Research data consistency auditing system and method based on customer interaction service

Abstract

The invention relates to the technical field of voice interaction investigation data processing, and discloses an investigation data consistency auditing system and method based on customer interaction service, at least one piece of auditing item data is obtained, and the auditing item data comprises basic data, visitors and a plurality of voice samples interacted with corresponding investigated objects; analyzing each voice sample to form corresponding interaction data; consistency auditing is carried out on the process of analyzing the voice sample to form interaction data, wherein the consistency auditing comprises basic record consistency auditing and voice emotion consistency auditing; creating a task library, matching and / or updating auditing tasks in the task library according to the basic data of the auditing items, obtaining a plurality of target auditing tasks, and performing basic record consistency auditing; constructing an emotion identification model, respectively identifying emotions of the visitor and the investigated object in the voice sample, and performing voice emotion consistency auditing; performing exception marking and secondary re-checking on the voice segments with exceptions in the voice sample consistency checking; and the voice interaction investigation and examination efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of voice interaction research data processing, and specifically to a research data consistency audit system and method based on customer interaction services. Background Art

[0002] Market research, as an important tool for business decision-making, focuses on collecting, analyzing, and interpreting data through systematic methods to help enterprises better understand the market environment, consumer behavior, and competitive landscape. Effective market research can not only reveal potential opportunities and threats in the market but also provide a basis for enterprises to formulate or adjust marketing strategies, thereby optimizing products and services and enhancing competitiveness.

[0003] When conducting market research, researchers can choose various ways to obtain information, including but not limited to questionnaires, telephone interviews, face-to-face interviews, online surveys, etc. Each method has its own characteristics and applicable scenarios. For example, in the service industry, telephone interviews based on questionnaires are widely used because they can directly reach the target audience, answer questions immediately, and ensure a high response rate, making the collected research data highly valuable for application.

[0004] The in-depth exploration of the application value of research data is inseparable from the audit and analysis of data. However, with the expansion of enterprise scale and the increase in service projects, traditional audit methods are facing challenges. The traditional method of manually listening to voice samples obtained through the above-mentioned research methods is not only time-consuming and laborious but also easily affected by subjective factors, such as the professional quality and concentration of auditors, which will all affect the final quality assessment results. In addition, when faced with a large number of voice samples, full audit becomes impractical, while sampling audit cannot guarantee coverage of all key issues and may miss important feedback, thus having a negative impact on the accuracy and reliability of research results. Summary of the Invention

[0005] The present invention aims to provide a research data consistency audit system and method based on customer interaction services, adopting a dual audit method combining intelligent audit and secondary review to reduce the probability of misaudit and missed audit, solve the limitations of traditional manual listening audit methods, greatly improve the audit efficiency, and reduce labor costs; at the same time, based on the data characteristics of the audit project itself, the created task library, and the emotion recognition model, it can not only achieve data consistency audit but also dynamically adjust audit strategies to improve the adaptability and timeliness of audit work.

[0006] The basic solution provided by the present invention is: a research data consistency audit method based on customer interaction services, the method comprising:

[0007] S100. Obtain at least one audit project data, where the audit project data includes basic data and a number of voice samples obtained from the interaction between a number of visitors and corresponding respondents;

[0008] S200. Analyze each voice sample to form corresponding interaction data, where the interaction data includes multiple questions and their answers;

[0009] S210. Conduct consistency audits on the process of analyzing voice samples to form interaction data, including basic record consistency audits and voice emotion consistency audits;

[0010] S211. Create a task library to represent the mapping relationship between a number of audit tasks and audit projects; according to the basic data of the audit project, match and / or update audit tasks in the task library to obtain multiple target audit tasks; conduct basic record consistency audits on voice samples and corresponding interaction data based on the multiple target audit tasks;

[0011] S212. Build an emotion recognition model to respectively recognize the emotions of visitors and respondents in voice samples, and conduct voice emotion consistency audits on voice samples and corresponding interaction data;

[0012] S300. Mark abnormal voice segments in the voice samples during the consistency audit, and send the abnormal voice segments to a number of review terminals according to a preset distribution method. The review terminals conduct secondary reviews and corresponding review processing based on review criteria.

