analyzing the excellent customer service response records, determining the called speech data and the order in which the speech data are called, and recording the combination of the called speech data and the order in which the speech data are called as an excellent response template;

[0008] Step S400: Based on the call features of the excellent response template for the speech data, the corresponding data modules are connected, and based on the call order of the speech data, the directions of the connection lines are marked to obtain the call logic on the diagram, and several call logics on the diagram are expressed on the same speech call expression diagram. If two adjacent data modules do not belong to the same summary node, they need to go through the summary node to which they belong in the process of connecting to each other;

[0009] Step S500, obtain the real-time customer service response, analyze the real-time customer service response, determine the real-time call speech data, map the real-time call speech data on the speech call expression diagram, and based on the distribution characteristics of the excellent response template on the speech call expression diagram, determine the data module for subsequent mapping of the real-time customer service response, and call the corresponding speech data for prompting.

[0010] In some embodiments disclosed in the present invention, the method for classifying and analyzing the dialogue data includes:

[0011] Step 101, based on the preset speech type, the speech data is first classified, based on the keywords and the order of the keywords appearing in the speech data, the meaning expression scope of the speech data is determined, and the speech data after the first classification is secondly classified;

[0012] Step S102, using semantic analysis technology, independently split each speech data after the second classification, the splitting method includes symbol segmenting the speech data, and locking the keywords in each speech data segment. If the locked keywords do not exist in the segments after the preset number of speech data segments, the current speech data segment is judged to be a node speech data segment, and the combination of several speech data segments between the node speech data segments and the subsequent node speech data segments is identified as an independent split of the speech data, which is recorded as a speech data segment group;

[0013] Step S103, record the combination of several speech data segment groups after the second classification as a speech data type set, and arrange the speech data type set in sequence to form several levels of speech data type sets.

[0014] In some embodiments disclosed in the present invention, a method for arranging a set of speech data types in sequence includes:

[0015] Step S104, a plurality of meaning expression scope sequences are preset, and based on the meaning expression scopes corresponding to the discourse data type sets, the discourse data type sets are sorted to form a discourse data type set sequence.

[0016] In some embodiments disclosed in the present invention, the classification label of a speech type set includes the sequence of speech data type sets to which it belongs, and its order in the sequence of speech data type sets.

[0017] In some embodiments disclosed in the present invention, the method of analyzing excellent customer service response records and determining the speech data to be called and the order of calling the speech data includes:

[0018] Step S301, analyzing the excellent customer service response records, determining the keywords, the keyword order of the keywords, and the word intervals between the keywords in the excellent customer service response records, and recording the determined combination of keywords, keyword order, and word intervals as a keyword statistical feature group;

[0019] Step S302: Based on the preset emphasis keyword table, the keyword statistical feature group is scanned and analyzed, and the mapped keywords are emphasized and marked, and based on the emphasized and marked keywords, a number of speech data containing equivalent keywords are called out, and the equivalent keywords are emphasized and marked, and the called out speech data are recorded as the speech data for comparison;

[0020] Step S303, determine the keywords, keyword order and word interval of the comparison speech data, obtain the keyword statistical feature group, substitute the keyword statistical feature group of the comparison speech data into the keyword statistical feature group of the excellent customer service answer record according to the requirements corresponding to the key words marked with emphasis, and if the degree of coincidence between the keyword statistical feature groups meets the preset standard, the corresponding comparison speech data is used as the called speech data;

[0021] Step S304, determine the called speech data, and determine the calling order of the speech data according to the position of the keyword statistical feature group of each speech data relative to the keyword statistical feature group of the excellent customer service response record.

[0022] In some embodiments disclosed in the present invention, the method for determining whether the degree of consistency between keyword statistical feature groups meets the preset standard includes:

[0023] Step S3031, determining a statistical feature group segment for comparison in the keyword statistical feature group corresponding to the excellent customer service answer record, and comparing the statistical feature group segment with the keyword statistical feature group of the speech data, and calculating the keyword matching ratio, keyword order matching ratio and word spacing matching ratio;

[0024] Step S3032: If the keyword matching ratio, keyword order matching ratio and word spacing matching ratio are all greater than or equal to the preset values, it is determined that the degree of coincidence between the keyword statistical feature groups meets the preset standard.

