Model training method and related device

Abstract

The embodiment of the invention provides a model training method and a related device, and the method comprises the steps: inputting customer service call data into a first model for feature extraction, and obtaining a first feature table which comprises the call features of each call round and the position information of each call feature, if a first call feature screened from the first feature table is obtained, detecting the first call feature through a pre-configured second feature table to obtain a detection result, and writing the customer service call data and the position information of the first call feature into a parameter adjustment prompt template when the detection result is that the detection is passed, and inputting the first parameter adjustment prompt data into the first model, and performing parameter adjustment by the first model according to the first parameter adjustment prompt data. According to the embodiment of the invention, the feature extraction accuracy of the first model can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a model training method and related devices. Background Art

[0002] With the continuous development and progress of the Internet and information technology, more and more services are being provided through the Internet. In order to improve users' perception of services and the convenience of users' participation in services, users can call the service provider's customer service to provide feedback and inquiries on service-related content to solve problems encountered by users in participating in services. Customer service can also recommend and remind users of related services through calls with users. Users are becoming more and more involved in services based on the Internet, and calling customers to provide feedback and inquiries is becoming more and more common. How to improve the management of data generated by calls between users and customer service is a focus of increasing attention for service providers. Summary of the Invention

[0003] In a first aspect, an embodiment of the present application provides a model training method, comprising: Inputting customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table to obtain a detection result; If the detection result is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0004] It can be seen that in the embodiment of the present application, the customer service call data is first input into the first model for feature extraction to obtain a first feature table including call features of each call round and location information of each call feature. In this way, feature extraction is performed through the first model to improve the efficiency and accuracy of feature extraction, so that screening can be performed through the first feature table, thereby improving the efficiency of call feature screening; then, the first call feature screened out from the first feature table is detected through the pre-configured second feature table to obtain a detection result. If the detection result is a pass, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template to obtain first parameter adjustment prompt data and input into the first model. The first model performs parameter adjustment according to the first parameter adjustment prompt data. In this way, the first call feature screened out from the first feature table is detected through the pre-configured second feature table, which not only realizes the detection of the screening ability, but also realizes the detection of the feature extraction ability of the first model. Further, if the detection result is a pass, the customer service call data and the location information of the first call feature are used to adjust the parameters of the first model to improve the accuracy of the location information of each call feature in the first feature table output by the first model, thereby improving the convenience of call feature screening.

[0005] In a second aspect, an embodiment of the present application provides a model training device, comprising: a feature extraction module, configured to input customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, a detection module is run, wherein the detection module is configured to detect the first call feature using a pre-configured second feature table to obtain a detection result; When the detection result is passed, the writing module is run, and the writing module is used to write the customer service call data and the location information of the first call feature into the parameter adjustment prompt template, obtain the first parameter adjustment prompt data and input it into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0006] In a third aspect, an embodiment of the present application provides a model training device, comprising: a processor; and a memory configured to store computer-executable instructions, which, when executed, cause the processor to execute the model training method described in the first aspect.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium for storing computer-executable instructions, which, when executed by a processor, implement the model training method as described in the first aspect.

[0008] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the model training method described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in this specification. Those skilled in the art can also derive other drawings based on these drawings without inventive work. Figure 1 A schematic diagram of an implementation environment for a model training method provided in an embodiment of the present application; Figure 2 A flowchart of a model training method provided in an embodiment of the present application; Figure 3 A flowchart of a model training method for call feature screening provided in an embodiment of the present application; Figure 4 A schematic diagram of a model training device provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of a model training device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0011] In actual applications, customer service call data is too long and contains irrelevant data that does not represent user needs or feedback. Direct processing based on customer service call data will result in a large amount of ineffective processing, resulting in loss of processing resources and a waste of processing time. Therefore, feature extraction can be performed on customer service call data to characterize the customer service call data using its call features. In order to achieve effective and accurate call feature screening, an embodiment of the present application provides a model training method, which uses a first model to extract features from customer service call data to obtain a first feature table including call features of each call round and location information of each call feature, so that the first call feature representing the customer service call data can be screened out from the first feature table, thereby improving the efficiency of feature extraction and also improving the efficiency and convenience of call feature screening; in the process of call feature screening with the help of the first model, the accuracy of feature extraction performed by the first model affects the efficiency of call feature screening to a certain extent. In order to improve the feature extraction ability of the first model, after extracting features from the customer service call data through the first model and obtaining the first feature table, if the first call feature screened out from the first feature table is obtained, the second feature table is introduced to detect the first call feature, thereby achieving dual detection of screening ability and feature extraction ability; if the detection passes, the first parameter adjustment prompt data containing at least the customer service call data and the location information of the first call feature is input into the first model, so that the first model performs parameter adjustment, thereby improving the accuracy of feature extraction of the first model.

[0012] The model training method provided in one or more embodiments of this specification can be applied to the implementation environment of the model training of the first model, such as Figure 1 As shown, the implementation environment includes at least a server 101 and a first model 102; The server 101 may be one or more servers, a server cluster consisting of several servers, or a cloud server of a cloud computing platform. The server 101 is used to call the first model 102 to extract features from customer service call data and to prepare data for adjusting parameters of the first model 102. The server 101 may also store a second feature table for detecting call features obtained through screening. It should be noted that the second feature table may also be stored in a database independent of the server 101, which is not limited in this embodiment. The first model 102 may be a Large Language Model (LLM). The first model 102 uses advanced algorithms such as the Transformer architecture for feature extraction and parameter adjustment. In addition, the implementation environment may also include a customer service terminal device 103 and a user terminal device 104, wherein the customer service terminal device 103 and the user terminal device 104 may be a mobile phone, a personal computer, a tablet computer, an e-book reader, a device for information interaction based on VR (Virtual Reality), an in-vehicle terminal, an IoT device, a wearable smart device, a laptop computer, a desktop computer, etc.; customer service and users communicate via the customer service terminal device 103 and the user terminal device 104; the customer service terminal device 103 is used to interact with the server 101 to implement call feature screening; In this implementation environment, the server 101 inputs the customer service call data into the first model 102 for feature extraction, and obtains a first feature table including the call features of each call round and the location information of each call feature. If the first call feature screened out from the first feature table is obtained, the first call feature is detected through the pre-configured second feature table to obtain a detection result. If the detection result is that the detection passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, and the first parameter adjustment prompt data is obtained and input into the first model 102. The first model 102 adjusts the parameters according to the first parameter adjustment prompt data. In this way, by extracting features and adjusting parameters at the same time, the convenience of call feature screening is improved while the accuracy of feature extraction is improved.

