Customer service quality inspection method and device and electronic equipment

Abstract

This invention relates to the field of service quality inspection technology, and discloses a customer service quality inspection method, apparatus, and electronic device. The method involves converting service call recordings into corresponding voice-to-text; obtaining semantic slices corresponding to the voice-to-text; determining the target service node corresponding to the first semantic slice; determining the target quality inspection rules corresponding to the target service node and performing rule quality inspection to obtain rule quality inspection results; performing large-scale model quality inspection on the first semantic slice based on a large model to obtain large-scale model quality inspection results; and cross-validating the rule quality inspection results and the large-scale model quality inspection results to obtain cross-validation results. By determining the target service node for rule quality inspection of the first semantic slice, rule writers can design quality inspection conditions for individual nodes, reducing the difficulty of rule writing. Using a large model for large-scale model quality inspection improves semantic understanding during the quality inspection process. Cross-validating the rule quality inspection results and the large-scale model quality inspection results yields more accurate service quality inspection results.

Description

Technical Field

[0001] This invention relates to the field of service quality inspection technology, specifically to customer service quality inspection methods, devices, and electronic equipment. Background Technology

[0002] In customer service, especially in after-sales service calls, effective quality control of customer service personnel is crucial for ensuring customer experience and mitigating corporate risks. However, current intelligent voice process quality control technologies rely on traditional regular expression matching methods. This requires exhaustively listing all possible business keywords and writing precise regular expressions, a process that is labor-intensive, time-consuming, and demands high levels of expertise from personnel, resulting in low overall implementation efficiency. Furthermore, relying on regular expression matching methods only allows for the recognition of preset keywords and cannot truly understand the contextual relationships of the voice content. Its semantic understanding is weak, making it difficult to accurately judge call quality in complex contexts, polysemous expressions, or colloquialisms, leading to frequent misjudgments or omissions. Additionally, fixed regular expressions are difficult to adapt to new needs, requiring continuous adjustment and optimization, which not only increases maintenance costs but may also affect the real-time performance and accuracy of quality control. Summary of the Invention

[0003] This invention provides a customer service quality inspection method, device, and electronic device to solve the problems of time-consuming and labor-intensive rule writing and the inability to truly understand the contextual relationships of voice content in related technologies.

[0004] In a first aspect, the present invention provides a customer service quality inspection method, comprising: Acquire service call recordings between customers and customer service representatives, and convert the service call recordings into corresponding voice-to-text. The speech text is segmented based on semantic integrity to obtain at least one semantically complete semantic slice; the semantic slice includes role tags, which include customer and customer service. For any first semantic slice with the role label "customer service", the target service node corresponding to the first semantic slice is determined based on multiple pre-defined service nodes; Determine the target quality inspection rules corresponding to the target service node, perform rule quality inspection on the first semantic slice according to the target quality inspection rules, and obtain the rule quality inspection results; Based on the large model, a large model quality check is performed on the first semantic slice to obtain the large model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node. Cross-validation is performed on the rule quality inspection results and the large model quality inspection results to obtain the cross-validation results; the cross-validation results are used to represent the service quality inspection results corresponding to the first semantic slice.

[0005] The customer service quality inspection method provided in this embodiment converts service call recordings into corresponding voice-text and semantic slices, determines the target service node corresponding to the first semantic slice, and performs rule-based quality inspection on the first semantic slice based on the target service node. This ensures that the rule-based quality inspection is only effective within the target service node, and rule writers only need to design quality inspection conditions for a single node, reducing the difficulty of rule writing. Furthermore, using a large-scale model for quality inspection helps understand complex contexts and handle colloquial and grammatically incorrect expressions, improving semantic understanding during the quality inspection process. By combining rule-based quality inspection with large-scale model quality inspection, the results of the rule-based and large-scale model quality inspections are cross-validated to obtain more accurate service quality inspection results.

[0006] In some optional implementations, the target quality inspection rules corresponding to the target service node are determined, and the first semantic slice is subjected to rule quality inspection according to the target quality inspection rules to obtain the rule quality inspection results, including: For any quality inspection rule used to determine whether a violation has occurred, determine the set of effective nodes for the quality inspection rule, and take the quality inspection rule that includes the target service node in the set of effective nodes as the target quality inspection rule; Based on the slice information of the first semantic slice, determine whether the first semantic slice meets the triggering conditions of the target quality inspection rule; the slice information includes the time information, text information and node identifier of the target service node of the first semantic slice, and the triggering conditions include at least one of the time conditions, text conditions and node conditions; If the first semantic slice satisfies the triggering condition of the target quality inspection rule, determine the rule quality inspection result indicating the violation; If the first semantic slice does not meet the triggering conditions of the target quality inspection rule, determine the rule quality inspection result that indicates compliance.

[0007] In some optional implementations, a large model quality check is performed on the first semantic slice based on the large model to obtain a large model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node, including: Obtain the target service requirements of the target service node, and construct a large model prompt word containing the first semantic slice and the target service requirements; Input the large model prompt words into the large model to instruct the large model to perform a large model quality check on the first semantic slice based on the large model prompt words, and obtain the large model quality check results.

[0008] In some optional implementations, the rule-based quality inspection results and the large model quality inspection results are cross-validated to obtain cross-validation results, including: If the results of the rule quality inspection are consistent with the results of the large model quality inspection, the results of either the rule quality inspection or the large model quality inspection will be used as the cross-validation results. If the rule quality inspection result indicates a violation, but the large model quality inspection result indicates compliance, a conflict check is performed, and the conflict check result corresponding to the conflict check is used as the cross-validation result. If the rule quality inspection result indicates compliance, but the large model quality inspection result indicates non-compliance, case verification is performed, and the case verification result corresponding to the case verification is used as the cross-verification result.

[0009] In some optional implementations, conflict checking is performed, and the conflict checking result is used as the cross-validation result, including: Construct conflict verification prompts that include the first semantic slice, the target quality inspection rule, and the rule quality inspection results; Input the conflict check prompt into the large model to instruct the large model to perform conflict check based on the conflict check prompt and obtain the conflict check result.

[0010] In some optional implementations, case validation is performed, and the case validation results are used as cross-validation results, including: From the preset case verification rules, determine the target case verification rules corresponding to the target service node; the case verification rules include a case library; Select a target case from the case library corresponding to the target case validation rule, perform case validation on the first semantic slice based on the target case, and use the case validation result generated by the case validation as the cross-validation result.