[0025] In some embodiments disclosed in the present invention, based on the distribution characteristics of excellent response templates on the speech call expression graph, a method for determining a data module for subsequent mapping of a real-time customer service response includes:

[0026] Step S501, analyzing the real-time call speech data, determining the data modules mapped by the real-time call speech data on the speech call expression diagram and the calling order of the data modules, connecting the mapped data modules, and marking the upward direction of the connection lines between the data modules based on the calling order of the data modules, to form a real-time response template;

[0027] Step S502, selecting excellent answer templates whose similarity with the front section of the real-time answer template on the speech call expression diagram is greater than or equal to a preset value, and sorting the excellent answer templates based on the similarity, and marking the excellent answer templates in the preset order before sorting as triggered;

[0028] Step S503, based on the template ratio parameter of the excellent response templates marked as triggered to all excellent response templates and the similarity with the real-time response template, determine the locked excellent response template, and identify the locked excellent response template as a reference object for the subsequent call speech data of the real-time customer service response.

[0029] In some embodiments disclosed in the present invention, the method for determining a locked excellent answer template includes:

[0030] Step S5031, comparing the real-time answer template with the excellent answer template, including comparing the data modules mapped on the speech call diagram and the connection lines between the data modules, and constructing a similarity operator between the excellent answer template and the real-time answer template based on a first matching ratio of the data modules and a second matching ratio of the connection lines;

[0031] Step S5032, based on the similarity operator and the template ratio parameter, determining the degree of adaptation between different excellent response templates and the real-time response template, and determining the locked excellent response template based on the degree of adaptation;

[0032] Among them, the expression for calculating the degree of adaptation is:

[0033] ;

[0034] Among them, P is the degree of adaptation, Adjust the weight coefficients for module matching, The weight coefficient is adjusted to match the connection line. n is the number of all data modules corresponding to the real-time response template. is the matching judgment function of the ith data module corresponding to the real-time response template. If the ith data module has a matching data module in the excellent response template, then Output 1, is the matching judgment function of the ith connection line corresponding to the real-time response template. If the ith connection line has a matching connection line in the excellent response template, then Output 1, is the template ratio parameter, is the ratio adjustment coefficient of the proportion parameter, and b is the proportion parameter adjustment constant.

[0035] In some embodiments disclosed in the present invention, there is also disclosed a customer service answering intelligent prompting system based on speech data, including:

[0036] The first module is used to classify and analyze the dialogue data to form several levels of dialogue data type sets, and configure classification labels for the dialogue data type sets based on the classification characteristics of each level of the dialogue data type sets;

[0037] The second module is used to set a data module for each speech data, and based on the relationship between the classification labels, set the summary nodes between the data corresponding modules, connect the summary nodes, and connect the corresponding data modules to the summary nodes to form a speech call expression diagram;

[0038] The third module is used to screen out excellent customer service response records, analyze the excellent customer service response records, determine the called speech data and the order in which the speech data are called, and record the combination of the called speech data and the order in which the speech data are called as an excellent response template;

[0039] The fourth module is used to connect the corresponding data modules based on the call features of the dialogue data of the excellent response template, and mark the direction of the connection line based on the call order of the dialogue data to obtain the call logic on the graph, and express several call logics on the graph on the same dialogue call expression graph. Among them, if two adjacent data modules do not belong to the same summary node, they need to go through the summary node to which they belong in the process of connecting with each other;

[0040] The fifth module is used to obtain real-time customer service responses, analyze the real-time customer service responses, determine the real-time call speech data, map the real-time call speech data on the speech call expression diagram, and based on the distribution characteristics of excellent response templates on the speech call expression diagram, determine the data module for subsequent mapping of the real-time customer service responses, and call the corresponding speech data for prompts.

[0041] The present invention discloses a method and system for intelligent prompting of customer service responses based on speech data, which relate to the technical field of customer service. A data module is set for each speech data, a summary node is set based on a classification label, and a speech call expression graph is constructed; excellent customer service response records are screened out, the speech data and order of the calls are analyzed and determined, and an excellent response template is formed; according to the call characteristics of the speech data of the excellent response template, a call logic on the graph is constructed, and expressed on the speech call expression graph; real-time customer service responses are obtained, the speech data of the real-time calls are analyzed and determined, and mapped on the speech call expression graph, and based on the distribution characteristics of the excellent response templates, subsequent speech data is predicted and called for prompting. The above technical scheme realizes efficient management and utilization of speech data, provides customer service personnel with real-time and accurate speech prompts, and helps to improve the quality and efficiency of customer service responses.