[0013] Figure 2 This is a processing flow chart of a model training method provided in this embodiment, refer to Figure 2 , a model training method provided in this embodiment specifically includes steps S202 to S206: Step S202: Input the customer service call data into the first model for feature extraction to obtain a first feature table.

[0014] The customer service call data in this embodiment includes voice data obtained by recording customer service calls with users. The first model can be a Large Language Model (LLM). Furthermore, the first model in this embodiment can also be a pre-trained natural language model. The model can utilize foundation models or pretrained models. The specific architecture of the first model can be a neural network architecture with large batch parameters, a Transform architecture, or other architectures. During implementation, the first model can directly utilize the foundation model or pretrained model. Furthermore, the foundation model or pretrained model can be fine-tuned (Supervised Fine-Tuning (SFT)) based on the foundation model or pretrained model for tasks such as feature extraction and parameter adjustment to obtain a model capable of performing these tasks. Examples include various open-source large-scale language models such as chatGPT (chatGenerative Pre-trained Transformer). The first model in this embodiment can be trained on a large language model using pre-labeled training samples.

[0015] In practical applications, if customer service call data is directly stored and processed, since customer service call data is voice data and may include voice data that does not represent the user's call intent, it is possible to extract the user's call intent as a call feature of the customer service call data and label the customer service call data based on the call features. This allows for efficient and convenient perception and management of the call intent corresponding to the customer service call data through call features. The call features in this embodiment include data representing the call intent corresponding to the customer service call data; in addition, the call features may also include a traffic summary related to the business and / or call intent obtained by summarizing the customer service call data. To further improve the accuracy of call features in representing customer service call data, the call features in this embodiment may be in the form of "first-level classification-second-level classification-third-level classification"; for example, complaint-resource return reminder-caused interruption, complaint-abnormal feedback-denial of resource request, post-resource request-negotiated resource return-additional resource reduction, post-resource request-resource return completion-early resource return.

[0016] During specific implementation, the customer service call data is input into the first model for feature extraction to obtain a first feature table; optionally, the first feature table includes the call features of each call round and the location information of each call feature.

[0017] If the customer service call data is voice data, to enable the first model to effectively extract features, the customer service call data may be preprocessed to obtain input data, which is then input into the first model for feature extraction to obtain the first feature table. Optionally, the preprocessing includes voice recognition and text conversion.

[0018] Specifically, voice recognition may be performed on the customer service call data to obtain call text, and then the call text may be converted into text to obtain input data, which may then be input into the first model for feature extraction.

[0019] For example, ASR (Automatic Speech Recognition) is first used to perform speech recognition on customer service call data to obtain call text. Natural language processing (NLP) is then used to convert the call text into text to obtain input data. The input data includes machine language that the first model can recognize.

[0020] In this embodiment, in order to improve the effectiveness of call feature screening and avoid the situation where feature extraction only outputs one call feature but the call feature is not accurate enough and call feature screening cannot be performed, in this embodiment, the first model can perform feature extraction on the sub-call data of each call round contained in the customer service call data to obtain the call features of each call round, and then generate a first feature table based on the call features of each call round; in order to improve the efficiency of call feature screening, the first model can determine the location information of each call feature in the process of generating the first feature table, and then generate the first feature table based on the call features of each call round and the location information of each call feature, so that when screening from the first feature table, the call features of the front position points can be preferentially screened, thereby improving the convenience and efficiency of call feature screening; based on this, in an optional implementation manner provided by this embodiment, the first model uses the following method to perform feature extraction: Performing feature extraction on the sub-call data of each call round contained in the customer service call data to obtain call features of each call round; Calculate the similarity between the sub-call data of each call round and the corresponding benchmark call data; The location information of each call feature is determined according to the similarity.

[0021] Optionally, the sub-call data of each call round includes the customer service conversation data and user conversation data corresponding to each call round; in addition, in order to improve the accuracy of the call features of each call round, the sub-call data of each call round may also include the customer service conversation data and user conversation data of the previous call round.

[0022] During the specific execution process, we first perform feature extraction on the sub-call data of each call round contained in the customer service call data to obtain the call features of each call round, and then determine the location information of the call features of each call round based on the similarity between the sub-call data of each call round and the corresponding benchmark call data.

[0023] Specifically, the benchmark call data corresponding to the customer service conversation data can be pre-configured. In the process of determining the location information of each call feature, the corresponding benchmark call data is first read based on the customer service conversation data of each call round, and then the similarity between the user conversation data of each call round and the corresponding benchmark call data is calculated. The call features are arranged in descending order according to the similarity to obtain the location information of each call feature. At this point, a first feature table including the call features of each call round and the location information of each call feature is obtained.

