Processing method and application for matching real-time call questions and answers

Abstract

The invention belongs to the technical field of call service processing, and discloses a real-time call question and answer matching processing method, which comprises the following steps of: converting voice into characters in real time, and sending preprocessed voice data into an ASR service for real-time conversion; real-time text processing and grouping are carried out, and texts are extracted, sorted and grouped through role detection and grouping and timestamp correction; and processing the text after the call, and matching the text with the identified groups in the call process. According to the method, calling of the TEXTCNN model is dispersed to the call process, so that the processing method remarkably reduces the time required for centralized text processing after the call is ended, which means that even if the call duration is relatively long, the response duration of filing information can be effectively controlled due to the fact that the processing is dispersed to the whole call period, and the user experience is improved. Therefore, the performance requirement of outputting filing information within 5 seconds after on-hook can be met even in peak hours.

Description

Technical Field

[0001] The present invention belongs to the technical field of call service processing, and specifically relates to a processing method and application of real-time call question and answer matching. Background Art

[0002] In today's customer-centric market environment, telephone calls are increasingly valued by companies as a direct and efficient way of communication. With the popularization of traditional communication methods such as telephones, faxes, and e-mails, and the increasing diversification of customer needs, companies have begun to explore new marketing strategies and service models. Telephone call services came into being under this background and have gradually developed into an important means for companies to enhance their market competitiveness and optimize customer service.

[0003] Currently, after the staff completes a call, the system needs to extract and analyze the call content immediately after the call is hung up, and input the model to complete the user's information file. It will first sort the text according to the time when the speech was generated, then pair and group it according to the role, and finally input the entire call text into the TEXTCNN model for classification and recognition at one time;

[0004] Although this method is convenient, in actual use, the duration of the call will affect the response time of the model. The longer the call duration, the more groups are grouped by role, the more times the TEXTCNN model is called, and the longer the response time of the file information. When the call duration exceeds 30 minutes, it cannot meet the performance requirement of outputting the file information within 5 seconds after hanging up. Moreover, when multiple calls are hung up at the same time, high-concurrency calls to the TEXTCNN model for classification and recognition can easily lead to an avalanche of server resources, making the response time of the file information too long, affecting the file structure, so it needs to be improved and optimized. Summary of the invention

[0005] The purpose of the present invention is to provide a processing method and application for real-time call question and answer matching to solve the problems raised in the above background technology.

[0006] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a processing method for real-time call question and answer matching, comprising the following steps: real-time speech to text conversion, through ASR technology, the pre-processed speech data is sent to the ASR service for real-time conversion; real-time text processing and grouping, through role detection and grouping and timestamp correction, the text is extracted, sorted and grouped; post-call text processing, the text is matched with the groups that have been recognized during the call, and the matched groups are input into the TEXTCNN model to obtain the classification recognition results.

[0007] Preferably, in the real-time text processing and grouping, the specific steps of role detection and grouping are:

[0008] A1, first reads the role information in the message to determine whether the sender of the current message is a consultant or a user;

[0009] A2, when the role of the text message is consultant, check the last message received;

[0010] A3, when the last message is a user message, create a new group; when the last message is not a user message, directly add it to the latest group;

[0011] A4, when the role of the text message is a user, the message is stored in the latest group;

[0012] A5, finally, the current group is input into the TEXTCNN model to obtain the recognition result.

[0013] Preferably, in the real-time text processing and grouping, the specific process of timestamp correction is:

[0014] B1, at the beginning of a call, assign a unique and increasing timestamp to each received text message;

[0015] B2, when receiving a new text message, first extract its timestamp and text content;

[0016] B3, comparing the newly received text message with the stored text message list, in particular with the most recently received text message;

[0017] B4, when the timestamp of the newly received text message is later than the timestamp of the previous text message, grouping is performed according to the normal process, that is, the text message is added to the group of the current speaker;

[0018] B5, when the timestamp of the newly received text message is earlier than the timestamp of the previous text message, reorder them;

[0019] B6, regrouping using the reordered text message list to ensure that the text messages in each group are arranged in the order of actual calls;

[0020] B7, finally, the regrouped text messages are input into the TEXTCNN model to complete the text processing.