[0042] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a method step diagram of a method for intelligent prompting of customer service responses based on speech data disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0045] The following will be combined with the accompanying drawings and specific embodiments to clearly and completely describe the technical solution of the present invention. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and cannot be understood as limiting the scope of protection of the present invention. Those skilled in the art in this field can make some non-essential improvements and adjustments based on the content of the present invention described below. In the present invention, unless otherwise clearly specified and limited, the technical terms used in the present invention should be the common meanings understood by the technical personnel described in the present invention.

[0046] Example:

[0047] The purpose of the present invention is to provide a customer service answer prompt method and system that can intelligently prompt speech based on real-time conversation records.

[0048] The present invention discloses a method and system for intelligent prompting of customer service response based on speech data, see Figure 1 ,include:

[0049] Step S100, classify and analyze the dialogue data to form several levels of dialogue data type sets, and configure classification labels for the dialogue data type sets based on the classification characteristics of each level of the dialogue data type sets.

[0050] The purpose of this step is to systematically organize and classify a large amount of disordered speech data for subsequent management and use. By analyzing the content of the speech, the commonalities and differences between different speech are identified, and then they are classified into different levels of data types. Each level of data type set reflects a certain specific attribute or function of the speech data. For example, it can be classified according to the topic, usage scenario, emotional tendency, etc. of the speech. For example, suppose there is a batch of customer service speech data, including different types of responses such as product introduction, after-sales service, and complaint handling. Through classification analysis, these speech can be divided into data type sets at the levels of "product introduction", "after-sales service", "complaint handling", etc., and the corresponding classification label is configured for each type set.

[0051] In some embodiments disclosed in the present invention, the method for classifying and analyzing the dialogue data includes:

[0052] Step 101, based on the preset speech type, the speech data is classified for the first time, based on the keywords appearing in the speech data and the order of the keywords, the meaning expression scope of the dialogue data is determined, and the speech data after the first classification is classified for the second time.

[0053] The core of this step is to make a preliminary classification of the speech data according to the pre-defined speech type framework. The preset speech types may be set based on multiple dimensions such as business scenarios, customer problem types, emotional tendencies, etc. By identifying the keywords and their order in the speech data, the system can determine the main meaning or intention expressed by the speech data and classify it into the most appropriate preset speech type. This process is similar to the text classification task and is widely used in the field of natural language processing (NLP). Through the first classification, a large amount of disordered speech data can be organized into a structured form, laying the foundation for subsequent in-depth analysis.

[0054] Step S102, using semantic analysis technology, independently split each speech data after the second classification. The splitting method includes symbol segmenting the speech data and locking the keywords in each speech data segment. If the locked keywords do not exist in the segments after the preset number of speech data segments, the current speech data segment is judged to be a node speech data segment, and the combination of several speech data segments between the node speech data segments and the subsequent node speech data segments is identified as an independent split of the speech data, which is recorded as a speech data segment group.

[0055] This step aims to further refine the speech data and split it into smaller independent units for more precise understanding and use. Semantic analysis technology is the key, which allows the system to understand the deep meaning and contextual relationships in the text. The speech data is initially segmented by symbolic segmentation (such as punctuation marks such as commas and periods), and the keywords in each segment are locked. The system can determine which segments are semantically independent. In particular, when the locked keywords do not appear in a certain number of segments after a segment, the segment is regarded as a node speech data segment, marking the transition of the semantic content before and after. In this way, the system can automatically split the speech data into several speech data segment groups, the segments within each group are closely related semantically, and the groups are relatively independent.

[0056] Step S103, record the combination of several speech data segment groups after the second classification as a speech data type set, and arrange the speech data type set in sequence to form several levels of speech data type sets.