[0024] It should be noted that in this embodiment, a customer service call data may include one or more call rounds. After the customer service announces the customer service conversation data to the user, the customer service may receive user feedback, and one customer service announcement and one user feedback are considered a call round. In addition, user questions and customer service feedback can also be considered a call round. That is, the order of customer service conversation data and user conversation data within a call round is not limited. It should also be noted that the customer service conversation data in this embodiment includes not only voice data generated by customer service calls with users, but also voice data generated by user-to-user calls. In addition, it can also be text data generated by conversations in the form of text messages or in-app messages, which are not limited in this embodiment.

[0025] In this embodiment, the customer service call data input into the first model can be the sub-call data of a call round after the end of a call round, or the sub-call data of the call round and the sub-call data of historical call rounds, or the sub-call data of all call rounds after the end of a call process. For example, during a call between a customer service representative and a user, after obtaining the sub-call data of each call round (customer service conversation data and user conversation data), the sub-call data is input into the first model for feature extraction. The first model extracts features based on the sub-call data and the sub-call data of historical call rounds to obtain the call features of the call round. At this time, the first model does not output the call features. The above process is repeated until the extraction end prompt data is received. If the extraction end prompt data is received, the location information of each call feature is determined based on the extraction end prompt data, and a first feature table is obtained and output. In an optional implementation provided by this embodiment, the process of inputting the customer service call data into the first model for feature extraction to obtain the first feature table can be implemented as follows: Obtaining sub-call data of a first call round, and inputting the sub-call data of the first call round and feature extraction prompt data into a first model; the first model extracts features based on the sub-call data of the first call round to obtain call features of the first call round, and stores first candidate call features according to the feature extraction prompt; Obtaining sub-call data of the second call round and inputting the sub-call data of the second call round into the first model; the first model extracting features based on the sub-call data of the first call round and the second call round to obtain and store call features of the second call round; Obtaining sub-call data of the third call round and inputting the sub-call data of the third call round into the first model; the first model extracting features based on the sub-call data of the first call round, the second call round, and the third call round to obtain and store call features of the third call round; Obtain the sub-call data of the fourth call round and the call end instruction submitted by the customer service, and input the sub-call data of the fourth call round and the extracted end prompt data into the first model; the first model performs feature extraction based on the sub-call data of the first call round, the second call round, the third call round and the fourth call round to obtain the call features of the fourth call round, and generates and outputs a first feature table including the call features of each call round and the location information of each call feature based on the extracted end prompt data.

[0026] The feature extraction prompt data includes prompt data for prompting the first model to start feature extraction, for example, "Start a call, please start feature extraction." The extraction end prompt data includes prompt data for prompting the first model to end a call, end feature extraction, and generate a feature table, for example, "End a call, please generate a feature table according to."

[0027] Specifically, during feature extraction by the first model, feature extraction may be performed based on the sub-call data of each call turn and the sub-call data of historical call turns to obtain call features for each call turn. If extraction end prompt data is detected, a first feature table is generated based on the call features of each call turn and outputted. Feature extraction may also be performed based on the sub-call data of each call turn and the sub-call data of historical call turns to obtain call features and similarities for each call turn. If extraction end prompt data is detected, location information for each call feature is determined based on the similarity for each call turn, to obtain a first feature table including the call features of each call turn and the location information for each call feature, and outputted.

[0028] Correspondingly, if the server detects a call start instruction, it inputs the acquired sub-call data and feature extraction prompt data into the first model for feature extraction. After acquiring the sub-call data for each call round, it inputs the sub-call data into the first model for feature extraction. If a call end instruction is detected, the extracted end prompt data is input into the first model to generate a feature table. In other words, step S202 can also be replaced by inputting the customer service call data into the first model for feature extraction; if a call end instruction is detected, the extracted end prompt data is input into the first model to generate a feature table, resulting in a first feature table. This, together with one or more other processing steps provided in this embodiment, constitutes a new implementation.

[0029] It should be noted that each of the above call turns and historical call turns refers to call turns within the same call. The above description uses the first, second, third, and fourth call turns as examples. In actual applications, the number of call turns can be fewer or more than the above, and similar processing can be used. This embodiment does not limit this.

[0030] In addition to obtaining sub-call data for each call turn and inputting this sub-call data into the first model for feature extraction, and generating a first feature table output by the first model after detecting a call end instruction, the customer service call data may also be written into a feature extraction prompt template to generate feature extraction prompt data, which is then input into the first model for feature extraction to generate a first feature table. Here, the customer service call data includes sub-call data for all call turns of the call. During this feature extraction process, the first model may first divide the customer service call data by call turn to obtain sub-call data for each call turn, then perform feature extraction on the sub-call data for each call turn to obtain call features for each call turn, and then generate a first feature table based on the call features for each call turn. Specifically, the process of generating the first feature table may be based on the call features and similarity of each call turn. The specific details are similar to those described above and will not be further described in this embodiment.

[0031] During specific implementation, in order to determine the call features of the customer service call data, call feature screening can be performed after obtaining the first feature table; the call feature screening in this embodiment includes call feature selection by the customer service.

[0032] Step S204: If the first call feature selected from the first feature table is obtained, the first call feature is detected using the pre-configured second feature table to obtain a detection result.

[0033] The second feature table includes a pre-configured feature table consisting of call features that have passed validity detection.

[0034] During specific implementation, if the first call feature filtered out from the first feature table is obtained, in order to improve the effectiveness of the first call feature obtained by filtering out, and to avoid the first call feature filtered out by the customer service being unable to accurately represent the customer service call data due to insufficient customer service experience, etc., in this embodiment, the first call feature is detected through a pre-configured second feature table to obtain a detection result, thereby improving the standardization of the call features of the customer service call data.

[0035] During the specific execution process, the first call feature is detected through a pre-configured second feature table. In the process of obtaining the detection result, it is detected whether the first call feature is a call feature in the second feature table, that is, whether the first call feature is in the second feature table. If so, it is determined that the detection has passed; if not, it is determined that the detection has failed.

