Data generation method and device, storage medium and computer program product

Abstract

The embodiment of the invention provides a data generation method and device, a storage medium and a computer program product. In the embodiment of the invention, operations such as semantic extraction, clustering and semantic induction can be performed on the question and answer data contained in the real question and answer record samples, so that all question and answer dimensions involved in each real question and answer record sample are mined, the question and answer dimensions involved in each question and answer record sample are flexibly combined, and the question and answer accuracy of each question and answer record sample is improved. And deriving an incremental question and answer dimension combination for each question and answer record sample, and associating the incremental question and answer dimension combination with the corresponding question and answer record sample to generate an incremental training sample corresponding to the question and answer record sample. According to the method, the high-quality data set can be constructed based on the incremental training sample, so that incremental training of the target large model can be realized based on the high-quality data set, the question and answer effect of the target large model is improved, and the interaction experience of a customer in an intelligent question and answer process can be further improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a data generation method, device, storage medium and computer program product. Background Art

[0002] With the rapid development of artificial intelligence technology, intelligent question-answering systems have been widely used in various industries as an important means for modern enterprises to communicate with customers. The large models used in traditional intelligent question-answering systems often rely on limited predefined dialogue trees, which are difficult to cope with complex and changing customer needs and interaction scenarios. The large models need to be trained so that they can adapt to specific customer needs or interaction scenarios.

[0003] Currently, real interaction data is usually used as a data set to train large models. The amount of real interaction data is limited and has certain limitations. It is difficult to generate a representative high-quality data set for the large model, which makes it impossible for the large model to learn effective question-and-answer knowledge, resulting in poor results in intelligent question-and-answering using the large model and poor customer interaction experience. Summary of the invention

[0004] Multiple aspects of the present application provide a data generation method, device, storage medium and computer program product to improve the customer's interactive experience during the intelligent question and answer process.

[0005] The present application provides a data generation method, including:

[0006] Obtain at least one question and answer record sample, each question and answer record sample containing at least one question and answer data;

[0007] Performing semantic feature extraction operations on the acquired question and answer data respectively to obtain semantic features corresponding to each question and answer data;

[0008] Based on the semantic features, clustering the acquired question and answer data to generate at least one cluster;

[0009] Under the at least one cluster, semantic induction is performed respectively to obtain the question-answering dimension corresponding to each of the at least one cluster;

[0010] For any question and answer record sample, the question and answer dimension corresponding to the cluster in which each question and answer data contained therein is located is determined as the question and answer dimension associated with the question and answer record sample;

[0011] According to a preset dimension combination rule, the question and answer dimensions associated with the question and answer record sample are combined to derive an incremental question and answer dimension combination from the question and answer record sample;

[0012] Associating the obtained incremental question-answer dimension combinations with the question-answer record samples respectively to generate incremental training samples corresponding to the question-answer record samples;

[0013] Generate incremental training samples for other question-answer record samples;

[0014] The generated incremental training samples are input into a target large model to train the target large model, and the target large model is used for intelligent question answering.

[0015] The embodiment of the present application also provides a computing platform, including: a memory, a processor, and a communication component;

[0016] The memory is used to store one or more computer instructions;

[0017] The processor is coupled to the memory and the communication component, and is configured to execute the one or more computer instructions to perform the aforementioned data generation method.

[0018] An embodiment of the present application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed by one or more processors, the one or more processors are caused to execute the aforementioned data generation method.

[0019] An embodiment of the present application also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the processor is caused to execute the aforementioned data generation method.

[0020] In an embodiment of the present application, a semantic feature extraction operation can be performed on the question and answer data contained in each of the multiple question and answer record samples, and based on the extracted semantic features, the question and answer data can be clustered into at least one cluster. Semantic induction is performed on the semantic features corresponding to the question and answer data contained in each cluster to obtain the question and answer dimensions corresponding to each cluster. For any question and answer record sample, the question and answer dimension corresponding to the cluster in which each question and answer data contained therein is located is determined as the question and answer dimension associated with the question and answer record sample. According to the preset dimension combination rules, the question and answer dimensions associated with the question and answer record sample are combined to derive an incremental question and answer dimension combination for the question and answer record sample, and the incremental question and answer dimension combination is associated with the corresponding question and answer record sample to generate an incremental training sample corresponding to the question and answer record sample, and the incremental training sample is input into the target large model to train the target large model for intelligent question and answer.