[0011] In some optional implementations, the customer service quality inspection method also includes: If the cross-validation result is a violation, an evidence chain is generated and archived; the evidence chain includes at least one of the following: violation location evidence, violation original text evidence, rule matching evidence, causal relationship evidence, case matching evidence, and rectification guidance evidence.

[0012] In some optional implementations, the customer service quality inspection method also includes: For any first semantic slice with the role label "customer service", the first semantic slice is processed according to the preset customer service emotion quantification model to determine the customer service emotion value of the first semantic slice. For any second semantic slice labeled as "customer", the second semantic slice is processed according to a preset customer sentiment quantification model to determine the customer sentiment value of the second semantic slice. Based on the customer sentiment value or customer service sentiment value corresponding to each semantic slice, and the time identifier of the semantic slice, a temporally aligned sentiment state graph is constructed. The sentiment state graph includes sentiment nodes and causal relationship edges, which are used to characterize the degree of influence of customer service behavior on subsequent changes in customer sentiment. If the cross-validation result is found to be non-compliant, the impact of the first semantic slice on subsequent changes in customer sentiment is determined based on the sentiment profile.

[0013] The customer service quality inspection method provided in this embodiment converts service call recordings into corresponding voice-text and semantic slices, determines the target service node corresponding to the first semantic slice, and performs rule-based quality inspection on the first semantic slice based on the target service node. This ensures that the rule-based quality inspection is only effective within the target service node, and rule writers only need to design quality inspection conditions for a single node, reducing rule writing costs. Furthermore, using a large-scale model for quality inspection helps understand complex contexts and handle colloquial and grammatically incorrect expressions, improving semantic understanding during the quality inspection process. By combining rule-based quality inspection with large-scale model quality inspection, the results are cross-validated to obtain more accurate service quality inspection results. By automatically generating a multi-dimensional evidence chain when a violation occurs, the credibility and business acceptance of the quality inspection results are improved.

[0014] Secondly, the present invention provides a customer service quality inspection device, comprising: The voice-to-text module is used to acquire service call recordings between customers and customer service representatives and convert the service call recordings into corresponding voice-to-text.

[0015] The speech slicing module is used to segment speech text based on semantic integrity, obtaining at least one semantically complete semantic slice. The semantic slice includes role tags, which include "customer" and "customer service".

[0016] The target service node module is used to determine the target service node corresponding to any first semantic slice with the role label "customer service" based on multiple pre-defined service nodes.

[0017] The rule quality inspection module is used to determine the target quality inspection rules corresponding to the target service node, perform rule quality inspection on the first semantic slice according to the target quality inspection rules, and obtain the rule quality inspection results.

[0018] The large model quality inspection module is used to perform large model quality inspection on the first semantic slice based on the large model, and obtain the large model quality inspection result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node.

[0019] The cross-validation module is used to cross-validate the rule-based quality inspection results and the large model quality inspection results to obtain the cross-validation result. The cross-validation result is used to represent the service quality inspection result corresponding to the first semantic slice.

[0020] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the customer service quality inspection method of the first aspect or any corresponding embodiment described above.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions, which are used to cause a computer to perform the customer service quality inspection method of the first aspect or any corresponding embodiment described above.

[0022] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the customer service quality inspection method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first type of customer service quality inspection method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a second process for a customer service quality inspection method according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a customer service quality inspection device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0027] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] As an optional application scenario of this invention, such as Figure 1 As shown, application 101 is installed in terminal device 110, and user 130 can interact with application 101 through terminal device 110 and / or access device of terminal device 110.

[0029] For example, application 101 can be any application that provides customer service quality inspection. Figure 1 In the application scenario shown, if application 101 is active, the terminal device 110 can display the interface 102 of application 101. The interface 102 may include various pages that application 101 can provide, such as a service voice upload page, a quality inspection page, a settings page, etc.

[0030] In some embodiments, terminal device 110 is communicatively connected to server 120 to provide services to application 101. Terminal device 110 may be a mobile terminal, fixed terminal, or portable terminal, etc., including but not limited to desktop computers, laptop computers, multimedia tablets, or any combination thereof, including accessories and peripherals of these devices, or any combination thereof. In some embodiments, terminal device 110 may also support any type of interface, and server 120 may be various types of computing systems or servers capable of providing computing power, including but not limited to mainframes, edge computing nodes, computing devices in cloud environments, etc.

[0031] It should be noted that, Figure 1 This is merely an example of an application scenario and does not limit the scope of protection of this invention.

[0032] The embodiments of the present invention will now be described with reference to the accompanying drawings. It should be understood that the pages shown in the drawings are merely examples, and various page designs are possible in practice. The various graphic elements on the page may have different arrangements and different visual representations; one or more elements may be omitted or replaced, and one or more other elements may also be present, without any limitation in the embodiments of the present invention. Furthermore, the embodiments described below primarily pertain to terminal device 110. It should be understood that the actions described relative to terminal device 110 can be performed by application 101 on terminal device 110, or can be performed by application 101 in conjunction with its server (e.g., server 120).

[0033] This invention provides a customer service quality inspection method. It converts recorded service calls into corresponding voice-to-text and semantic segments. Rule-based quality inspection is then performed on the first semantic segment based on the target service node. Rule writers only need to design inspection conditions for individual nodes, reducing the difficulty of rule writing. Furthermore, large-scale model-based quality inspection helps understand complex contexts and handle colloquial and grammatically incorrect expressions, improving semantic understanding during the quality inspection process. By combining rule-based quality inspection with large-scale model-based quality inspection, the results are cross-validated to obtain more accurate service quality inspection results.

[0034] According to an embodiment of the present invention, a customer service quality inspection method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0035] This embodiment provides a customer service quality inspection method, which can be used in the aforementioned terminal devices, such as desktop computers and laptops, as well as in the aforementioned servers, such as mainframes and edge computing nodes. Figure 2 This is a flowchart of a customer service quality inspection method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the service call recording between the customer and the customer service representative, and convert the service call recording into the corresponding voice text.