[0057] This step is to abstract and summarize the dialogue data at a higher level based on the previous two steps. By grouping the dialogue data into segments after the second classification into a dialogue data type set, and arranging them in sequence according to the internal connections between them, several levels of dialogue data type sets are formed. This hierarchical structure not only helps to systematically manage and query dialogue data, but also reveals the logical relationships and dependencies between different dialogue data. For customer service personnel or automated systems, this means that dialogue data can be called and combined more flexibly to cope with complex customer service scenarios and improve the accuracy and efficiency of responses.

[0058] In some embodiments disclosed in the present invention, a method for arranging a set of speech data types in sequence includes:

[0059] Step S104, a plurality of meaning expression scope sequences are preset, and based on the meaning expression scopes corresponding to the discourse data type sets, the discourse data type sets are sorted to form a discourse data type set sequence.

[0060] Step S200, a data module is set for each speech data, and based on the relationship between the classification labels, a summary node is set between the data corresponding modules, and the summary nodes are connected, and the corresponding data modules are connected to the summary nodes to form a speech call expression diagram.

[0061] This step is based on step S100, and the classified speech data is represented in a graphical way to facilitate more intuitive understanding and calling. By setting data modules and summary nodes, these data modules are connected through summary nodes according to the relationship between classification labels to form a speech call expression diagram. In this way, when a certain speech needs to be called, the corresponding data module can be quickly located, and the relevant speech can be found along the path in the diagram. For example, suppose that in the "product introduction" data type set, there are multiple introduction speeches about different product features. When constructing a speech call expression diagram, these introduction speeches can be set as data modules respectively, and summary nodes can be set according to the product categories to which they belong (such as "electronic products", "household items", etc.). Then, these data modules are connected through summary nodes to form a clear speech call expression diagram.

[0062] Step S300, screen out excellent customer service response records, analyze the excellent customer service response records, determine the called speech data and the order in which the speech data is called, and record the combination of the called speech data and the order in which the speech data is called as an excellent response template.

[0063] The purpose of this step is to select high-quality responses from a large number of customer service response records as samples for subsequent analysis and learning. Through manual review or machine learning algorithms, those response records with clear expression, strict logic, and high customer satisfaction can be identified and saved and analyzed as excellent customer service response records. In actual operation, a series of evaluation indicators (such as response time, resolution rate, customer satisfaction, etc.) can be set to score customer service response records. Then, based on the scoring results, the response records with higher scores are selected as excellent customer service response records. For example, a response record about a product return was selected as an excellent customer service response record because of its quick response, clear solution, and high evaluation by customers.

[0064] In some embodiments disclosed in the present invention, the method of analyzing excellent customer service response records and determining the speech data to be called and the order of calling the speech data includes:

[0065] Step S301, analyze the excellent customer service response records, determine the keywords, keyword order of keywords and word intervals between keywords in the excellent customer service response records, and record the determined combination of keywords, keyword order and word intervals as a keyword statistical feature group.

[0066] The purpose of this step is to extract key information from excellent customer service response records for subsequent matching and calling of speech data. By analyzing the keywords, the order of keywords, and the word spacing between them in the response records, a unique set of keyword statistical feature groups can be formed. These feature groups reflect the key elements and arrangement of speech data in excellent responses, and are the basis for subsequent matching and calling of speech data.

[0067] Step S302, based on the preset emphasis keyword table, the keyword statistical feature group is scanned and analyzed, the mapped keywords are marked with emphasis, and based on the marked keywords, several speech data containing equivalent keywords are called out, and the equivalent keywords are marked with emphasis, and the called out speech data are recorded as speech data for comparison.

[0068] This step introduces a preset emphasis keyword table for more accurate screening and marking of the speech data. By scanning and analyzing the keyword statistical feature group and comparing the mapped keywords with the emphasis keyword table, these keywords can be marked with emphasis. Then, based on these emphasized keywords, speech data containing the same keywords are called out from the speech database and marked with emphasis accordingly. These called speech data will be used as candidate sets for subsequent comparisons.

[0069] Step S303, determine the keywords, keyword order and word spacing of the comparison speech data, obtain the keyword statistical feature group, and substitute the keyword statistical feature group of the comparison speech data into the keyword statistical feature group of the excellent customer service response record according to the corresponding requirements of the emphasized marked keywords. If the degree of consistency between the keyword statistical feature groups meets the preset standard, the corresponding comparison speech data will be used as the called speech data.