[0021] It can be seen that by preprocessing, clustering and normalizing real question and answer record samples, all question and answer dimensions involved in the question and answer record samples are first mined out, and then the question and answer dimensions mined out under each question and answer record sample are flexibly combined to obtain at least one question and answer dimension combination associated with the question and answer record sample. Afterwards, based on the question and answer record sample and the question and answer dimension combination associated with the question and answer record sample, incremental training samples can be derived, and a high-quality data set can be constructed based on the incremental training samples, so that incremental training of the target large model can be achieved based on the high-quality data set, thereby improving the question and answer effect of the target large model, thereby further improving the customer's interactive experience in the intelligent question and answer process. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0023] Figure 1 A system architecture diagram of a data processing system provided for an exemplary embodiment of the present application;

[0024] Figure 2 A flowchart of a data generation method provided for an exemplary embodiment of the present application;

[0025] Figure 3 A logical schematic diagram of generating incremental training samples in an exemplary application scenario provided by another exemplary embodiment of the present application;

[0026] Figure 4 A schematic diagram of the structure of a computing device is provided for yet another exemplary embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0028] At present, large models are usually used to conduct intelligent question and answer for customers. The data set used for training large models has great limitations, which leads to the technical problem that the effect of using large models for intelligent question and answer is poor and the customer interaction experience is poor. The embodiment of the present application provides a solution. The basic idea is: extract the semantic features corresponding to each question and answer data in the real question and answer record sample, first cluster the question and answer data based on the semantic features, obtain at least one cluster cluster and the semantic feature group corresponding to each cluster cluster, and then semantically summarize the semantic features in each semantic feature group to obtain the question and answer dimension corresponding to each cluster cluster. For any question and answer record sample, the question and answer dimension corresponding to the cluster cluster where each question and answer data contained in it is located is determined as the question and answer dimension associated with the question and answer record sample, so as to mine all the question and answer dimensions involved in the question and answer record sample. The question and answer dimensions mined under each question and answer record sample are flexibly combined to derive an incremental question and answer dimension combination for each question and answer record sample, and the incremental question and answer dimension combination is associated with the corresponding question and answer record sample to generate an incremental training sample corresponding to the question and answer record sample. A high-quality data set can be constructed based on incremental training samples, which enables incremental training of the target large model, thereby improving the question-answering effect of the target large model and further improving the customer's interactive experience during the intelligent question-answering process.

[0029] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0030] Figure 1 A system architecture diagram of a data processing system provided by an exemplary embodiment of the present application. Figure 1 As shown, the system architecture may include: a client 101 and a server 102, the client 101 and the server 102 are connected in communication, and the connection method between the client 101 and the server 102 may be a wired or wireless communication link or an optical fiber cable, etc. For example, the server may be connected to the client through a mobile network, and the network standard of the mobile network may be any one of 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), 5G, WiMax, etc. The client may also establish a communication connection with the server through Bluetooth, WiFi, infrared, etc.

[0031] The data processing system provided in this embodiment can be applied to various scenarios that require the use of large models for intelligent question and answer, for example, scenarios that use large models for intelligent outbound calls, scenarios that use large models to implement intelligent customer service, etc. Of course, these are only exemplary, and this embodiment does not limit the application scenarios.

[0032] Among them, the client 101 can be a browser, an APP (Application), a web application such as H5 (HyperText Markup Language 5, Hypertext Markup Language Version 5) application, a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The server 102 may include servers that provide various services, such as a server for background training that provides support for the model used on the client 101, a server that processes the interactive information sent by the client, etc. In physical implementation, the client 101 can be a mobile phone, a tablet computer, a personal computer, a desktop computer, a smart speaker, a smart watch, etc. The server 102 can be a single server, a distributed server cluster composed of multiple servers, a cloud server, a cloud host, a virtual center and other server devices. The composition of the server device mainly includes a processor, a hard disk, a memory, a system bus, etc., which is similar to a general computer architecture.

[0033] It should be noted that the data generation method provided in the embodiment of the present application is generally executed by the server 102, and the corresponding data processing device is generally arranged in the server 102. However, in other embodiments of the present application, the client 101 may also have similar functions to the server 102, so as to execute the data generation method provided in the embodiment of the present application. In other embodiments, the video generation method provided in the embodiment of the present application may also be jointly executed by the client 101 and the server 102.

[0034] In practical applications, the data processing system can have any number of clients and servers according to actual needs. Figure 1 Only one client and one server are shown exemplarily, but this should not cause loss of the protection scope of this application.

[0035] Figure 2 A flowchart of a data generation method provided by an exemplary embodiment of the present application. The method can be executed by a data processing device, which can be implemented as a combination of software and / or hardware, and the data processing device can be integrated in a computing device. Figure 2 , the method comprising:

[0036] Step 200, obtaining at least one question and answer record sample, each question and answer record sample containing at least one question and answer data;

[0037] Step 201, performing semantic feature extraction operations on the acquired question and answer data to obtain semantic features corresponding to each question and answer data;

[0038] Step 202: clustering the acquired question and answer data based on semantic features to generate at least one cluster;

[0039] Step 203, performing semantic induction under at least one cluster to obtain a question-answering dimension corresponding to each of the at least one cluster;

[0040] Step 204: for any question and answer record sample, determine the cluster where each question and answer data contained therein is located as the question and answer dimension associated with the question and answer record sample;

[0041] Step 205: combining the question and answer dimensions associated with the question and answer record sample according to the preset dimension combination, so as to derive an incremental question and answer dimension combination for the question and answer record sample;

[0042] Step 206, associating each incremental question-and-answer dimension obtained with the question-and-answer record sample to generate an incremental training sample corresponding to the question-and-answer record sample;

[0043] Step 207, generating incremental training samples for other question and answer record samples;

[0044] Step 208, input the generated incremental training samples into the target large model to train the target large model, and the target large model is used for intelligent question answering.