[0013] The research data consistency audit method based on customer interaction services of the present invention also provides a research data consistency audit system based on customer interaction services. The system includes a voice sample acquisition module, an analysis module, a consistency audit module, a secondary review module, and review terminals;

[0014] The voice sample acquisition module is used to obtain the basic data of at least one audit project and a number of voice samples obtained from the interaction between a number of visitors and corresponding respondents;

[0015] The analysis module is used to analyze voice samples to form interaction data, where the interaction data includes multiple questions and their answers;

[0016] The consistency audit module is used to conduct consistency audits on the process of analyzing voice samples to form interaction data, including basic record consistency audits and voice emotion consistency audits;

[0017] Among them, the consistency review module includes a task library module for creating a task library to represent the mapping relationship between several review tasks and review items; it also includes a basic record consistency review sub-module that, based on the basic data of the review items, matches and / or updates the review tasks in the task library to obtain multiple target review tasks, and conducts a basic record consistency review of the voice samples and the corresponding interaction data based on the multiple target review tasks; it further includes a voice emotion consistency review sub-module for constructing an emotion recognition model to respectively recognize the emotions of the visitor and the surveyed object in the voice sample, and conduct a voice emotion consistency review of the voice sample and the corresponding interaction data.

[0018] The secondary review module is used to mark the abnormal voice segments in the voice samples with consistency review anomalies, and send the voice segments with anomalies to several review terminals according to a preset distribution method.

[0019] The review terminal is used to conduct a secondary review and corresponding review processing based on the review criteria.

[0020] The working principle and advantages of the present invention are as follows:

[0021] Compared with the prior art, this solution introduces advanced information technology means, adopts a combination of intelligent review and secondary review, solves the limitations of the traditional manual listening review method, greatly improves the review efficiency, and reduces the labor cost. Due to the introduction of intelligent review, it can increase the total amount of research samples to a certain extent, reach more research objects, improve the research promotion efficiency, and at the same time recover more questionnaire samples to support the data analysis of the research, further enhancing the effect of the research and providing more solid data support for the strategic planning of the enterprise. In order to ensure the accuracy of intelligent review and balance the review quantity of secondary review, a consistency review of the intelligent review process and dynamic adjustment of the review strategy are proposed. Based on the data characteristics of the review items themselves and the constructed task library, by using the matching and update of review items and review tasks, the review strategy is adjusted to adapt to the changing review requirements, improving the adaptability and timeliness of the review work. Based on the emotion recognition model, it can more deeply understand the true intentions, attitudes and other emotion information of the surveyed object and the visitor by targeting the non-verbal information such as the tone, emotion and intonation retained in the voice sample, improving the accuracy of the review. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flowchart of the research data consistency review method based on customer interaction services provided by Embodiment 1 of the present invention;

[0023] Figure 2 It is a structural diagram of the research data consistency review system based on customer interaction services provided by Embodiment 1 of the present invention;

[0024] Figure 3 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 1 ;

[0025] Figure 4 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 2 ;

[0026] Figure 5 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 3 ;

[0027] Figure 6 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 4 ;

[0028] Figure 7 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 5 ;

[0029] Figure 8 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 6 ;

[0030] Figure 9 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 7 ;

[0031] Figure 10 Interface schematic of the research data consistency audit system based on customer interaction services provided by Embodiment 1 of the present invention Figure 8 。 Detailed implementation manners

[0032] The following is a further detailed description through specific implementation manners:

[0033] Embodiment 1

[0034] Basically as shown in the appendix: A research data consistency audit method based on customer interaction services, the method includes: Figure 1 S100, obtaining the basic data of at least one audit item and a number of voice samples obtained by interacting with a number of visitors and corresponding respondents

[0035] S100, obtaining the basic data of at least one audit item and a number of voice samples obtained by interacting with a number of visitors and corresponding respondents

[0036] Specifically, for audit projects such as satisfaction surveys in industries like communication and services, questionnaire-based voice interaction surveys are adopted, usually by phone interviews. Of course, it can also be any other channel that can achieve voice interaction. The basic data of the audit project includes user information, questionnaire content, access standards, and quality inspection standards. Among them, user information includes information such as company name and industry; the questionnaire content includes multiple questions and their multiple preset answers; the access standards are the access requirements for the interviewer when interacting with the surveyed object by voice, such as standard introduction language, standard closing language, etc.; the quality inspection standards are the requirements for consistency review of the voice samples obtained from voice interaction, including review indicators and their corresponding standard conditions.

[0037] S200, Analyze the voice samples to form interaction data. Among them, the interaction data includes multiple questions and their answers.