[0070] This step is to further screen and determine the comparison speech data called in the previous step. By comparing the degree of consistency between the keyword statistical feature groups of these speech data and the keyword statistical feature groups of excellent customer service response records, it can be determined which speech data is more consistent with the characteristics of excellent responses. When the degree of consistency meets the preset standard, the corresponding speech data is determined as the speech data that needs to be called.

[0071] Step S304, determine the called speech data, and determine the calling order of the speech data according to the position of the keyword statistical feature group of each speech data relative to the keyword statistical feature group of the excellent customer service response record.

[0072] This step is to further determine the calling order of these data after determining the called speech data. Since the arrangement order of speech data in the response has an important impact on the meaning and effect of the expression, it is necessary to determine their calling order based on the position of the keyword statistical feature group of each speech data in the keyword statistical feature group of the excellent customer service response record. In this way, it can be ensured that when a new response is generated, the arrangement order of the speech data is consistent with the excellent response, thereby improving the quality and efficiency of the response.

[0073] In some embodiments disclosed in the present invention, the method for determining whether the degree of consistency between keyword statistical feature groups meets the preset standard includes:

[0074] Step S3031, determine the statistical feature group segment used for comparison in the keyword statistical feature group corresponding to the excellent customer service response record, and compare the statistical feature group segment with the keyword statistical feature group of the speech data to calculate the keyword matching ratio, keyword order matching ratio and word spacing matching ratio.

[0075] Step S3032: If the keyword matching ratio, keyword order matching ratio and word spacing matching ratio are all greater than or equal to the preset values, it is determined that the degree of coincidence between the keyword statistical feature groups meets the preset standard.

[0076] Step S400, based on the calling features of the excellent response template for the dialogue data, the corresponding data modules are connected, and based on the calling order of the dialogue data, the directions of the connection lines are marked to obtain the calling logic on the graph, and several calling logics on the graph are expressed on the same dialogue call expression graph. Among them, if two adjacent data modules do not belong to the same summary node, then in the process of connecting with each other, they need to go through the summary node to which they belong.

[0077] This step is based on step S300, and according to the call features of the speech data in the excellent response template, the call logic is constructed on the speech call expression diagram. By analyzing and extracting the call sequence and dependency of the speech data in the excellent response template, the connection method and direction between the data modules can be determined, thereby forming a clear call logic on the diagram. In this way, when a new response needs to be generated, the speech that meets the requirements can be quickly combined according to the call logic. For example, suppose that an excellent response template contains a speech call sequence of greeting the customer first, then confirming the problem, and finally giving a solution. When constructing the call logic on the diagram, the corresponding data modules can be connected according to this order, and the connection direction can be ensured to be correct. In this way, when a new response is generated, it is only necessary to call the corresponding data modules in sequence according to the call logic on the diagram.

[0078] Step S500, obtain the real-time customer service response, analyze the real-time customer service response, determine the real-time call speech data, map the real-time call speech data on the speech call expression diagram, and based on the distribution characteristics of the excellent response template on the speech call expression diagram, determine the data module for subsequent mapping of the real-time customer service response, and call the corresponding speech data for prompting.

[0079] This step is to apply the above technical solution to the real-time customer service response scenario to provide intelligent prompts and assistance to customer service personnel. By analyzing the content of the customer service response in real time, the current call speech data is determined and mapped to the speech call expression graph. Then, based on the distribution characteristics of the excellent response template on the graph, the speech data that may need to be called later is predicted and prompted to help customer service personnel generate responses more quickly and accurately. In actual operation, when the customer service personnel start to enter the response content, the system can capture and analyze the content in real time. If it is identified that a certain speech data is currently being called (such as "Hello, how can I help you?"), the data module is located on the speech call expression graph. Then, based on the call logic on the graph and the distribution characteristics of the excellent response template, the speech data that may need to be called later (such as "For the problem you mentioned, we can provide the following solutions...") is predicted and prompted. In this way, customer service personnel can generate high-quality responses more efficiently with the assistance of the system.

[0080] In some embodiments disclosed in the present invention, the classification label of a speech type set includes the sequence of speech data type sets to which it belongs, and its order in the sequence of speech data type sets.