[0036] In order to increase the diversity of call features in the second feature table, in an optional implementation provided by this embodiment, the second call features can be created in the following manner: Input customer service call data within the time interval into the first model for feature extraction to obtain a historical feature table; A validity check is performed on each call feature in the historical feature table, and a second feature table is constructed based on the call features that pass the validity check.

[0037] The validity check in this embodiment includes generating an audit task for the call feature and sending it to the audit user so that the audit user can audit the call feature. If the audit user passes the audit, it is determined that the validity check of the call feature has passed. If the audit user fails the audit, it is determined that the validity check of the call feature has failed.

[0038] In addition, in addition to performing validity checks on the call features in the historical feature table and constructing a second feature table based on the call features that pass the validity check, in order to further improve the diversity of the call features in the second feature table, call features can also be added to the second feature table by reviewing user configurations. This is not limited in this embodiment.

[0039] Step S206: If the detection result is passed, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain first parameter adjustment prompt data and input it into the first model.

[0040] During specific implementation, when the first call feature is detected through a pre-configured second feature table and the detection result is that the detection is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model, and the first model adjusts the parameters according to the first parameter adjustment prompt data, thereby realizing feature extraction and parameter adjustment at the same time.

[0041] During the specific execution process, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model, and the first model performs parameter adjustment according to the first parameter adjustment prompt data, that is, the process of performing reinforcement learning on the first model based on the customer service call data and the location information of the first call feature. In an optional implementation manner provided by this embodiment, during the process of the first model performing parameter adjustment according to the first parameter adjustment prompt data, the following operations are performed: calculating a first feedback deviation based on the preset location information included in the first parameter adjustment prompt data and the location information of the first call feature, and adjusting the parameters of the first model based on the first feedback deviation to obtain a second model; Inputting the customer service call data included in the first parameter adjustment prompt data into the second model for feature extraction to obtain a third feature table; Calculating a second feedback deviation based on the preset position information and the position information of the first call feature in the third feature table; If the second feedback deviation is greater than a deviation threshold, parameters of the second model are adjusted according to the second feedback deviation.

[0042] Optionally, the preset position information includes the first order.

[0043] Specifically, first, the first feedback deviation is calculated based on the preset position information and the position information of the first call feature contained in the first parameter adjustment prompt data, the parameters of the first model are adjusted according to the first feedback deviation to obtain the second model, the customer service call data contained in the first parameter adjustment prompt data is input into the second model for feature extraction to obtain a third feature table, and the second feedback deviation is calculated based on the preset position information and the position information of the first call feature in the third feature table. If the second feedback deviation is equal to the deviation threshold, the first model is replaced by the second model, that is, the second model is used as the model for feature extraction of the next customer service call data; if the second feedback deviation is greater than the deviation threshold, the parameters of the second model are adjusted according to the second feedback deviation, and the above process is repeated until the feedback deviation of the obtained model is equal to the deviation threshold.

[0044] In the process of adjusting the parameters of the second model based on the second feedback deviation, in order to improve the effectiveness of the parameter adjustment, the reward can be determined based on the second feedback deviation and the first feedback deviation, the training loss can be calculated based on the reward, and then the parameters of the second model can be adjusted based on the training loss.

[0045] For example, the position information of the first call feature is the fifth place, and the preset position information is the first place. The preset reward mechanism in reinforcement learning is: if the first call feature approaches the first place, a positive reward is given; if the position of the first call feature remains unchanged or remains in the first place, a negative reward is given; the first feedback deviation is 5-1=4; if after calculation, the second feedback deviation is 3, and the second feedback deviation is reduced relative to the first feedback deviation, it means that the first call feature in the third feature table output by the second model is close to the first place, then a positive reward is given, and the parameters of the second model are adjusted based on the positive reward.

[0046] In specific implementation, when the first call feature is detected through a pre-configured second feature table and the detection result is that the detection fails, in order to improve the comprehensiveness of the second feature table, this embodiment provides an optional implementation method. When the first call feature is detected through a pre-configured second feature table and the detection result is that the detection fails, the first call feature is tested for validity; if the validity test passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the second parameter adjustment prompt data is obtained and input into the first model, and the first model performs parameter adjustment according to the second parameter adjustment prompt data; in order to further improve the diversity of the second feature table, the first call feature can also be updated to the second feature table after the validity test passes; if the validity test fails, no processing is performed.

[0047] In this embodiment, in order to improve the customer service's perception of call feature screening, in addition to screening from the first feature table, a pre-configured second feature table and a fourth feature table determined according to the number of times screened are provided to screen the call features of the customer service call data; that is, in addition to inputting the customer service call data into the first model for feature extraction to obtain the first feature table, the second feature table can also be read, and / or the feature table generation prompt data can be input into the first model for feature table generation to obtain the fourth feature table, and the first feature table, the second feature table and the fourth feature table are sent to the customer service for call feature screening.

[0048] The above specifically describes the processing procedure after obtaining the first call feature filtered out from the first feature table; in addition, call features can also be filtered out from the second feature table or the fourth feature table. The following specifically describes the processing procedure after filtering out call features from the second feature table and the fourth feature table.

[0049] (1) For the second call feature selected from the fourth feature table, in order to improve the effectiveness of the second call feature obtained by the selection, the second call feature is detected through the second feature table to obtain the detection result. Since the second call feature is not the call feature output by the first model, it is impossible to perform reinforcement learning on the first model based on the location information. Therefore, when the detection result is that the detection passes, the customer service call data can be used as sample data and the second call feature as the sample label to adjust the parameters of the first model, that is, to fine-tune the first model.