[0021] Preferably, the process of real-time speech-to-text conversion is as follows:

[0022] C1, start the ASR service, configure the parameters required for speech recognition, and connect to the call system;

[0023] C2, during the call, captures the voice data of the consultant and the user from the audio stream of the call system;

[0024] C3, preprocessing the captured speech data to improve the recognition accuracy of ASR;

[0025] C4, sends the pre-processed voice data to the ASR service for real-time conversion, and waits for the ASR service to return the converted text results.

[0026] C5, create a message queue for storing the converted text messages, and add each text message to the queue in the form of a message.

[0027] Preferably, the specific process of post-call text processing is:

[0028] D1, after the call ends, all remaining messages are taken out from the message queue and integrated into a complete call text;

[0029] D2, finally grouping the texts according to the role and timestamp information to ensure that the order of the texts is consistent with the actual call order;

[0030] D3, matching the text with the groups that have been identified during the call;

[0031] D4, after the group is fully matched, the cached result is used directly;

[0032] D5, after the grouping fails to match, the grouping is input into the TEXTCNN model to obtain the classification recognition result.

[0033] Preferably, the post-call text processing also includes performance monitoring and model optimization;

[0034] The performance monitoring can monitor the performance indicators of post-call text processing, and record and analyze the performance data.

[0035] Preferably, the model optimization can continuously adjust and optimize the parameters and configurations of the ASR model and the TEXTCNN model according to performance monitoring results and user feedback.

[0036] Preferably, the processing method is applied to telephone call services, responsible for extracting and analyzing the call content and completing user information archiving.

[0037] The beneficial effects of the present invention are as follows:

[0038] 1. The present invention significantly reduces the time required for centralized text processing after the call by distributing the call of the TEXTCNN model throughout the call process. This means that even if the call duration is long, the response time of the archiving information can be effectively controlled because the processing is distributed throughout the call period, thereby improving the real-time performance of the system and user experience.

[0039] 2. In the present invention, when multiple calls are hung up at the same time, since most of the text processing work has been completed during the call, the pressure of high-concurrency calling the TEXTCNN model for classification and recognition is significantly alleviated. This decentralized processing method helps to optimize the utilization of server resources and reduce resource contention and avalanche effects caused by high concurrency, thereby ensuring that the performance requirement of outputting file information within 5 seconds after hanging up can be met even during peak hours.

[0040] 3. The present invention ensures that text messages can be accurately grouped according to the actual call order and roles through strict role detection and timestamp correction mechanisms. This grouping method helps the model to better understand the context of the conversation, thereby improving the accuracy of classification and recognition. At the same time, it further improves the processing efficiency and recognition accuracy of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a question-answer matching flow chart of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, an embodiment of the present invention provides a processing method for real-time call question and answer matching, including the following steps: real-time speech to text conversion, through ASR technology, the pre-processed speech data is sent to the ASR service for real-time conversion; real-time text processing and grouping, through role detection and grouping and timestamp correction, the text is extracted, sorted and grouped; post-call text processing, the text is matched with the groups that have been recognized during the call, and the matched groups are input into the TEXTCNN model to obtain classification recognition results.

[0044] By distributing the calls to the TEXTCNN model throughout the call, this approach significantly reduces the time required to centrally process the text after the call ends. This means that even if the call is long, the response time for archiving information can be effectively controlled because the processing is distributed throughout the call, thereby improving the real-time nature of the system and user experience.

[0045] When multiple calls are on hold at the same time, since most of the text processing work has been completed during the call, the pressure of high-concurrency calls to the TEXTCNN model for classification and recognition is significantly alleviated. This decentralized processing method helps to optimize the utilization of server resources and reduce resource contention and avalanche effects caused by high concurrency, thereby ensuring that the performance requirement of outputting file information within 5 seconds after hanging up can be met even during peak hours.