[0036] To ensure the quality of customer service, the service call recordings can be converted into corresponding voice-to-text. In this embodiment, the original service call recording files (e.g., MP3 or WAV format recording files) can be preprocessed, such as by noise reduction and echo cancellation, to obtain the preprocessed service call recordings. Then, Automatic Speech Recognition (ASR) technology is used to convert the preprocessed service call recordings into easily manipulated and processed text strings (i.e., voice-to-text). This voice-to-text string is a text sequence arranged in chronological order, which may also include start and end time markers for each segment of text.

[0037] By converting service call recordings into corresponding voice-to-text, speech that is difficult to analyze directly can be transformed into structured text, which facilitates parallel processing and reduces processing costs.

[0038] Step S202: Segment the speech text based on semantic integrity to obtain at least one semantically complete semantic slice. The semantic slice includes role tags, which include customer and customer service.

[0039] First, initial segments are generated based on the role changes in the dialogue. Within each initial segment, the role label (i.e., customer or customer service) is unique. If two adjacent initial segments belong to the same role (i.e., the role label is the same), and the pause time is short (in this embodiment, a preset pause time threshold can be added; for example, a pause time of less than 2 seconds is considered short, which can be determined by the start and end time identifiers corresponding to the initial semantic segments), and the large model determines that their semantic similarity is high (e.g., a semantic similarity greater than 0.85 is considered high semantic similarity), it means that these two initial segments actually belong to the same semantic topic, and are automatically merged into a more complete semantic segment, thus ensuring semantic integrity. Segmenting speech text based on semantic integrity ensures that each semantic segment can independently express a complete dialogue content in terms of content and logic (e.g., the dialogue content can be a request, an answer, etc.), avoiding the problem of the same semantic content being segmented into different semantic segments due to fixed-time segmentation.

[0040] If an initial slice is too long (e.g., exceeding 60 seconds or 200 words), it indicates that it may contain multiple independent semantic topics. In this case, the larger model is invoked to split the initial semantic slice using splitting prompts.

[0041] In this embodiment, the splitting prompt can be: "You are an after-sales dialogue semantic segmentation expert. Only perform semantic topic segmentation on the input text, strictly adhering to the following rules: 1. Each segmented sub-slice corresponds to a complete semantic expression; segmenting the complete semantics of a single sentence is prohibited; 2. Sub-slices must fit the after-sales service dialogue scenario, distinguishing different requests, questions, and response topics; 3. Only output a list of segmented sub-slices, each labeled with its corresponding topic tag; no other content should be output. Text to be segmented: {S i}”. Among them, {S i} represents the text to be split, i.e., the initial slice that is too long.

[0042] Step S203: For any first semantic slice with the role label "customer service", determine the target service node corresponding to the first semantic slice based on multiple pre-defined service nodes.

[0043] For the first semantic slice with the role label "customer service," the target service node corresponding to this first semantic slice can be determined. In this embodiment, multiple service nodes can be pre-defined, such as: opening procedure node, request confirmation node, policy notification node, problem handling node, request response node, closed-loop confirmation node, closing procedure node, and exception handling node. Each node has a corresponding number (ID), triggering condition, start and end boundaries, and core compliance requirements. For example, for the opening procedure node, its triggering condition could be "the customer service representative speaks for the first time," the start and end boundaries could be "from the customer service representative speaking to the customer's first complete statement of their request," and the core compliance requirement could be "the customer service representative must proactively state their employee ID, brand name, and offer a polite greeting." Based on the triggering conditions and start and end boundaries of multiple service nodes, the first semantic slice and the corresponding service node among the pre-defined multiple service nodes can be determined, i.e., the target service node.

[0044] In this embodiment, the IDs, triggering conditions, start and end boundaries, and core compliance requirements corresponding to multiple service nodes are shown in Table 1 below: Table 1

[0045] In this embodiment, the target service node corresponding to the first semantic slice can be determined using a large model and corresponding node prompts. For example, the node prompts can be: "You are an expert in identifying nodes in the after-sales service process. Based solely on the context of the conversation, you determine the standardized node ID of the current conversation, strictly adhering to the following rules:" 1. The standardized service node IDs are 1-8, and the corresponding content is: {the triggering conditions and start and end boundaries of the 8 service nodes}; 2. Only output the number corresponding to the service node ID; do not output any other content.

[0046] Given the dialogue context: {all current semantic slices + role tags + sequential IDs of semantic slices set in order}, the current semantic slice is: {Sn}.

[0047] Please output the service node ID corresponding to the current semantic slice. Here, {Sn} is the semantic slice (i.e., the first semantic slice) for which the service node needs to be determined.

[0048] By identifying the target service node for the first semantic slice, it is ensured that the quality inspection rules for subsequent quality inspection processes only take effect within the target service node, avoiding cross-node misjudgments. Furthermore, rule writers only need to design quality inspection conditions for a single service node, reducing the difficulty of rule writing.

[0049] Step S204: Determine the target quality inspection rules corresponding to the target service node, and perform rule quality inspection on the first semantic slice according to the target quality inspection rules to obtain the rule quality inspection results.

[0050] After identifying the target service node, a target quality inspection rule can be selected from a set of pre-defined quality inspection rules. This target quality inspection rule can be a regular expression or a structured trigger condition, such as a time condition or a text condition. Based on the target quality inspection rule, a rule-based quality inspection can be performed on the first semantic slice to obtain the corresponding rule inspection result. This result can be either "compliant" or "non-compliant".

[0051] Step S205: Perform a large model quality check on the first semantic slice based on the large model to obtain the large model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node.

[0052] During the rule-based quality inspection process, a large-scale model quality inspection can be performed in parallel. This involves inputting the first semantic slice and the target service requirements corresponding to the target service node into the large model. The large model then determines whether the first semantic slice is compliant, generating the large-scale model quality inspection result. The target service requirements are the service requirements corresponding to the target service node, such as the "core compliance requirements" for the target service node in Table 1 above. These requirements are input into the large model to determine the compliance of the first semantic slice. For example, for a request confirmation node, the service requirement could be "must fully restate the customer's core request, confirm product information, and confirm customer identity information." In this embodiment, the large model can be a Large Language Model (LLM), and the large-scale model quality inspection of the first semantic slice is performed based on the semantic understanding capabilities pre-trained in the large model.

[0053] By using large-scale model quality inspection, we can understand the contextual relationships of speech content. When faced with complex contexts, polysemous expressions, or colloquial expressions, we can accurately judge the call quality and make the quality inspection results more accurate.