[0050] In an optional implementation provided by this embodiment, if the second call feature filtered out from the fourth feature table is obtained, the second call feature is detected through the second feature table to obtain a detection result; if the detection result is that the detection passes, the customer service call data is used as sample data, and the second call feature is used as the sample label to adjust the parameters of the first model.

[0051] Optionally, the position information of the second call feature in the fourth feature table is determined according to the number of times the second call feature is screened.

[0052] Furthermore, in order to improve the comprehensiveness of the second feature table, this embodiment provides an optional implementation. When the test result of the second call feature through the second feature table is that the test fails, the second call feature is tested for validity. If the validity test passes, the customer service call data is used as sample data and the second call feature is used as the sample label to adjust the parameters of the first model. If the validity test fails, no processing is performed.

[0053] It should be noted that after the validity check of the second call feature is passed, the second call feature can be updated to the second feature table.

[0054] In a specific implementation, the fourth feature table may include historical call features within a time interval and location information of each historical call feature. In an optional implementation provided by this embodiment, the fourth feature table may be generated in the following manner: Read historical call features within a time interval; The location information of each historical call feature is determined according to the number of times each historical call feature is screened within the time interval.

[0055] Specifically, the historical call features are sorted in descending order according to the number of times they are screened within the time interval to obtain the location information of each historical call feature, thereby obtaining a fourth feature table containing the historical call features within the time interval and the location information of each historical call feature.

[0056] The time interval includes historical time intervals, such as the most recent week, the most recent month, or the most recent six months. The historical call features include call features that are filtered as call features of customer service call data within the time interval. The number of times the historical call features are filtered includes the number of times the historical call features are selected by customer service within the time interval; the fourth feature table can be a feature table obtained by arranging the historical call features within the most recent week in descending order according to the number of times they are filtered. It should be noted that the fourth feature table can be obtained by sorting the number of times the historical call features are filtered within a time interval, or by sorting the number of times the historical call features are filtered within multiple time intervals, that is, the fourth feature table can be one or more, and this embodiment does not limit this.

[0057] In the process of sorting the historical call features within a time interval according to the number of times they were screened to obtain the fourth feature table, the historical call features may be sorted according to the number of times they were screened within the time interval, which is pre-stored, to obtain the fourth feature table. Furthermore, feature table generation prompt data may be input into the first model to generate the feature table, thereby obtaining the fourth feature table. The feature table generation prompt data may include generating a feature table based on the number of times the historical call features were screened within the most recent week, and the first model generating the fourth feature table based on the feature table generation prompt data.

[0058] (2) With respect to the third call feature selected from the second feature table, since it is selected from the second feature table, it is not necessary to test the third call feature using the second feature table. In an optional implementation provided by this embodiment, if the third call feature selected from the second feature table is obtained, the customer service call data is used as sample data and the third call feature is used as a sample label to adjust the parameters of the first model. Optionally, the third call feature in the second feature table includes call features that have passed the validity test.

[0059] The above describes the process of sending the first feature table, the second feature table, and / or the fourth feature table together to customer service for call feature screening, and obtaining the screened call features. In addition, feature table switching can also be performed. For example, customer service call data is input into the first model for feature extraction to obtain a first feature table and send it to customer service. If a switching request is detected, prompt data is generated from the feature table and input into the first model to generate a feature table to obtain a fourth feature table and send it to customer service. If a second call feature selected from the fourth feature table is obtained, the second call feature is tested using the second feature table to obtain a test result. If the test result is that the test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model. If the test result is that the test fails, the second call feature is tested for validity. If the validity test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model. After sending the fourth feature table to customer service, if a second switching request is detected, the second feature table is read and sent to customer service. If the third call feature filtered out from the second feature table is obtained, the parameters of the first model are adjusted using the customer service call data as sample data and the third call feature as the sample label.

[0060] It should be noted that, in order to further ensure the accuracy of the determined call features of the customer service call data, in this embodiment, the call features of the customer service call data may also be tested according to a test cycle. Specifically, if it is detected that the test cycle has expired, a preset number of call features of the customer service call data within the test cycle are randomly sampled and the call features of the customer service call data are tested for validity. If the validity test passes, no processing is performed. If the validity test fails, the parameters of the first model are adjusted using the customer service call data as sample data and the call features determined by the review user as sample labels. In addition, customer service personnel may be trained to enable them to accurately and effectively screen call features. During the validity test of the call features of the customer service call data, an audit task for the customer service call data and call features is generated and sent to the review user. If the audit user passes the audit, the validity test of the call features of the customer service call data is determined to have passed. If the audit user fails the audit, the validity test of the call features of the customer service call data is determined to have failed.

[0061] In summary, the model training method provided in this embodiment inputs customer service call data into the first model for feature extraction to obtain a first feature table including call features of each call round and location information of each call feature, inputs feature table generation prompt data into the first model for feature table generation to obtain a fourth feature table, and reads a pre-configured second feature table, and sends the first feature table, the second feature table, and the fourth feature table to the customer service. In this way, feature extraction is performed through the first model, and the first feature table that better matches the customer service call data is sent to the customer service so that the customer service performs feature screening from the first feature table, thereby improving the efficiency and effectiveness of feature extraction and the efficiency of call feature screening. On this basis, the second feature table and the fourth feature table are also provided to the customer service so that the customer service can screen the call features from the second feature table or the fourth feature table if the call features cannot be screened out through the first feature table. If a first call feature filtered out from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table. If the detection result is that the detection is passed, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain first parameter adjustment prompt data and input the data into the first model. The first model performs parameter adjustment according to the first parameter adjustment prompt data. If the detection result is that the detection fails, if the validity test of the first call feature is passed, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain second parameter adjustment prompt data and input the data into the first model. The first model performs parameter adjustment according to the second parameter adjustment prompt data. In this way, the accuracy of the location information of the call feature output by the first model is improved. If the second call feature selected by the customer service in the fourth feature table is detected, the second call feature is tested according to the second feature table. If the test result is that the test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model. If the test result is that the test fails, if the validity test of the second call feature passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model; thereby improving the accuracy and validity of the call features output by the first model; If the third call feature selected by the customer service representative in the second feature table is detected, the customer service representative's call data is used as sample data and the third call feature is used as a sample label to adjust the parameters of the first model. This improves the accuracy and effectiveness of the call features output by the first model. In this way, parameters are updated while feature extraction is being performed, which not only improves the efficiency of call feature screening, but also improves the accuracy and effectiveness of feature extraction of the first model.