[0038] Specifically, natural language processing technology, text vectorization technology, semantic analysis technology, question-and-answer technology, and large language model technology are adopted, including but not limited to speech-to-text conversion, text segmentation, matching and extraction of questions and answers. Finally, the interaction data of all questions and corresponding answers in the corresponding voice samples of the surveyed object is obtained. There are situations where the surveyed object answers all the questions in the questionnaire content, or only answers some questions, or raises some new questions, etc.

[0039] S210, Conduct a consistency review on the process of analyzing voice samples to form interaction data, including basic record consistency review and voice emotion consistency review. In this embodiment, if the review of this process is all normal and there is no abnormal mark, it is considered that the review is passed and no secondary review is required. If there is an abnormal mark, the review is not passed and a secondary review is required.

[0040] Specifically, deeply mining the text content and emotion content in the voice samples for review is beneficial to improving the accuracy of the correspondence between the voice samples and the interaction data.

[0041] S211, Create a task library to represent the mapping relationship between several audit tasks and audit projects.

[0042] Specifically, a task library is established. According to the audit indicators and their corresponding standard conditions in the regular audit quality inspection standards, each audit indicator is transformed into a specific audit task. For example, if an audit indicator requires a complete recording, an audit task is created accordingly to check whether all questions and corresponding answers are included in the voice sample; if an audit indicator requires the visitor to inform the surveyed object that "the call will be recorded", or the visitor to inform the surveyed object that "it is free to answer calls from others locally, so it is also free to answer our calls", or the visitor to use the name of a third party to inform that "we are an independent survey company and collect your opinions as an independent third party", an audit task is created accordingly to check whether the visitor accurately informs the corresponding content in the voice sample; if an audit indicator requires auditing the outbound call time, an audit task is created accordingly to check whether the outbound call time in the voice sample meets the corresponding duration requirements. The specific audit tasks can be initially set according to the audit indicators in the regular audit quality inspection standards and dynamically adjusted later.

[0043] During the audit, based on the basic data of the audit project, the audit tasks are matched and / or updated in the task library to obtain multiple target audit tasks; the basic record consistency audit of the voice sample and the corresponding interaction data is carried out based on the multiple target audit tasks.

[0044] Specifically, the basic data of the audit project includes the quality inspection standards. All audit indicator features in the quality inspection standards are extracted and matched with all audit tasks in the current task library. If a match is found, it is used as a target audit task. If no match is found, the audit task is updated with the content of the quality inspection standards corresponding to the audit indicator feature and used as a target audit task. Among them, the matching method can use a coding model to code the content of the quality inspection standards, and clustering is carried out using the codes of the content of the quality inspection standards. The same audit task is used for the same class.

[0045] Updating the audit tasks includes the level of the audit tasks and their combined arrangement. Specifically, when a new audit indicator is added, the new indicator is matched with the existing audit tasks and the tasks are rearranged. If the new audit task contains multiple subtasks, the audit task will reinstantiate all relevant subtasks and then combine them into a large task in order. There is a sequence between the small tasks in the new audit task, but it does not affect the original tasks. In addition, if the new audit indicator has the highest matching degree with one of the existing audit tasks, it is updated as a subtask under the existing audit task. Multiple audit tasks are formed in the task library, and each audit task has multiple sequential subtasks. Each audit project matches the corresponding target audit task from the task library based on the quality inspection standards.

[0046] Based on multiple target review tasks, perform a consistency review of the basic records of voice samples and corresponding interaction data. The review time depends on the longest target review task, and not all review tasks will be used for each review item. For example, Project A uses 8 review tasks, and Project B uses 10 review tasks. The review time for Project A depends on the longest one among its 8 review tasks, and the review time for Project B depends on the longest one among its 10 review tasks.

[0047] After that, summarize the results: When all the subtasks under the corresponding review item of this review project are completed and pass the detection, the consistency review of the basic records of this voice sample passes; otherwise, the consistency review of the basic records of this voice sample fails, and the failed tasks and their error information are marked, that is, the corresponding voice segments are marked as abnormal, and text annotations are made in the interaction data. For example, there are abnormalities in Q1, Q3, and incomplete recording, etc. All subtasks can be processed independently and in parallel to improve the data processing efficiency.

[0048] After updating the review tasks, check the review tasks against the review metrics. If some tasks are found to be missing, supplement these tasks to the task library in a timely manner to complete the supplementary missing tasks and ensure that the review tasks fully cover the quality inspection standards of the review project.