[0046] Through strict role detection and timestamp correction mechanisms, text messages can be accurately grouped according to the actual call order and roles. This grouping method helps the model better understand the context of the conversation, thereby improving the accuracy of classification and recognition, and at the same time, further improving the processing efficiency and recognition accuracy of the entire system.

[0047] Among them, the specific steps of role detection and grouping in real-time text processing and grouping are:

[0048] After receiving the text message, process it according to the role;

[0049] A1, first reads the role information (consultant / user) in the message to determine whether the sender of the current message is a consultant or a user;

[0050] A2, when the role of the text message is consultant, check the last message received;

[0051] A3, when the last message is a user message, create a new group; when the last message is not a user message, directly add it to the latest group;

[0052] A4, when the role of the text message is a user, the message is stored in the latest group;

[0053] A5, finally, the current group is input into the TEXTCNN model to obtain the recognition result.

[0054] Among them, the specific process of timestamp correction in real-time text processing and grouping is as follows:

[0055] Since the consultant and customer words received during the call are transmitted through dual channels, the information may be different from the actual speaking order of the two parties. In order to reduce the large error between the matched grouping and the actual grouping during the call, a call text timestamp mark is added.

[0056] B1, at the beginning of a call, assign a unique and increasing timestamp to each received text message;

[0057] This timestamp should be accurate to the millisecond level to ensure that it accurately reflects the actual time the text message was received.

[0058] B2, when receiving a new text message, first extract its timestamp and text content;

[0059] B3, comparing the newly received text message with the stored text message list, in particular with the most recently received text message;

[0060] B4, when the timestamp of the newly received text message is later than the timestamp of the previous text message, grouping is performed according to the normal process, that is, the text message is added to the group of the current speaker;

[0061] B5, when the timestamp of the newly received text message is earlier than the timestamp of the previous text message, reorder them;

[0062] B6, regrouping using the reordered text message list to ensure that the text messages in each group are arranged in the order of actual calls;

[0063] B7, finally, the regrouped text messages are input into the TEXTCNN model to complete the text processing.

[0064] Each complete group (i.e., a round of conversation) is preliminarily classified and recognized, and the TEXTCNN model is used to quickly process the text in the group to obtain the classification and recognition results.

[0065] The process of real-time speech-to-text conversion is as follows:

[0066] C1, start the ASR service, configure the parameters required for speech recognition, and connect to the call system;

[0067] Such as language and sampling rate, ensuring that the ASR service can seamlessly connect with the call system and receive voice data in real time;

[0068] C2, during the call, captures the voice data of the consultant and the user from the audio stream of the call system;

[0069] C3, pre-processing the captured speech data, such as denoising and enhancement, to improve the recognition accuracy of ASR;

[0070] C4, sends the pre-processed voice data to the ASR service for real-time conversion, and waits for the ASR service to return the converted text results.

[0071] The ASR service returns the converted text results, including text content, confidence level and other information;

[0072] C5, create a message queue for storing the converted text messages, and add each text message (including text content, role, and timestamp) to the queue in the form of a message.

[0073] The specific process of post-call text processing is as follows:

[0074] D1, after the call ends, all remaining messages are taken out from the message queue and integrated into a complete call text;

[0075] D2, finally grouping the texts according to the role and timestamp information to ensure that the order of the texts is consistent with the actual call order;

[0076] D3, matching the text with the groups that have been identified during the call;

[0077] D4, after the group is fully matched, the cached result is used directly;

[0078] D5, after the grouping fails to match, the grouping is input into the TEXTCNN model to obtain the classification recognition result.

[0079] Among them, post-call text processing also includes performance monitoring and model optimization;

[0080] Performance monitoring can monitor the performance indicators of post-call text processing, such as processing time, throughput, accuracy, etc., record and analyze performance data to identify potential problems and optimization points.