[0062] The following uses the application of a model training method provided by this embodiment in a call feature screening scenario as an example to further illustrate the model training method provided by this embodiment for use in a call feature screening scenario. Figure 3 ,The model training method applied to the call feature screening scenario specifically includes the following steps.

[0063] Step S302: Write the customer service call data into a feature extraction prompt template to obtain feature extraction prompt data.

[0064] Step S304: Input the feature extraction prompt data into the first model to perform feature extraction to obtain a first feature table.

[0065] Step S306 , reading the pre-configured second feature table, and inputting the feature table generation prompt data into the first model to generate the feature table, thereby obtaining a fourth feature table.

[0066] It should be noted that steps S302 to S304 and step S306 may be performed first, and then step S306, or step S306 may be performed first, and then step S302 to S304, or steps S302 to S304 and step S306 may be performed simultaneously. This embodiment does not limit the execution order of steps S302 to S304 and step S306.

[0067] Step S308: If the first call feature selected from the first feature table is obtained, the first call feature is detected using the second feature table.

[0068] Step S310: When the detection result is that the detection passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template to obtain the first parameter adjustment prompt data and input it into the first model.

[0069] Optionally, the first model performs parameter adjustment according to the first parameter adjustment prompt data.

[0070] Step S312: When the detection result is failure, the validity of the first call feature is detected.

[0071] Step S314: If the validity check passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, and the second parameter adjustment prompt data is obtained and input into the first model.

[0072] Optionally, the first model performs parameter adjustment according to the second parameter adjustment prompt data.

[0073] Step S316: If the second call feature selected from the fourth feature table is obtained, the second call feature is detected using the second feature table.

[0074] Step S318: If the detection result is passed, the customer service call data is used as sample data and the second call feature is used as the sample label to adjust the parameters of the first model.

[0075] Step S320: If the detection result is failure, the second call feature is tested for validity.

[0076] In step S322, if the validity test passes, the customer service call data is used as sample data and the second call feature is used as the sample label to adjust the parameters of the first model.

[0077] In step S324, if the third call feature selected from the second feature table is obtained, the customer service call data is used as sample data and the third call feature is used as a sample label to adjust the parameters of the first model.

[0078] It should be noted that any one of steps S302 to step S324 or any combination of multiple steps can be combined with any one of steps S202 to step S206 to form a new implementation method according to the needs of implementation deployment; in addition, according to the needs of actual deployment, any one or multiple technical features can be selected from steps S302 to step S324 and combined with any one or multiple technical features provided by steps S202 to step S206 to form a new implementation method; or, any one or multiple technical features in steps S302 to step S324 can be replaced by any one or multiple technical features provided by steps S202 to step S206 to form a new implementation method according to the needs of actual deployment, which will not be repeated here.

[0079] An embodiment of a model training device provided in this specification is as follows: In the above embodiment, a model training method is provided, and correspondingly, a model training device is also provided, which is described below with reference to the accompanying drawings.

[0080] Reference Figure 4 , which shows a schematic diagram of the structure of a model training device provided by this embodiment. Since the device embodiment corresponds to the method embodiment, the description is relatively simple. For the relevant parts, please refer to the corresponding description of the method embodiment provided above. The device embodiment described below is merely illustrative.

[0081] This embodiment provides a model training device, including: A feature extraction module 402 is configured to input customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If the first call feature selected from the first feature table is obtained, the detection module 404 is executed, and the detection module 404 is configured to detect the first call feature using the pre-configured second feature table to obtain a detection result; When the detection result is passed, the writing module 406 is run, and the writing module 406 is used to write the customer service call data and the location information of the first call feature into the parameter adjustment prompt template, obtain the first parameter adjustment prompt data and input it into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0082] In one embodiment, when the first model performs parameter adjustment according to the first parameter adjustment prompt data, the first model performs the following steps: calculating a first feedback deviation based on the preset location information included in the first parameter adjustment prompt data and the location information of the first call feature, and adjusting the parameters of the first model based on the first feedback deviation to obtain a second model; Inputting the customer service call data included in the first parameter adjustment prompt data into the second model for feature extraction to obtain a third feature table; Calculating a second feedback deviation based on the preset position information and the position information of the first call feature in the third feature table; If the second feedback deviation is greater than a deviation threshold, parameters of the second model are adjusted according to the second feedback deviation.

[0083] In one embodiment, the feature extraction module 402 performs the following steps during feature extraction: Performing feature extraction on the sub-call data of each call round contained in the customer service call data to obtain call features of each call round; Calculate the similarity between the sub-call data of each call round and the corresponding benchmark call data; The location information of each call feature is determined according to the similarity.

[0084] In one embodiment, after detecting the first call feature using the pre-configured second feature table and obtaining the detection result, the detection module 404 further performs the following steps: If the detection result is failure, performing a validity check on the first call feature; If the validity check passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, and the second parameter adjustment prompt data is obtained and input into the first model; the first model performs parameter adjustment according to the second parameter adjustment prompt data.