[0081] In some embodiments disclosed in the present invention, based on the distribution characteristics of excellent response templates on the speech call expression graph, a method for determining a data module for subsequent mapping of a real-time customer service response includes:

[0082] Step S501, analyze the real-time call speech data, determine the data modules mapped by the real-time call speech data on the speech call expression diagram and the calling order of the data modules, connect the mapped data modules, and based on the calling order of the data modules, mark the connection lines between the data modules in the upward direction to form a real-time response template.

[0083] Step S502, screen out excellent response templates whose similarity with the front section of the real-time response template on the speech call expression diagram is greater than or equal to a preset value, and sort the excellent response templates based on the similarity, and mark the excellent response templates in the preset order before sorting as triggered.

[0084] Step S503, based on the template ratio parameter of the excellent response templates marked as triggered to all excellent response templates and the similarity with the real-time response template, determine the locked excellent response template, and identify the locked excellent response template as a reference object for the subsequent call speech data of the real-time customer service response.

[0085] In some embodiments disclosed in the present invention, the method for determining a locked excellent answer template includes:

[0086] Step S5031, compare the real-time response template with the excellent response template, including comparing the data modules mapped on the speech call diagram and the connecting lines between the data modules, and constructing a similarity operator between the excellent response template and the real-time response template based on the first matching ratio of the data modules and the second matching ratio of the connecting lines.

[0087] Step S5032: Based on the similarity operator and the template ratio parameter, the degree of adaptation between different excellent response templates and the real-time response template is determined, and based on the degree of adaptation, the locked excellent response template is determined.

[0088] Among them, the expression for calculating the degree of adaptation is:

[0089] .

[0090] Among them, P is the degree of adaptation, Adjust the weight coefficients for module matching, The weight coefficient is adjusted to match the connection line. n is the number of all data modules corresponding to the real-time response template. is the matching judgment function of the ith data module corresponding to the real-time response template. If the ith data module has a matching data module in the excellent response template, then Output 1, is the matching judgment function of the ith connection line corresponding to the real-time response template. If the ith connection line has a matching connection line in the excellent response template, then Output 1, is the template ratio parameter, is the ratio adjustment coefficient of the proportion parameter, and b is the proportion parameter adjustment constant.

[0091] In some embodiments disclosed in the present invention, there is also disclosed a customer service answering intelligent prompting system based on speech data, including:

[0092] The first module is used to classify and analyze the dialogue data to form several levels of dialogue data type sets, and configure classification labels for the dialogue data type sets based on the classification characteristics of each level of the dialogue data type sets;

[0093] The second module is used to set a data module for each speech data, and based on the relationship between the classification labels, set the summary nodes between the data corresponding modules, connect the summary nodes, and connect the corresponding data modules to the summary nodes to form a speech call expression diagram;

[0094] The third module is used to screen out excellent customer service response records, analyze the excellent customer service response records, determine the called speech data and the order in which the speech data are called, and record the combination of the called speech data and the order in which the speech data are called as an excellent response template;

[0095] The fourth module is used to connect the corresponding data modules based on the call features of the dialogue data of the excellent response template, and mark the direction of the connection line based on the call order of the dialogue data to obtain the call logic on the graph, and express several call logics on the graph on the same dialogue call expression graph. Among them, if two adjacent data modules do not belong to the same summary node, they need to go through the summary node to which they belong in the process of connecting with each other;

[0096] The fifth module is used to obtain real-time customer service responses, analyze the real-time customer service responses, determine the real-time call speech data, map the real-time call speech data on the speech call expression diagram, and based on the distribution characteristics of excellent response templates on the speech call expression diagram, determine the data module for subsequent mapping of the real-time customer service responses, and call the corresponding speech data for prompts.

[0097] The present invention discloses a method and system for intelligent prompting of customer service responses based on speech data, which relate to the technical field of customer service. A data module is set for each speech data, a summary node is set based on a classification label, and a speech call expression graph is constructed; excellent customer service response records are screened out, the speech data and order of the calls are analyzed and determined, and an excellent response template is formed; according to the call characteristics of the speech data of the excellent response template, a call logic on the graph is constructed, and expressed on the speech call expression graph; real-time customer service responses are obtained, the speech data of the real-time calls are analyzed and determined, and mapped on the speech call expression graph, and based on the distribution characteristics of the excellent response templates, subsequent speech data is predicted and called for prompting. The above technical scheme realizes efficient management and utilization of speech data, provides customer service personnel with real-time and accurate speech prompts, and helps to improve the quality and efficiency of customer service responses.