[0054] Step S206: Cross-validate the rule quality inspection results and the large model quality inspection results to obtain the cross-validation results. The cross-validation results are used to represent the service quality inspection results corresponding to the first semantic slice.

[0055] Because rule-based quality inspection results are obtained through rigid rules (such as regular expressions or triggering conditions), they may lead to misjudgments due to the inflexibility of the rules. Meanwhile, large-scale model quality inspection results are based on the semantic understanding capabilities of a large model, and may also be subject to misjudgments due to factors such as the training data of the large model. In this step, cross-validation can be performed on the rule-based and large-scale model quality inspection results. For example, cross-comparing and comprehensively judging the rule-based and large-scale model quality inspection results yields a more accurate cross-validation result, which can represent the service quality inspection result corresponding to the first semantic slice. Furthermore, based on the cross-validation results of each first semantic slice, the service quality inspection result of the service call recording can be comprehensively determined.

[0056] By cross-validating the results of rule-based quality inspection and the results of large-scale model quality inspection, quality inspection can be performed using rigid rules to correct omissions in large-scale model quality inspection, and the quality inspection of large-scale model can correct misjudgments of rigid rules in special contexts, thereby obtaining more accurate service quality inspection results.

[0057] The customer service quality inspection method provided in this embodiment converts service call recordings into corresponding voice-text and semantic slices, determines the target service node corresponding to the first semantic slice, and performs rule-based quality inspection on the first semantic slice based on the target service node. This ensures that the rule-based quality inspection is only effective within the target service node, and rule writers only need to design quality inspection conditions for a single node, reducing the difficulty of rule writing. Furthermore, using a large-scale model for quality inspection helps understand complex contexts and handle colloquial and grammatically incorrect expressions, improving semantic understanding during the quality inspection process. By combining rule-based quality inspection with large-scale model quality inspection, the results of the rule-based and large-scale model quality inspections are cross-validated to obtain more accurate service quality inspection results.

[0058] This embodiment provides a customer service quality inspection method, which can be used in the aforementioned terminal devices, such as desktop computers and laptops, as well as in the aforementioned servers, such as mainframes and edge computing nodes. Figure 3 This is a flowchart of a customer service quality inspection method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain the service call recording between the customer and the customer service representative, and convert the service call recording into the corresponding voice-to-text.

[0059] Please see details Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0060] Step S302: Segment the speech text based on semantic integrity to obtain at least one semantically complete semantic slice. The semantic slice includes role tags, which include customer and customer service.

[0061] Please see details Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0062] Step S303: For any first semantic slice with the role label "customer service", determine the target service node corresponding to the first semantic slice based on multiple pre-defined service nodes.

[0063] Please see details Figure 2 Step S203 of the illustrated embodiment will not be described again here.

[0064] Step S304: Determine the target quality inspection rules corresponding to the target service node, perform rule quality inspection on the first semantic slice according to the target quality inspection rules, and obtain the rule quality inspection results.

[0065] Please see details Figure 2 Step S204 of the illustrated embodiment will not be described again here.

[0066] In some optional implementations, step S304, "determine the target quality inspection rule corresponding to the target service node, perform rule quality inspection on the first semantic slice according to the target quality inspection rule, and obtain the rule quality inspection result", includes steps a1 to a4.

[0067] Step a1: For any quality inspection rule used to determine whether a violation has occurred, determine the set of effective nodes for the quality inspection rule, and take the quality inspection rule that includes the target service node in the set of effective nodes as the target quality inspection rule.

[0068] Step a2: Based on the slice information of the first semantic slice, determine whether the first semantic slice meets the triggering conditions of the target quality inspection rule. The slice information includes the time information, text information, and node identifier of the target service node of the first semantic slice. The triggering conditions include at least one of the time conditions, text conditions, and node conditions.

[0069] Step a3: If the first semantic slice satisfies the triggering condition of the target quality inspection rule, determine the rule quality inspection result indicating the violation.

[0070] Step a4: If the first semantic slice does not meet the triggering conditions of the target quality inspection rule, determine the rule quality inspection result that indicates compliance.

[0071] Since a large number of quality inspection rules are pre-defined, the target quality inspection rule can be selected from these rules first. In this embodiment, the effective node set of each quality inspection rule can be determined first. The effective node set of a quality inspection rule indicates which service nodes(s) the quality inspection rule applies to. Furthermore, based on the target service node and the effective node sets of each quality inspection rule, the corresponding target quality inspection rule is determined. For example, the quality inspection rule that includes the target service node in all effective node sets can be used as the target quality inspection rule.

[0072] Each target quality inspection rule, in addition to binding to the effective node, also includes a triggering condition. When this triggering condition is triggered, the first semantic slice can be determined to be in violation. In this embodiment, the triggering condition may include at least one of the following: time condition (e.g., the time condition may be "within 30 seconds before the call"), node condition (e.g., the node condition may be "the node identifier is 1"), and text condition (e.g., the text condition may be "the text does not contain the expression of employee number + number"). Correspondingly, the slice information of the first semantic slice also includes corresponding time information (i.e., the start and end time of the first semantic slice), text information (i.e., the text of the first semantic slice), and the node identifier of the target service node.

[0073] In this embodiment, the structure of the quality inspection rule can be: r i = (Nid_set, C i A i , T i W i ); Where, r i For the i-th quality inspection rule, Nid_set is the set of node IDs (i.e., the set of effective nodes) where the quality inspection rule takes effect, and C i The corresponding trigger condition can be a structured logical expression, supporting logical operations, time conditions, node conditions, and text conditions. For example, the trigger condition could be "(within 30 seconds before the call and node Nid=1) and the text does not contain an employee ID + number expression." i This refers to the actions taken after a quality inspection rule is triggered, including marking violations, triggering alerts, pushing standardized scripts, generating rectification work orders, and synchronizing with the performance appraisal system. i Standardize the coding for violation types. For example, the standardized coding for a violation type could be FLOW-VIOLATE-001 "Process Violation - Failure to Report Employee Number at the Start". i This is the rule weight, representing the severity of the violation, used for priority ranking when multiple rules are matched.