[0085] In one embodiment, after inputting the customer service call data into the first model for feature extraction to obtain the first feature table, the feature extraction module 402 further performs the following steps: If a second call feature filtered out from the fourth feature table is obtained, the second call feature is detected using the second feature table to obtain a detection result, and position information of the second call feature in the fourth feature table is determined based on the number of times the second call feature is filtered out; When the detection result is that the detection passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

[0086] In one embodiment, after detecting the second call feature using the second feature table and obtaining the detection result, the feature extraction module 402 further performs the following steps: If the detection result is failure, performing a validity check on the second call feature; If the validity test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

[0087] In one embodiment, after inputting the customer service call data into the first model for feature extraction to obtain the first feature table, the feature extraction module 402 further performs the following steps: If the third call feature filtered out from the second feature table is obtained, the customer service call data is used as sample data, the third call feature is used as the sample label, and the parameters of the first model are adjusted. The third call feature in the second feature table includes call features that have passed the validity test.

[0088] In the model training device provided in this embodiment, customer service call data is first input into a first model for feature extraction, obtaining a first feature table including call features for each call round and location information of each call feature. Feature extraction is performed using the first model, thereby improving the efficiency and accuracy of feature extraction and further enabling screening using the first feature table, thereby improving the efficiency of call feature screening. The first call features screened out from the first feature table are then tested using a pre-configured second feature table to obtain a test result. If the test result is a pass, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain first parameter adjustment prompt data, which is then input into the first model. The first model performs parameter adjustment based on the first parameter adjustment prompt data. In this way, the first call features screened out from the first feature table are tested using the pre-configured second feature table, thereby testing not only the screening capability but also the feature extraction capability of the first model. Furthermore, if the test result is a pass, the customer service call data and the location information of the first call feature are used to adjust the parameters of the first model, thereby improving the accuracy of the location information of each call feature in the first feature table output by the first model, thereby improving the convenience of call feature screening.

[0089] An embodiment of a model training device provided in this specification is as follows: Corresponding to the model training method described above, based on the same technical concept, an embodiment of the present application further provides a model training device, which is used to execute the model training method provided above. Figure 5 A schematic diagram of the structure of a model training device provided in an embodiment of the present application.

[0090] This embodiment provides a model training device, including: like Figure 5 As shown, the model training device may vary significantly due to different configurations or performance, and may include one or more processors 501 and memory 502. The memory 502 may store one or more applications or data. The memory 502 may be a temporary storage or a persistent storage. The application stored in the memory 502 may include one or more modules (not shown), each of which may include a series of computer-executable instructions in the model training device. Furthermore, the processor 501 may be configured to communicate with the memory 502 to execute the series of computer-executable instructions in the memory 502 on the model training device. The model training device may also include one or more power supplies 503, one or more wired or wireless network interfaces 504, one or more input / output interfaces 505, one or more keyboards 506, and the like.

[0091] In a specific embodiment, the model training device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer-executable instructions in the model training device, and the one or more programs are configured to be executed by one or more processors, including computer-executable instructions for performing the following: Inputting customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table to obtain a detection result; If the detection result is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0092] In one embodiment, when the processor adjusts the parameters according to the first parameter adjustment prompt data, the processor performs the following steps: calculating a first feedback deviation based on the preset location information included in the first parameter adjustment prompt data and the location information of the first call feature, and adjusting the parameters of the first model based on the first feedback deviation to obtain a second model; Inputting the customer service call data included in the first parameter adjustment prompt data into the second model for feature extraction to obtain a third feature table; Calculating a second feedback deviation based on the preset position information and the position information of the first call feature in the third feature table; If the second feedback deviation is greater than a deviation threshold, parameters of the second model are adjusted according to the second feedback deviation.

[0093] In one embodiment, the processor performs the following steps during feature extraction: Performing feature extraction on the sub-call data of each call round contained in the customer service call data to obtain call features of each call round; Calculate the similarity between the sub-call data of each call round and the corresponding benchmark call data; The location information of each call feature is determined according to the similarity.

[0094] In one embodiment, after detecting the first call feature using a preconfigured second feature table and obtaining a detection result, the processor further performs the following operations: If the detection result is failure, performing a validity check on the first call feature; If the validity check passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, and the second parameter adjustment prompt data is obtained and input into the first model; the first model performs parameter adjustment according to the second parameter adjustment prompt data.

[0095] In one embodiment, the processor further performs the following operations: If a second call feature filtered out from the fourth feature table is obtained, the second call feature is detected using the second feature table to obtain a detection result, and position information of the second call feature in the fourth feature table is determined based on the number of times the second call feature is filtered out; When the detection result is that the detection passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

[0096] In one embodiment, after detecting the second call feature using the second feature table and obtaining a detection result, the processor further performs the following operations: If the detection result is failure, performing a validity check on the second call feature; If the validity test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

[0097] In one embodiment, the processor further performs the following operations: If the third call feature filtered out from the second feature table is obtained, the customer service call data is used as sample data, the third call feature is used as the sample label, and the parameters of the first model are adjusted. The third call feature in the second feature table includes call features that have passed the validity test.

[0098] In the model training device provided in this embodiment, customer service call data is first input into a first model for feature extraction, obtaining a first feature table including call features for each call round and location information of each call feature. Feature extraction is performed using the first model, thereby improving the efficiency and accuracy of feature extraction and further enabling screening using the first feature table, thereby improving the efficiency of call feature screening. The first call features screened out from the first feature table are then tested using a pre-configured second feature table to obtain a test result. If the test result is a pass, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain first parameter adjustment prompt data, which is then input into the first model. The first model performs parameter adjustment based on the first parameter adjustment prompt data. In this way, the first call features screened out from the first feature table are tested using the pre-configured second feature table, thereby testing not only the screening capability but also the feature extraction capability of the first model. Furthermore, if the test result is a pass, the customer service call data and the location information of the first call feature are used to adjust the parameters of the first model, thereby improving the accuracy of the location information of each call feature in the first feature table output by the first model, thereby improving the convenience of call feature screening.