[0098] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A customer service response intelligent prompt method based on speech data, characterized in that: include: Step S100, classifying and analyzing the dialogue data to form several levels of dialogue data type sets, and configuring classification labels for the dialogue data type sets based on the classification features of each level of the dialogue data type sets; Step S200, a data module is set for each speech data, and based on the relationship between the classification labels, a summary node is set between the data corresponding modules, and the summary nodes are connected, and the corresponding data modules are connected to the summary nodes to form a speech call expression graph; Step S300, screening out excellent customer service response records, analyzing the excellent customer service response records, determining the called speech data and the order in which the speech data are called, and recording the combination of the called speech data and the order in which the speech data are called as an excellent response template; Step S400: Based on the call features of the excellent response template for the speech data, the corresponding data modules are connected, and based on the call order of the speech data, the directions of the connection lines are marked to obtain the call logic on the diagram, and several call logics on the diagram are expressed on the same speech call expression diagram. If two adjacent data modules do not belong to the same summary node, they need to go through the summary node to which they belong in the process of connecting to each other; Step S500, obtaining a real-time customer service response, analyzing the real-time customer service response, determining the real-time call speech data, mapping the real-time call speech data on the speech call expression graph, and determining the data module for subsequent mapping of the real-time customer service response based on the distribution characteristics of the excellent response template on the speech call expression graph, and calling the corresponding speech data for prompting; Based on the distribution characteristics of excellent response templates on the speech call expression diagram, the method for determining the data module for subsequent mapping of real-time customer service response includes: Step S501, analyzing the real-time call speech data, determining the data modules mapped by the real-time call speech data on the speech call expression diagram and the calling order of the data modules, connecting the mapped data modules, and marking the upward direction of the connection lines between the data modules based on the calling order of the data modules, to form a real-time response template; Step S502, selecting excellent answer templates whose similarity with the front section of the real-time answer template on the speech call expression diagram is greater than or equal to a preset value, and sorting the excellent answer templates based on the similarity, and marking the excellent answer templates in the preset order before sorting as triggered; Step S503, based on the ratio parameter of the excellent answer templates marked as triggered to all excellent answer templates and the similarity with the real-time answer template, determine the locked excellent answer template, and identify the locked excellent answer template as a reference object for the subsequent call speech data of the real-time customer service answer; Methods for identifying good answer templates to target include: Step S5031, comparing the real-time answer template with the excellent answer template, including comparing the data modules mapped on the speech call diagram and the connection lines between the data modules, and constructing a similarity operator between the excellent answer template and the real-time answer template based on a first matching ratio of the data modules and a second matching ratio of the connection lines; Step S5032, based on the similarity operator and the template ratio parameter, determining the degree of adaptation between different excellent response templates and the real-time response template, and determining the locked excellent response template based on the degree of adaptation; Among them, the expression for calculating the degree of adaptation is: ； Among them, P is the degree of adaptation, Adjust the weight coefficients for module matching, The weight coefficient is adjusted to match the connection line. n is the number of all data modules corresponding to the real-time response template. is the matching judgment function of the ith data module corresponding to the real-time response template. If the ith data module has a matching data module in the excellent response template, then Output 1, is the matching judgment function of the ith connection line corresponding to the real-time response template. If the ith connection line has a matching connection line in the excellent response template, then Output 1, is the template ratio parameter, is the ratio adjustment coefficient of the proportion parameter, is the proportional parameter adjustment constant.

2. A customer service response intelligent prompt method based on speech data, characterized in that: Methods for classifying and analyzing dialogue data include: Step 101, based on the preset speech type, the speech data is first classified, based on the keywords and the order of the keywords appearing in the speech data, the meaning expression scope of the speech data is determined, and the speech data after the first classification is secondly classified; Step S102, using semantic analysis technology, independently split each speech data after the second classification, the splitting method includes symbol segmenting the speech data, and locking the keywords in each speech data segment. If the locked keywords do not exist in the segments after the preset number of speech data segments, the current speech data segment is judged to be a node speech data segment, and the combination of several speech data segments between the node speech data segments and the subsequent node speech data segments is identified as an independent split of the speech data, which is recorded as a speech data segment group; Step S103, record the combination of several speech data segment groups after the second classification as a speech data type set, and arrange the speech data type set in sequence to form several levels of speech data type sets.