[0074] If the first semantic slice meets the triggering condition of a target quality inspection rule, it is determined that the first semantic slice violates the target quality inspection rule, and the rule quality inspection result is set to "violation"; if it is determined that the first semantic slice does not meet the triggering conditions of all target quality inspection rules, the system determines that the first semantic slice is compliant under the rule quality inspection, and the rule quality inspection result is set to "compliant".

[0075] By filtering by target service nodes, the matching scope can be narrowed from all quality inspection rules to target quality inspection rules only related to the current node, thereby improving quality inspection efficiency.

[0076] Step S305: Perform large model quality inspection on the first semantic slice based on the large model to obtain the large model quality inspection result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node.

[0077] In some optional implementations, step S305, "performing a large model quality check on the first semantic slice based on the large model to obtain a large model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node," includes steps b1 and b2.

[0078] Step b1: Obtain the target service requirements of the target service node and construct a large model prompt word containing the first semantic slice and the target service requirements.

[0079] Step b2: Input the large model prompt words into the large model to instruct the large model to perform a large model quality check on the first semantic slice based on the large model prompt words, and obtain the large model quality check results.

[0080] After determining the target service requirements corresponding to the first semantic slice, corresponding large model prompt words can be generated based on the first semantic slice and the target service requirements. In this embodiment, a preceding semantic slice corresponding to the first semantic slice can also be added to the large model prompt words to facilitate the large model's understanding of the context. In this embodiment, the large model prompt word template can be: "You are a dedicated quality inspection expert for after-sales service scenarios. You are responsible for verifying the compliance of dialogue texts based solely on the target service requirements of the current stage, strictly adhering to the following rules:" 1. Current service node ID: {Nid}, service node compliance requirements: {target service requirements for the service node}; 2. Validation should only be performed on the dialogue text corresponding to the current service node, taking into account the context; validation across service nodes is not allowed. 3. The output must include: compliance / violation determination result; violation type (if a violation occurs); evidence of violation (excerpt from the original text, accurate to the sentence level); confidence level Sim (0-1, 1 = complete violation, 0 = complete compliance); and the basis for the determination.

[0081] The dialogue text to be validated includes: {first semantic slice, corresponding dialogue context, role label}. After generating the large model prompts, the prompts can be sent to the large model to perform a quality check on the first semantic slice, and the quality check result can be obtained. For example, the quality check result can be "compliant" or "non-compliant".

[0082] Step S306: Perform cross-validation on the rule quality inspection results and the large model quality inspection results to obtain the cross-validation result. The cross-validation result is used to represent the service quality inspection result corresponding to the first semantic slice.

[0083] Specifically, step S306, "cross-validating the rule quality inspection results and the large model quality inspection results to obtain cross-validation results", includes steps S3061 to S3063.

[0084] Step S3061: If the rule quality inspection result is consistent with the large model quality inspection result, the rule quality inspection result or the large model quality inspection result shall be used as the cross-validation result.

[0085] When both the rule-based quality inspection result and the large-model quality inspection result are deemed "compliant," or both are deemed "non-compliant" with the same type of violation, it can be determined that the two quality inspection methods (i.e., rule-based quality inspection and large-model quality inspection) have reached a consistent conclusion. In this case, the reliability of the quality inspection result for the first semantic slice is relatively high, and either the rule-based quality inspection result or the large-model quality inspection result can be directly used as the cross-validation result (in this case, the two quality inspection results are consistent). If the cross-validation result is a violation, the execution action A corresponding to the rule-based quality inspection result can be executed. i Record the standardized code T for this violation type. i wait.

[0086] Step S3062: If the rule quality inspection result indicates a violation, but the large model quality inspection result indicates compliance, perform conflict verification and use the conflict verification result as the cross-validation result.

[0087] When the rule quality inspection result indicates a violation, but the large model quality inspection result indicates compliance, conflict verification can be performed. Conflict verification is used to perform secondary verification of the rule quality inspection result to determine whether the rule quality inspection result is correct. In this embodiment, conflict verification can be performed through the large model or through manual review. After completing the conflict verification, the conflict verification result can be used as the cross-validation result.

[0088] In some optional implementations, step S3062, "perform conflict checking and use the conflict checking result corresponding to the conflict checking as the cross-check result", includes steps c1 and c2.

[0089] Step c1: Construct conflict verification prompt words that include the first semantic slice, the target quality inspection rule, and the rule quality inspection result.

[0090] Step c2: Input the conflict check prompt into the large model to instruct the large model to perform conflict check according to the conflict check prompt and obtain the conflict check result.

[0091] To reduce manpower costs during conflict verification, a large model can be used to perform the corresponding conflict verification. In this case, conflict verification prompts can be constructed to be input into the large model. These prompts can include the first semantic slice, the target quality control rule, and the rule quality control result.

[0092] In this embodiment, the conflict check prompt can be: "Given the definition of the target quality inspection rule: {r i The complete structured definition}, the current dialogue context: {dialogue text of the first semantic slice}. The rule quality check determines that the target quality check rule has been hit, and the violation type is {T}. i The large model determines it to be compliant. Please combine the target quality inspection rules and the first semantic slice to determine whether it is a rule-based quality inspection misjudgment. Output only "Yes / No," along with the complete basis for the judgment; do not output any other content. Where, r i The target quality inspection rule to be triggered.

[0093] After generating conflict check hints, these hints can be input into the large model. The large model can then perform corresponding conflict checks based on these hints and generate the corresponding conflict check results. If the conflict check result indicates a false positive, the rule quality inspection result can be rejected, and "compliant" can be output as the conflict check result. If the conflict check result indicates a valid decision, the target quality inspection rule r can then be executed. i Action A i It outputs "violation" as the conflict verification result and adds the judgment criteria of the large model to the evidence chain.

[0094] Step S3063: If the rule quality inspection result indicates compliance and the large model quality inspection result indicates violation, perform case verification and use the case verification result corresponding to the case verification as the cross-verification result.

[0095] When the rule quality check result indicates compliance, but the large model quality check result indicates non-compliance, case validation can be performed. Case validation can use existing cases as a reference to determine the correctness of the large model quality check result. For example, it can be done through manual review and comparison with similar cases, or it can be done by using the large model to perform case validation based on similar cases. After completing the case validation, the corresponding case validation results can be used as cross-validation results.

[0096] In some optional implementations, step S3063, "perform case verification and use the case verification result corresponding to the case verification as the cross-verification result", includes steps d1 and d2.