[0081] Among them, model optimization can continuously adjust and optimize the parameters and configurations of the ASR model and TEXTCNN model based on performance monitoring results and user feedback.

[0082] More advanced data augmentation, regularization, hyperparameter search and other techniques can be used to improve the generalization ability and recognition accuracy of the model.

[0083] Among them, this processing method is applied to telephone call services, responsible for extracting and analyzing the call content and completing the user's information archiving.

[0084] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0085] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing real-time call question and answer matching, Features: The following steps are involved: Real-time speech-to-text conversion: through ASR technology, the pre-processed speech data is sent to the ASR service for real-time conversion; Real-time text processing and grouping: extract, sort and group text through role detection and grouping and timestamp correction; post-call text processing: match the text with the groups that have been identified during the call, and input the matched groups into the TEXTCNN model to obtain classification and recognition results.

2. A method for processing real-time call question and answer matching according to claim 1, Features: In the real-time text processing and grouping, the specific steps of role detection and grouping are: A1, first reads the role information in the message to determine whether the sender of the current message is a consultant or a user; A2, when the role of the text message is consultant, check the last message received; A3, when the last message is a user message, create a new group; when the last message is not a user message, directly add it to the latest group; A4, when the role of the text message is a user, the message is stored in the latest group; A5, finally, the current group is input into the TEXTCNN model to obtain the recognition result.

3. A method for processing real-time call question and answer matching according to claim 1, Features: In the real-time text processing and grouping, the specific process of timestamp correction is as follows: B1, at the beginning of a call, assign a unique and increasing timestamp to each received text message; B2, when receiving a new text message, first extract its timestamp and text content; B3, comparing the newly received text message with the stored text message list, in particular with the most recently received text message; B4, when the timestamp of the newly received text message is later than the timestamp of the previous text message, grouping is performed according to the normal process, that is, the text message is added to the group of the current speaker; B5, when the timestamp of the newly received text message is earlier than the timestamp of the previous text message, reorder them; B6, regrouping using the reordered text message list to ensure that the text messages in each group are arranged in the order of actual calls; B7, finally, the regrouped text messages are input into the TEXTCNN model to complete the text processing.

4. A method for processing real-time call question and answer matching according to claim 1, Features: The process of real-time speech-to-text conversion is as follows: C1, start the ASR service, configure the parameters required for speech recognition, and connect to the call system; C2, during the call, captures the voice data of the consultant and the user from the audio stream of the call system; C3, preprocessing the captured speech data to improve the recognition accuracy of ASR; C4, sending the pre-processed voice data to the ASR service for real-time conversion, and waiting for the ASR service to return the converted text result; C5, create a message queue for storing the converted text messages, and add each text message to the queue in the form of a message.

5. A method for processing real-time call question and answer matching according to claim 1, Features: The specific process of post-call text processing is as follows: D1, after the call ends, all remaining messages are taken out from the message queue and integrated into a complete call text; D2, finally grouping the texts according to the role and timestamp information to ensure that the order of the texts is consistent with the actual call order; D3, matching the text with the groups that have been identified during the call; D4, after the group is fully matched, the cached result is used directly; D5, after the grouping fails to match, the grouping is input into the TEXTCNN model to obtain the classification recognition result.

6. A method for processing real-time call question and answer matching according to claim 1, Features: The post-call text processing also includes performance monitoring and model optimization; The performance monitoring can monitor the performance indicators of post-call text processing, and record and analyze the performance data.

7. A method for processing real-time call question and answer matching according to claim 6, Features: The model optimization can continuously adjust and optimize the parameters and configurations of the ASR model and the TEXTCNN model according to the performance monitoring results and user feedback.

8. Application of the method for real-time call question-answer matching according to claim 1, Features: The processing method is applied to telephone call services, responsible for extracting and analyzing the call content and completing the user information archiving.

Images