3. The method for intelligent prompting of customer service response based on speech data according to claim 2 is characterized in that: Methods for arranging the set of speech data types in sequence include: Step S104, a plurality of meaning expression scope sequences are preset, and based on the meaning expression scopes corresponding to the discourse data type sets, the discourse data type sets are sorted to form a discourse data type set sequence.

4. The method for intelligent prompting of customer service response based on speech data according to claim 3 is characterized in that: The classification label of a speech type set includes the sequence of speech data type sets to which it belongs and its order in the sequence of speech data type sets.

5. The method for intelligent prompting of customer service response based on speech data according to claim 1 is characterized in that: The method of analyzing excellent customer service response records and determining the speech data to be called and the order in which the speech data is called includes: Step S301, analyzing the excellent customer service response records, determining the keywords, the keyword order of the keywords, and the word intervals between the keywords in the excellent customer service response records, and recording the determined combination of keywords, keyword order, and word intervals as a keyword statistical feature group; Step S302: Based on the preset emphasis keyword table, the keyword statistical feature group is scanned and analyzed, and the mapped keywords are emphasized and marked, and based on the emphasized and marked keywords, a number of speech data containing equivalent keywords are called out, and the equivalent keywords are emphasized and marked, and the called out speech data are recorded as the speech data for comparison; Step S303, determine the keywords, keyword order and word interval of the comparison speech data, obtain the keyword statistical feature group, substitute the keyword statistical feature group of the comparison speech data into the keyword statistical feature group of the excellent customer service response record according to the requirements corresponding to the key words marked with emphasis, and if the degree of coincidence between the keyword statistical feature groups meets the preset standard, the corresponding comparison speech data is used as the called speech data; Step S304, determine the called speech data, and determine the calling order of the speech data according to the position of the keyword statistical feature group of each speech data relative to the keyword statistical feature group of the excellent customer service response record.

6. The method for intelligent prompting of customer service response based on speech data according to claim 5 is characterized in that: Methods for determining whether the degree of agreement between keyword statistical feature groups meets the preset standard include: Step S3031, determining a statistical feature group segment for comparison in the keyword statistical feature group corresponding to the excellent customer service answer record, and comparing the statistical feature group segment with the keyword statistical feature group of the speech data, and calculating the keyword matching ratio, keyword order matching ratio and word spacing matching ratio; Step S3032: If the keyword matching ratio, keyword order matching ratio and word spacing matching ratio are all greater than or equal to the preset values, it is determined that the degree of coincidence between the keyword statistical feature groups meets the preset standard.

7. The customer service response intelligent prompt system based on speech data is characterized by: The method for executing the customer service answering intelligent prompt method according to any one of claims 1 to 6 comprises: The first module is used to classify and analyze the dialogue data to form several levels of dialogue data type sets, and configure classification labels for the dialogue data type sets based on the classification characteristics of each level of the dialogue data type sets; The second module is used to set a data module for each speech data, and based on the relationship between the classification labels, set the summary nodes between the data corresponding modules, connect the summary nodes, and connect the corresponding data modules to the summary nodes to form a speech call expression diagram; The third module is used to screen out excellent customer service response records, analyze the excellent customer service response records, determine the called speech data and the order in which the speech data are called, and record the combination of the called speech data and the order in which the speech data are called as an excellent response template; The fourth module is used to connect the corresponding data modules based on the call features of the dialogue data of the excellent response template, and mark the direction of the connection line based on the call order of the dialogue data to obtain the call logic on the graph, and express several call logics on the graph on the same dialogue call expression graph. Among them, if two adjacent data modules do not belong to the same summary node, they need to go through the summary node to which they belong in the process of connecting with each other; The fifth module is used to obtain real-time customer service responses, analyze the real-time customer service responses, determine the real-time call speech data, map the real-time call speech data on the speech call expression diagram, and based on the distribution characteristics of excellent response templates on the speech call expression diagram, determine the data module for subsequent mapping of the real-time customer service responses, and call the corresponding speech data for prompts.

Images