[0049] S212. Build an emotion recognition model to respectively recognize the emotions of the visitor and the surveyed object in the voice sample, and perform a voice emotion consistency review of the voice sample and the corresponding interaction data.

[0050] Specifically, the voice sample retains non-verbal information such as tone, emotion, and intonation, which helps to more deeply understand the true intentions, access attitudes, and other emotion information of the surveyed object and the visitor. Therefore, based on the interaction data such as voice intonation, access time, and response time during the access process for consistency judgment, correlation analysis and detection can more deeply analyze the accuracy of the data.

[0051] Use a voiceprint recognition model to assign different voices to different speakers (interviewer, surveyed object), and then use an emotion recognition model to recognize the emotions of the visitor and the surveyed object in the voice sample. The emotions of the visitor can be expressed as positive and negative, and the emotions of the surveyed object can be expressed as whether there is a possibility that the surveyed object is induced by the visitor, so as to judge whether the consistency passes. Specifically as follows:

[0052] In S212, the emotion recognition model includes a visitor emotion recognition sub-model, which is used to judge positive and negative for the voice segments belonging to the visitor. If the judgment is negative, the voice emotion consistency review is abnormal, and the corresponding voice segments judged as negative are marked as abnormal.

[0053] In this embodiment, the emotion2vec model is used to label some data (using annotation) in the interaction scenario of this solution, and the visitor emotion recognition sub-model is obtained by fine-tuning and training according to the emotion2vec model. Specifically: based on the emotion2vec model, a neural network layer is extended, and the training labels are positive and negative, and overfitting to negative, so as to form a binary classification model. The speech and intonation of the visitor are detected, and the samples with negative speech samples of the interviewer are judged as abnormal samples; if the number of audits is small, random audits can be carried out on the positive speech samples.

[0054] The time patterns of most interactions are relatively fixed, and the time patterns of most interview visits are relatively fixed. Therefore, the answering time and the total visit time usually follow certain rules.

[0055] Therefore, in S212, the emotion recognition model includes a respondent emotion recognition sub-model, which is used to calculate the time difference between the answering duration of each question in the speech segment to which the respondent belongs and the average answering duration of the question. If the time difference does not meet the threshold requirement, the speech emotion consistency audit is abnormal, and the question and the corresponding speech segment of the answer are marked as abnormal. The length of time is used to judge the possibility that the visitor induces the respondent.

[0056] In this embodiment, the average answering duration of the question is the average of the answering durations of all respondents corresponding to the visitor corresponding to the respondent, or the average of the answering durations of all respondents for the question. The threshold requirement can be that if the answering duration is 30% to 50% higher than the average value, it is considered that the access time of the question is abnormal. In the case of a large number of audits, only the answers that do not meet the threshold requirement and the corresponding speech segments of the questions can be regarded as abnormal.

[0057] S300, mark the speech segments with abnormal consistency audits in the speech samples as abnormal, and send the speech segments with abnormalities to several review terminals according to a preset distribution method. The review terminals perform secondary review and corresponding review processing based on the review criteria.

[0058] Specifically, the preset distribution method includes evenly distributing the abnormal speech segments to the current valid review terminals, and only one review terminal can process one abnormal speech segment. When evenly distributing, the duration is balanced. For example, there are a total of 100 abnormal speech segments, 242 minutes, and 3 review terminals. Basically, each review terminal is allocated 80 - 85 minutes of abnormal speech segments, but it is necessary to ensure that only one review terminal can process one abnormal speech segment, and the duration difference between the review terminals is minimized for balanced distribution, so as to fairly distribute the review tasks to each review terminal and improve the review efficiency.

[0059] The review criteria integrate at least quality inspection criteria, review tasks, and emotion determination criteria, which can cover all abnormal review requirements and can give review processing methods based on the criteria, such as updating the answers to questions in the interaction data.

[0060] S211 also includes using a large language model to extract all the answers to each question for the same voice sample, repeating the answer extraction several times, judging whether the answers extracted each time are consistent, comparing the number of times of consistency and non - consistency, and determining whether the basic record consistency review passes. If the number of times of consistency is greater than the number of times of non - consistency, the basic record consistency review passes; otherwise, the voice segments corresponding to the question and its answer will be marked as abnormal. In this embodiment, the same question is judged 3 times or 5 times. This number of judgments is a relatively balanced choice that can balance system performance and judgment effect.