[0097] Step d1: Determine the target case verification rule corresponding to the target service node from the preset case verification rules. The case verification rules include a case library.

[0098] Step d2: Select a target case from the case library corresponding to the target case verification rule, perform case verification on the first semantic slice based on the target case, and use the case verification result generated by the case verification as the cross-verification result.

[0099] Similarly, to reduce labor costs, case validation can be performed using a large model. Target case validation rules corresponding to the target service node can be determined from preset case validation rules for case validation. In this embodiment, the structure of the case validation rules can be: s i = (Nid_set, Rule_desc, Case_lib, θ, T i W i ); Among them, s i Unlike the aforementioned quality inspection rules, the case verification rules only have a case library. A large model compares corresponding or similar cases within the case verification rules to generate verification results, achieving flexible verification. `Rule_desc` is the linguistic description of the case verification rule, such as "The use of evasive language is prohibited; responsibility must not be shifted to customers, manufacturers, or other departments; service responsibility must be proactively assumed, demonstrating empathetic response." `Case_lib` is the case library corresponding to the case verification rule. This library can be divided into positive example libraries, negative example libraries, and boundary case libraries. Each case in the library can include dialogue text, annotation results, node labels, confidence level, and addition time. Cases in the positive example library are compliant cases, i.e., cases judged as "compliant" by customer service quality inspection; cases in the negative example library are non-compliant cases; and cases in the boundary case library are boundary cases prone to misjudgment. `θ` is the confidence threshold corresponding to the case verification rule.

[0100] Based on the target service node ID set (i.e., Nid_set), the target case verification rule corresponding to the target service node can be determined in the case verification rule set, and the target case can be determined in the case library corresponding to the target case verification rule. In this embodiment, 3 cases can be selected from the positive example library, 5 cases from the negative example library, and 2 cases from the boundary case library as target cases. After determining the target cases, corresponding case verification prompts can be generated. Inputting the case verification prompts into the large model will determine the case verification result output by the large model.

[0101] In some optional implementations, inputting case validation prompts into the large model determines the case validation results output by the large model, including: Obtain the decision score Sim from the large model output. If Sim < θ - 0.2, use compliance as the case validation result; if Sim > θ + 0.2, use violation as the case validation result; if Sim ∈ [θ - 0.2, θ + 0.2], trigger secondary validation, call the boundary case library to generate exclusive secondary validation prompt words, let the large model re-determine, and output a more detailed chain of evidence. After secondary validation, if Sim ≥ θ, use violation as the case validation result; if Sim < θ, use compliance as the case validation result, and at the same time, add this case as a boundary case to the boundary case library.

[0102] In some optional implementations, the customer service quality inspection method further includes step e1.

[0103] Step e1: If the cross-validation result indicates a violation, generate and archive the evidence chain. The evidence chain includes at least one of the following: violation location evidence, original violation text evidence, rule matching evidence, causal relationship evidence, case matching evidence, and rectification guidance evidence.

[0104] When cross-validation results indicate a violation, a corresponding evidence chain can be generated. This evidence chain can include violation location evidence, original violation text evidence, rule matching evidence, causal relationship evidence, case matching evidence, and rectification guidance evidence. Violation location evidence includes the service node ID where the violation occurred, the ID corresponding to the semantic slice, the role tag, and the timestamp of the occurrence, used to locate the violation. Original violation text evidence is the original text of the violation in the dialogue, accurate to specific sentences or phrases, ensuring the clarity of the violation facts. Rule matching evidence includes the definition, effective node, violation type, and weight of the quality inspection rule triggered by this violation. Causal relationship evidence proves the actual impact of the violation. Case matching evidence consists of several negative examples most similar to the violation text, along with their corresponding semantic similarity scores. Rectification guidance evidence is generated by the large model based on the violation type and dialogue context, providing precise rectification suggestions, standardized wording references, and directions for subsequent service optimization. By automatically generating a multi-dimensional evidence chain when a violation occurs, the transparency and business value of the quality inspection system are improved.

[0105] In some optional implementations, the customer service quality inspection method further includes steps f1 to f4.

[0106] Step f1: For any first semantic slice with the role label "customer service", process the first semantic slice according to the preset customer service emotion quantification model to determine the customer service emotion value of the first semantic slice.

[0107] Step f2: For any second semantic slice labeled as "customer", process the second semantic slice according to the preset customer sentiment quantification model to determine the customer sentiment value of the second semantic slice.

[0108] Step f3 involves constructing a time-aligned emotional state graph based on the customer or customer service emotional values ​​corresponding to each semantic slice and the time identifier of the semantic slice. The emotional state graph includes emotional nodes and causal relationship edges, which represent the degree of influence of customer service behavior on subsequent changes in customer emotions.

[0109] Step f4: If the cross-validation result is a violation, determine the degree of influence of the first semantic slice on subsequent changes in customer sentiment based on the sentiment profile.

[0110] For the first semantic slice generated by the customer service role, a customer service emotion quantification model specifically designed for customer service is invoked to calculate a customer service emotion value (for example, the customer service emotion value can be 0 to 10), which represents the customer service's emotional state or service attitude tendency in the first semantic slice.

[0111] For the second semantic slice generated for the customer role, a customer emotion quantification model specifically designed for customers is invoked to calculate a customer emotion value (for example, the customer service emotion value can be 0 to 10), which represents the customer's emotional state in the second semantic slice.

[0112] After obtaining the customer service sentiment value or customer sentiment value corresponding to each semantic slice (including the first semantic slice and the second semantic slice), a sentiment state graph can be established. The temporal alignment of the sentiment state graph is ensured based on the time identifier corresponding to each semantic slice. The sentiment state graph G_emo can be represented as G_emo=(V_emo, E_emo), where V_emo is the sentiment node corresponding to each semantic slice in the graph, which can include role label, time sequence ID, process node ID, sentiment value (i.e., customer service sentiment value or customer sentiment value), and the text corresponding to the semantic slice; E_emo is the temporal association edge connecting the sentiment nodes corresponding to adjacent semantic slices of the same role in the graph, or the causal association edge connecting the first semantic slice node with the second semantic slice node in the next 1-2 rounds, indicating the extent to which the first semantic slice of the customer service led to the subsequent change in customer sentiment.