[0099] An embodiment of a computer-readable storage medium provided in this specification is as follows: Corresponding to the model training method described above, based on the same technical concept, an embodiment of the present application also provides a computer-readable storage medium.

[0100] The computer-readable storage medium provided in this embodiment is used to store computer-executable instructions. When the computer-executable instructions are executed, the following process is implemented: Inputting customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table to obtain a detection result; If the detection result is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0101] It should be noted that the embodiment of a computer-readable storage medium in this specification and the embodiment of a model training method in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0102] Another embodiment of the present disclosure further provides a computer program product, the computer program product including a computer program, which implements the following process when executed by a processor: Inputting customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table to obtain a detection result; If the detection result is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

[0103] In the computer program product provided in this embodiment, customer service call data is first input into a first model for feature extraction, resulting in a first feature table including call features for each call round and location information of each call feature. Feature extraction is performed using the first model, thereby improving the efficiency and accuracy of feature extraction and enabling screening using the first feature table, thereby improving the efficiency of call feature screening. A preconfigured second feature table is then used to test the first call features screened out from the first feature table to obtain a test result. If the test result is a pass, the customer service call data and the location information of the first call feature are written into a parameter adjustment prompt template to obtain first parameter adjustment prompt data, which is then input into the first model. The first model then performs parameter adjustment based on the first parameter adjustment prompt data. In this way, the first call features screened out from the first feature table are tested using the preconfigured second feature table, thereby testing not only the screening capability but also the feature extraction capability of the first model. Furthermore, if the test result is a pass, the customer service call data and the location information of the first call feature are used to adjust the parameters of the first model, thereby improving the accuracy of the location information of each call feature in the first feature table output by the first model, thereby improving the convenience of call feature screening.

[0104] The computer program product in the embodiment of the present disclosure can implement each process of the above-mentioned model training method embodiment and achieve the same effects and functions, which will not be repeated here.

[0105] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0106] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the embodiments of the present application may take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware. Furthermore, this specification may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable test equipment to produce a machine, so that the instructions executed by the processor of the computer or other programmable test equipment generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable test equipment to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device that implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable test device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process described in the flow. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0110] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0111] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0112] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology for information storage. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0113] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0114] The embodiments of the present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0115] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0116] The foregoing description is merely an example of the present invention and is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of the claims herein.

Claims

1. A model training method, characterized in that: The method comprises: Inputting customer service call data into a first model for feature extraction to obtain a first feature table, wherein the first feature table includes call features of each call round and location information of each call feature; If a first call feature selected from the first feature table is obtained, the first call feature is detected using a pre-configured second feature table to obtain a detection result; If the detection result is passed, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, the first parameter adjustment prompt data is obtained and input into the first model; the first model adjusts the parameters according to the first parameter adjustment prompt data.

2. The method according to claim 1, characterized in that The performing parameter adjustment according to the first parameter adjustment prompt data includes: calculating a first feedback deviation based on the preset location information included in the first parameter adjustment prompt data and the location information of the first call feature, and adjusting the parameters of the first model based on the first feedback deviation to obtain a second model; Inputting the customer service call data included in the first parameter adjustment prompt data into the second model for feature extraction to obtain a third feature table; Calculating a second feedback deviation based on the preset position information and the position information of the first call feature in the third feature table; If the second feedback deviation is greater than a deviation threshold, parameters of the second model are adjusted according to the second feedback deviation.

3. The method according to claim 1, characterized in that The feature extraction includes: Performing feature extraction on the sub-call data of each call round contained in the customer service call data to obtain call features of each call round; Calculate the similarity between the sub-call data of each call round and the corresponding benchmark call data; The location information of each call feature is determined according to the similarity.

4. The method according to claim 1, wherein After the operation of detecting the first call feature using the pre-configured second feature table to obtain a detection result is performed, the method further includes: If the detection result is failure, performing a validity check on the first call feature; If the validity check passes, the customer service call data and the location information of the first call feature are written into the parameter adjustment prompt template, and the second parameter adjustment prompt data is obtained and input into the first model; the first model performs parameter adjustment according to the second parameter adjustment prompt data.

5. The method according to claim 1, wherein The method further comprises: If a second call feature filtered out from the fourth feature table is obtained, the second call feature is detected using the second feature table to obtain a detection result, and position information of the second call feature in the fourth feature table is determined based on the number of times the second call feature is filtered out; When the detection result is that the detection passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

6. The method according to claim 5, characterized in that After the operation of detecting the second call feature using the second feature table to obtain a detection result is performed, the method further includes: If the detection result is failure, performing a validity check on the second call feature; If the validity test passes, the customer service call data is used as sample data and the second call feature is used as a sample label to adjust the parameters of the first model.

7. The method according to claim 1, characterized in that The method further comprises: If the third call feature filtered out from the second feature table is obtained, the customer service call data is used as sample data, the third call feature is used as the sample label, and the parameters of the first model are adjusted. The third call feature in the second feature table includes call features that have passed the validity test.

8. A model training device, characterized in that: The device comprises: A processor; and a memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the model training method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store computer-executable instructions, which, when executed by a processor, implement the model training method according to any one of claims 1 to 7.

10. A computer program product, characterized in that It includes a computer program, which, when executed by a processor, implements the model training method as described in any one of claims 1 to 7.

Images