[0061] It also includes S500, which is the statistical display of the results of the research process, including the number of system - reviewed samples, the number of re - reviewed samples, the number of samples that failed the review, the scrap rate, the number of people participating in the review, the analysis of the source of sample data, the analysis of re - reviewed samples, the analysis of the data of the executing interviewers, and the analysis of the types of scraped volumes; it also includes a list of abnormal voice segments, which is associated with information such as the reviewer, task progress, task status, review time, and review result.

[0062] As Figure 2 shown, it is to execute the above - mentioned research data consistency review method based on customer interaction services; this solution also provides a research data consistency review system based on customer interaction services.

[0063] The system includes a voice sample acquisition module, a parsing module, a consistency review module, a secondary review module, and a review terminal;

[0064] The voice sample acquisition module is used to acquire the basic data of at least one review item and several voice samples obtained from the interaction between several visitors and the corresponding research objects;

[0065] The parsing module is used to parse the voice samples to form interaction data, where the interaction data includes multiple questions and their answers;

[0066] The consistency review module is used to conduct consistency reviews on the process of parsing voice samples to form interaction data, including basic record consistency review and voice emotion consistency review;

[0067] Among them, the consistency review module includes a task library module for creating a task library to represent the mapping relationship between several review tasks and review items; it also includes a basic record consistency review sub-module that, based on the basic data of the review items, matches and / or updates review tasks in the task library to obtain multiple target review tasks, and conducts a basic record consistency review of voice samples and corresponding interaction data based on the multiple target review tasks; it further includes a voice emotion consistency review sub-module for constructing an emotion recognition model to respectively recognize the emotions of the visitor and the surveyed object in the voice sample, and conduct a voice emotion consistency review of the voice sample and corresponding interaction data.

[0068] The secondary review module is used to mark abnormal voice segments in the voice samples with consistency review abnormalities and send the voice segments with abnormalities to several review terminals according to a preset distribution method.

[0069] The review terminal is used to conduct a secondary review and corresponding review processing based on the review criteria.

[0070] It also includes a display terminal for statistically displaying the results of the research process. Specifically, as shown in Figure 3 、 Figure 4 、 Figure 5 、 Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 and Figure 10 shown.

[0071] It can be understood that the above system can fully execute the above method, and the specific process will not be elaborated here.

[0072] The research data consistency review system and method based on customer interaction services provided in this embodiment, by introducing advanced information technology means and adopting a combination of intelligent review and secondary review, solve the limitations of the traditional manual listening review method, greatly improve the review efficiency, and reduce the labor cost. Due to the introduction of intelligent review, it can increase the total sample volume to a certain extent, reach more surveyed objects, improve the research promotion efficiency, and at the same time recover more questionnaire samples to support the data analysis of the research, further enhancing the research effect and providing more solid data support for the strategic planning of the enterprise. In order to ensure the accuracy of intelligent review and balance the review quantity of secondary review, a consistency review of the intelligent review process and dynamic adjustment of the review strategy are proposed. Based on the constructed task library, by matching and updating review items and review tasks, the review strategy is adjusted to adapt to the changing review requirements, improving the timeliness and adaptability of the review work. Based on the emotion recognition model, it can more deeply understand the true intentions, access attitudes and other emotion information of the surveyed object and the visitor by targeting non-verbal information such as tone, emotion and intonation retained in the voice sample, and improve the accuracy of the review.

[0073] Example 2

[0074] Different from Example 1, an importance level is assigned to each review task, and the importance level of each review task and / or the review passing criteria of the review task are dynamically adjusted according to the review item data or standard differences, that is, the review degree is dynamically adjusted.

[0075] Specifically, the importance levels can be divided into three categories: "important", "general", and "unimportant".

[0076] Dynamically adjust the review passing criteria for review tasks at each importance level according to the review item data. Obtain the total number of voice segments corresponding to the "unimportant" review tasks from all voice samples of the review item data. After performing consistency review on all voice samples for all target review tasks, when the proportion of the number of abnormally marked voice segments detected for the "unimportant" review tasks does not exceed the threshold proportion or does not exceed the threshold number of the total number of voice segments corresponding to the "unimportant" review tasks, allow it to pass the review. The abnormal marks are only retained as the record of this review, but no secondary review statistics are performed.