[0113] If the cross-validation result indicates a violation, the impact of the first semantic slice on subsequent changes in customer sentiment is quantitatively assessed based on the sentiment profile. This impact can be used to determine the influence of the customer service representative's language on the user. If the impact is significant, the corresponding first semantic slice can be added to the negative example library or to the quality inspection rules, allowing for rule-based quality checks on similar language and wording during subsequent quality inspections. Furthermore, this impact can also be added to the evidence chain as causal evidence to prove the actual impact of the violation.

[0114] The customer service quality inspection method provided in this embodiment converts service call recordings into corresponding voice-text and semantic slices, determines the target service node corresponding to the first semantic slice, and performs rule-based quality inspection on the first semantic slice based on the target service node. This ensures that the rule-based quality inspection is only effective within the target service node, and rule writers only need to design quality inspection conditions for a single node, reducing rule writing costs. Furthermore, using a large-scale model for quality inspection helps understand complex contexts and handle colloquial and grammatically incorrect expressions, improving semantic understanding during the quality inspection process. By combining rule-based quality inspection with large-scale model quality inspection, the results are cross-validated to obtain more accurate service quality inspection results. By automatically generating a multi-dimensional evidence chain when a violation occurs, the credibility and business acceptance of the quality inspection results are improved.

[0115] This embodiment also provides a customer service quality inspection device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0116] This embodiment provides a customer service quality inspection device, such as... Figure 4 As shown, it includes: The voice-to-text module 401 is used to acquire service call recordings between customers and customer service representatives and convert the service call recordings into corresponding voice-to-text.

[0117] The speech slicing module 402 is used to segment speech text based on semantic integrity to obtain at least one semantically complete semantic slice. The semantic slice includes role tags, which include customer and customer service.

[0118] The target service node module 403 is used to determine the target service node corresponding to any first semantic slice with the role label "customer service" based on multiple pre-defined service nodes.

[0119] The rule quality inspection module 404 is used to determine the target quality inspection rules corresponding to the target service node, perform rule quality inspection on the first semantic slice according to the target quality inspection rules, and obtain the rule quality inspection results.

[0120] The large model quality inspection module 405 is used to perform large model quality inspection on the first semantic slice based on the large model, and obtain the large model quality inspection result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node.

[0121] The cross-validation module 406 is used to cross-validate the rule quality inspection results and the large model quality inspection results to obtain the cross-validation result. The cross-validation result is used to represent the service quality inspection result corresponding to the first semantic slice.

[0122] In some alternative implementations, the rule-based quality inspection module 404 includes: The target quality inspection rule submodule is used to determine the set of effective nodes for any quality inspection rule used to determine whether a violation has occurred, and to take the quality inspection rule that includes the target service node in the set of effective nodes as the target quality inspection rule.

[0123] The triggering condition submodule is used to determine whether the first semantic slice meets the triggering conditions of the target quality inspection rule based on the slice information of the first semantic slice. The slice information includes the time information, text information, and node identifier of the target service node of the first semantic slice, and the triggering conditions include at least one of the time conditions, text conditions, and node conditions.

[0124] The rule violation submodule is used to determine the rule quality inspection result indicating a violation when the first semantic slice meets the triggering condition of the target quality inspection rule.

[0125] The rule compliance submodule is used to determine the rule quality inspection result that indicates compliance when the first semantic slice does not meet the triggering conditions of the target quality inspection rule.

[0126] In some alternative implementations, the large model quality inspection module 405 includes: The prompt word submodule is used to obtain the target service requirements of the target service node and construct a large model prompt word containing the first semantic slice and the target service requirements.

[0127] The large model quality inspection submodule is used to input large model prompt words into the large model, so as to instruct the large model to perform large model quality inspection on the first semantic slice according to the large model prompt words, and obtain the large model quality inspection results.

[0128] In some alternative implementations, the cross-validation module 406 includes: The consistency of results submodule is used to use either the rule quality inspection result or the large model quality inspection result as the cross-validation result when the rule quality inspection result is consistent with the large model quality inspection result.

[0129] The conflict verification submodule is used to perform conflict verification when the rule quality inspection result indicates a violation, while the large model quality inspection result indicates compliance. The conflict verification result is then used as the cross-validation result.

[0130] The case validation submodule is used to perform case validation when the rule quality inspection result indicates compliance and the large model quality inspection result indicates violation, and the case validation result corresponding to the case validation is used as the cross-validation result.

[0131] In some optional implementations, the conflict checking submodule includes: The conflict prompt word unit is used to construct conflict verification prompt words that include the first semantic slice, the target quality inspection rule, and the rule quality inspection result.

[0132] The conflict checking unit is used to input conflict checking prompts into the large model, so as to instruct the large model to perform conflict checking according to the conflict checking prompts and obtain the conflict checking results.

[0133] In some optional implementations, the case verification submodule includes: The case rule unit is used to determine the target case verification rule corresponding to the target service node from the preset case verification rules. The case verification rules include a case library.

[0134] The case verification unit is used to select a target case from the case library corresponding to the target case verification rule, perform case verification on the first semantic slice based on the target case, and use the case verification result generated by the case verification as the cross-verification result.

[0135] In some optional implementations, the customer service quality inspection device further includes: The evidence chain module is used to generate and archive an evidence chain when the cross-validation result indicates a violation. The evidence chain includes at least one of the following: violation location evidence, original violation text evidence, rule matching evidence, causal relationship evidence, case matching evidence, and rectification guidance evidence.

[0136] In some optional implementations, the customer service quality inspection device further includes: The customer service emotion module is used to process the first semantic slice for any role labeled as customer service according to a preset customer service emotion quantification model, and determine the customer service emotion value of the first semantic slice.

[0137] The customer sentiment module is used to process any second semantic slice labeled as "customer" according to a preset customer sentiment quantification model to determine the customer sentiment value of the second semantic slice.

[0138] The sentiment graph module is used to construct a time-aligned sentiment landscape graph based on the customer or customer service sentiment values ​​corresponding to each semantic slice and the time identifier of the semantic slice. The sentiment landscape graph includes sentiment nodes and causal relationship edges, which are used to represent the degree of influence of customer service behavior on subsequent changes in customer sentiment.

[0139] The Emotion Influence Module is used to determine the degree of influence of the first semantic slice on subsequent customer emotion changes based on the emotion state graph when the cross-validation result is a violation.