[0077] Dynamically adjust the review passing criteria for review tasks at each importance level according to the standard differences. Since there is usually a time difference between the actual visit and the actual review, that is, the actual visit is completed in April and the actual review is completed in May, there may be a situation where the visit standard during the visit is different from the quality inspection standard during the review. For example, the visit standard does not require the visitor to inform the surveyed object that "the call will be recorded", while the quality inspection standard requires detecting that the visitor needs to inform the surveyed object that "the call will be recorded". This situation will cause almost all voice samples to be abnormal, further leading to inaccurate reviews.

[0078] Therefore, in order to avoid inaccurate reviews caused by standard differences and excessive abnormal quantities for secondary reviews, dynamically adjust the standard conditions of the corresponding review tasks based on the standard differences, relax the requirements, and make it easier for samples to pass the review. That is, first compare the quality inspection standard and the visit standard, mark the standard differences for the review indicators in the quality inspection standard. When the target review tasks that match carry the importance level, they may also carry the standard difference mark. After performing the review using the target review tasks, sort the abnormally detected voice segments for the review tasks with the standard difference mark according to the importance level of the review tasks. It can be allowed to pass according to the detection method for the unimportant review tasks mentioned above, or the passing review can be limited according to the current number of effective review terminals or the review pressure. The corresponding data that is allowed to pass the review is still marked as abnormal, and the abnormal marks are only retained as the record of this review, without performing secondary review statistics.

[0079] Correspondingly, the task library module is also used to classify the importance levels of the review tasks, and dynamically adjust the review passing criteria for the review tasks at each importance level according to the review project data and standard differences.

[0080] The research data consistency review system and method based on customer interaction services provided in this embodiment can still accurately perform abnormal reviews. For abnormal situations caused by unimportant levels and standard differences, abnormal markings are made but no secondary review push is performed, so as to reduce the review volume pushed to the secondary review, thereby balancing the labor cost of the secondary review and the accuracy of the review. At the same time, it improves the adaptability of the review work during the dynamic change of the review criteria.

[0081] Embodiment Three

[0082] Different from Embodiments One and Two, based on external condition changes, the review order and review degree of the consistency review are dynamically adjusted.

[0083] Specifically, external condition changes, such as an increase or decrease in the number of items to be reviewed, a shortening of the review time, a decrease in the number of review terminals, etc. Therefore, in order to ensure the effective development of the review work while external conditions change, an adaptive adjustment of the review method is required.

[0084] Set up a customer white list and its level. When there are multiple items to be reviewed, the review sorting is performed according to the customer level, and the review is carried out according to the review sorting during the secondary review.

[0085] Further set up the review terminals and their service quality levels. According to the number of review projects, the review terminals are divided into multiple groups according to the service quality level. Each group processes one review project, and the number of review terminals and service quality level in each group are correspondingly matched based on the customer level, ensuring that the review work can cover multiple review projects simultaneously during the same time period, ensuring the orderly development of the review work for multiple customers, and avoiding excessive waiting time for a single customer.

[0086] When the review time is shortened and the number of review terminals decreases, the review degree can be adjusted while the review order is determined, and the review degree can be adjusted in the same way as in Embodiment Two.

[0087] The research data consistency review system and method based on customer interaction services provided in this embodiment can not only ensure the efficient and timely completion of the review tasks according to the established standard requirements without affecting the review effects of other aspects, but also quickly respond when external conditions change, dynamically adapt to the changing situations, flexibly and automatically adjust the review strategy, maintain the timeliness of the review process, enhance the applicability and flexibility of the system, thereby providing more reliable data support and decision-making basis for the enterprise, and promoting the continuous improvement of service quality and the optimization of service experience.

[0088] The above are only embodiments of the present invention. Common general knowledge such as the specific structures and characteristics in the solutions is not described in detail herein. Those of ordinary skill in the art know all the common general knowledge in the technical field to which the invention pertains before the filing date or the priority date, are able to know all the prior arts in this field, and have the ability to apply the conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, complete and implement this solution in combination with their own abilities. Some typical well-known structures or well-known methods should not become obstacles for those of ordinary skill in the art to implement this application. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent.

Claims

1. A method for auditing the consistency of survey data based on customer interaction services, characterized in that: S100, obtaining at least one audit project data, the audit project data including basic data and a number of voice samples obtained by interaction between a number of interviewers and corresponding survey subjects; S200, parsing each voice sample to form corresponding interaction data, wherein the interaction data includes a plurality of questions and their answers; S210, performing consistency review on the process of parsing the voice sample to form interactive data, including basic record consistency review and voice emotion consistency review; S211, creating a task library for characterizing the mapping relationship between a number of audit tasks and audit items; matching and / or updating the audit tasks in the task library according to the basic data of the audit items to obtain multiple target audit tasks; and performing a basic record consistency audit of the voice samples and the corresponding interaction data based on the multiple target audit tasks; S212, constructing an emotion recognition model to respectively recognize the emotions of the interviewer and the surveyed object in the voice sample, and to conduct a voice emotion consistency review of the voice sample and the corresponding interaction data; S300, marking the speech segments with abnormalities in the consistency review of the speech sample as abnormal, and sending the abnormal speech segments to several review terminals according to a preset distribution method. The review terminals perform secondary review and corresponding review processing based on the review standards.

2. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: The basic data of the audit project includes quality inspection standards. All audit indicator features in the quality inspection standards are extracted and matched with all audit tasks in the current task library. If they match, they are used as target audit tasks. If they do not match, the audit tasks are updated with the quality inspection standard content corresponding to the audit indicator features and used as target audit tasks.

3. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: Update audit tasks include the attribution level of audit tasks and their combination arrangement; multiple audit tasks independently perform basic record consistency audits.

4. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: Assign importance levels to each audit task; dynamically adjust the importance level of each audit task and / or the audit pass criteria for the audit task based on audit project data or standard differences.

5. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: In S212, the emotion recognition model includes a visitor emotion recognition sub-model, which is used to make positive and negative judgments on the speech segment belonging to the visitor. If it is judged to be negative, the speech emotion consistency audit is abnormal, and the speech segment judged to be negative is marked as abnormal.

6. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: In S212, the emotion recognition model includes a sub-model of emotion recognition of the surveyed subject, which is used to calculate the time difference between the answer time of each question in the speech segment to which the surveyed subject belongs and the average answer time of the question. If the time difference does not meet the threshold requirement, the speech emotion consistency review is abnormal, and the speech segment corresponding to the question and its answer is marked as abnormal.

7. The method for checking the consistency of survey data based on customer interactive services according to claim 6, characterized in that: The average answering time for the question is the average answering time of all respondents corresponding to the corresponding visitors of the surveyed subject, or the average answering time of all respondents.

8. The method for checking the consistency of survey data based on customer interactive services according to claim 1, characterized in that: S211 includes, for the same voice sample, using a large language model to extract all answers to each question, repeatedly extracting answers several times, judging whether the answers extracted each time are consistent, comparing the number of consistent and inconsistent occurrences, and determining whether the basic record consistency review has passed.

9. The method for checking the consistency of survey data based on customer interactive services according to claim 8, characterized in that: If the number of consistent occurrences is greater than the number of inconsistent occurrences, the basic record consistency review is passed, otherwise the corresponding speech segments of the question and its answer will be marked as abnormal.

10. The survey data consistency review system based on customer interactive services is characterized by: Execute the method for consistency review of survey data based on customer interaction service according to any one of claims 1 to 9; the system comprises a voice sample acquisition module, a parsing module, a consistency review module, a secondary review module and a review terminal; A voice sample acquisition module, used to acquire basic data of at least one audit project and a number of voice samples obtained by interactions between a number of interviewers and corresponding survey subjects; A parsing module, used to parse the voice sample to form interactive data, wherein the interactive data includes multiple questions and their answers; The consistency review module is used to conduct consistency review on the process of parsing voice samples to form interactive data, including basic record consistency review and voice emotion consistency review; Among them, the consistency audit module includes a task library module, which is used to create a task library to characterize the mapping relationship between a number of audit tasks and audit projects; it also includes a basic record consistency audit submodule, which matches and / or updates the audit tasks in the task library according to the basic data of the audit project, obtains multiple target audit tasks, and performs basic record consistency audits on voice samples and corresponding interaction data based on the multiple target audit tasks; it also includes a voice emotion consistency audit submodule, which is used to build an emotion recognition model, respectively recognize the emotions of the visitor and the surveyed object in the voice sample, and perform voice emotion consistency audits on the voice sample and the corresponding interaction data; A secondary review module is used to mark abnormal speech segments in the speech sample that have abnormal consistency review, and send the abnormal speech segments to several review terminals according to a preset distribution method; The review end is used for secondary review and corresponding review processing based on the review standards.

Images