[0140] The customer service quality inspection device provided in this embodiment of the invention can execute the customer service quality inspection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0141] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0142] The following is a detailed reference. Figure 5 The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from memory 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0143] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0144] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the customer service quality inspection method of the embodiments of the present invention.

[0145] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0146] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the customer service quality inspection method shown in the above embodiments is implemented.

[0147] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0148] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A customer service quality inspection method, characterized in that, The method includes: Acquire service call recordings between customers and customer service representatives, and convert the service call recordings into corresponding voice-to-text. The speech text is segmented based on semantic integrity to obtain at least one semantically complete semantic slice; the semantic slice includes role tags, and the role tags include customer and customer service. For any of the first semantic slices with the role tag "customer service", the target service node corresponding to the first semantic slice is determined based on multiple pre-defined service nodes; Determine the target quality inspection rule corresponding to the target service node, and perform rule quality inspection on the first semantic slice according to the target quality inspection rule to obtain the rule quality inspection result; Based on the large model, a large model quality check is performed on the first semantic slice to obtain a large model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node. The rule quality inspection results and the large model quality inspection results are cross-validated to obtain cross-validation results; the cross-validation results are used to represent the service quality inspection results corresponding to the first semantic slice.

2. The method of claim 1, wherein, The step of determining the target quality inspection rule corresponding to the target service node, and performing rule quality inspection on the first semantic slice according to the target quality inspection rule to obtain the rule quality inspection result includes: For any quality inspection rule used to determine whether a violation has occurred, determine the set of effective nodes for the quality inspection rule, and take the quality inspection rule that includes the target service node in the set of effective nodes as the target quality inspection rule; Based on the slice information of the first semantic slice, determine whether the first semantic slice meets the triggering condition of the target quality inspection rule; the slice information includes the time information, text information and node identifier of the target service node of the first semantic slice, and the triggering condition includes at least one of time condition, text condition and node condition; If the first semantic slice satisfies the triggering condition of the target quality inspection rule, the rule quality inspection result indicating a violation is determined. If the first semantic slice does not meet the triggering condition of the target quality inspection rule, a rule quality inspection result indicating compliance is determined.

3. The method of claim 1, wherein, The step of performing a large-scale model quality check on the first semantic slice based on the large model to obtain a large-scale model quality check result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node includes: Obtain the target service requirements of the target service node, and construct a large model prompt word that includes the first semantic slice and the target service requirements; The large model prompt words are input into the large model to instruct the large model to perform a large model quality check on the first semantic slice based on the large model prompt words, and obtain the large model quality check result.

4. The method of claim 1, wherein, The cross-validation of the rule quality inspection results and the large model quality inspection results to obtain the cross-validation results includes: If the rule quality inspection result is consistent with the large model quality inspection result, the rule quality inspection result or the large model quality inspection result shall be used as the cross-validation result. If the rule quality inspection result indicates a violation and the large model quality inspection result indicates compliance, a conflict check is performed, and the conflict check result corresponding to the conflict check is used as the cross-validation result. If the rule quality inspection result indicates compliance, but the large model quality inspection result indicates violation, case verification is performed, and the case verification result corresponding to the case verification is used as the cross-verification result.

5. The method of claim 4, wherein, The step of performing conflict verification, and using the conflict verification result as the cross-validation result, includes: Construct conflict verification prompt words that include the first semantic slice, the target quality inspection rule, and the rule quality inspection result; The conflict check prompt is input into the large model to instruct the large model to perform conflict check according to the conflict check prompt and obtain the conflict check result.

6. The method of claim 4, wherein, The step of performing case validation and using the case validation results as the cross-validation results includes: From the preset case verification rules, the target case verification rules corresponding to the target service node are determined; the case verification rules include a case library; Select a target case from the case library corresponding to the target case verification rule, perform case verification on the first semantic slice based on the target case, and use the case verification result generated by the case verification as the cross-verification result.

7. The method of claim 1, wherein, The method further includes: If the cross-validation result indicates a violation, an evidence chain is generated and archived; the evidence chain includes at least one of the following: violation location evidence, violation original text evidence, rule matching evidence, causal relationship evidence, case matching evidence, and rectification guidance evidence.

8. The method of claim 1, wherein, The method further includes: For any of the aforementioned role tags as customer service, the first semantic slice is processed according to a preset customer service emotion quantification model to determine the customer service emotion value of the first semantic slice. For any of the aforementioned role labels as "customer" in the second semantic slice, the second semantic slice is processed according to a preset customer sentiment quantification model to determine the customer sentiment value of the second semantic slice. Based on the customer sentiment value or customer service sentiment value corresponding to each semantic slice, and the time identifier of the semantic slice, a time-aligned sentiment state graph is constructed; the sentiment state graph includes sentiment nodes and causal relationship edges, and the causal relationship edges are used to characterize the degree of influence of customer service behavior on subsequent changes in customer sentiment. If the cross-validation result is found to be in violation, the degree of influence of the first semantic slice on subsequent changes in customer sentiment is determined based on the sentiment profile.

9. A customer service quality inspection apparatus characterized by comprising: The device includes: The voice-to-text module is used to acquire service call recordings between customers and customer service representatives, and convert the service call recordings into corresponding voice-to-text. The speech slicing module is used to segment the speech text based on semantic integrity to obtain at least one semantically complete semantic slice; the semantic slice includes role tags, and the role tags include customer and customer service. The target service node module is used to determine the target service node corresponding to any first semantic slice with the role tag "customer service" based on a plurality of pre-defined service nodes. The rule quality inspection module is used to determine the target quality inspection rule corresponding to the target service node, and to perform rule quality inspection on the first semantic slice according to the target quality inspection rule to obtain the rule quality inspection result; The large model quality inspection module is used to perform large model quality inspection on the first semantic slice based on the large model, and obtain a large model quality inspection result indicating whether the first semantic slice meets the target service requirements corresponding to the target service node. The cross-validation module is used to cross-validate the rule quality inspection results and the large model quality inspection results to obtain cross-validation results; the cross-validation results are used to represent the service quality inspection results corresponding to the first semantic slice.

10. An electronic device, comprising: include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the customer service quality inspection method according to any one of claims 1 to 8 by executing the computer instructions.

Images