[0045] The data generation method provided in this embodiment can be applied to various scenarios that require the use of large models for intelligent question and answer. The scenarios of intelligent question and answer include but are not limited to intelligent outbound calls, intelligent customer service, intelligent operation guidance, etc. For example, incremental training samples can be constructed for the large model that initiates intelligent outbound calls to train the large model using the incremental training samples; incremental training samples can also be constructed for the large model integrated with intelligent customer service to train the large model using the incremental training samples. Of course, this embodiment is not limited to this.

[0046] Before going into detail, several technical concepts involved in this embodiment are briefly explained.

[0047] Big Model: In the field of machine learning and artificial intelligence, a big model refers to a deep learning model with a large parameter scale and complexity. These models are usually built with deep neural networks and have tens of trillions or even hundreds of billions of parameters, aiming to improve the model's expressiveness and predictive performance to handle more complex tasks and data.

[0048] Intelligent Q&A: Intelligent Q&A is a question-and-answer system based on artificial intelligence technology. Through natural language processing technology, it can quickly and accurately ask questions to customers or answer questions raised by customers.

[0049] Smart outbound calling: Smart outbound calling refers to the optimization and automation of outbound calls based on artificial intelligence technology to improve the efficiency and quality of outbound calls. Smart outbound calling can perform intelligent dialing, voice recognition, voice synthesis, automated question and answer operations based on customer needs and behaviors, thereby achieving more efficient and accurate customer communication.

[0050] During the research, the inventors found that large models used for intelligent question-answering usually use real interactive data as data sets. The number of samples in the data sets is limited and has certain limitations, which makes it impossible to effectively train the large models, resulting in poor results for intelligent question-answering using large models. To this end, the inventors proposed a new data set construction mechanism to use the newly constructed incremental data set to train the large model, so that the large model can learn more question-answering logic, thereby effectively improving the question-answering effect.

[0051] The following is an explanation of the technical implementation logic of this new mechanism introduced in this embodiment.

[0052] refer to Figure 2 In step 200, at least one question and answer record sample can be obtained. The question and answer record sample is the historical conversation data generated in the process of intelligent question and answer using a large model. A question and answer record sample represents a group of historical conversation data. Each group of historical conversation data can contain at least one question and answer. A single question or a single answer can be regarded as a question and answer data.

[0053] When obtaining question and answer record samples, question and answer record samples can be obtained from online and offline channels respectively. From the online question and answer record table for intelligent question and answer, extract the first question and answer record generated online. The online question and answer record table records the historical conversation data generated by intelligent question and answer through the large model. From the offline question and answer record table for manual question and answer, extract the second question and answer record generated offline. The offline question and answer record table records the historical conversation data generated by the real staff in the process of communicating with customers through telephone, face-to-face conversation, etc. The first question and answer record and the second question and answer record are used as at least one obtained question and answer record sample. In this way, the question and answer records obtained online and the question and answer records obtained offline are both used as question and answer record samples, which can reduce the deviation caused by a single channel, effectively ensure the richness of the samples, and enable the samples to more comprehensively and accurately reflect the real question and answer situation, thereby improving the accuracy and reliability of the incremental training samples generated for the question and answer record samples.

[0054] When extracting question and answer record samples from the online question and answer record table and / or the offline question and answer record table, at least one question and answer record belonging to the target application scenario can be extracted from the online question and answer record table and / or the offline question and answer record table according to the application scenarios to which each question and answer record belongs, as the question and answer record sample required for generating the incremental training sample. Alternatively, question and answer record samples can also be extracted from the online question and answer record table and / or the offline question and answer record table according to the data entry time, data size, etc., which is not limited in this embodiment.

[0055] Of course, all the question and answer records recorded in the online question and answer record table and the offline question and answer record table can also be directly used as question and answer record samples. In this case, the collected first question and answer record and the second question and answer record can be data cleaned. The first question and answer record or the second question and answer record can be empty. Through data cleaning, abnormal and repeated question and answer records or question and answer data can be screened out, and the content in the question and answer data can also be standardized, such as unifying the date format, converting the values ​​of different units into a unified unit, etc. For each question and answer record obtained after data cleaning, determine the application scenario to which it belongs, and each question and answer record obtained after data cleaning includes the first question and answer record and / or the second question and answer record. From the question and answer records obtained after data cleaning, screen out the question and answer records belonging to the target application scenario as at least one question and answer record sample required to generate incremental training samples.

[0056] It should be noted that the question and answer record samples recorded in the online question and answer record table and / or the offline question and answer record table include but are not limited to text form and voice form. If the question and answer record sample is in voice form, the question and answer data in the question and answer record sample may be converted from voice form to text form before performing the semantic feature extraction operation on the question and answer data in the question and answer record sample.

[0057] In this embodiment, question and answer record samples belonging to the target application scenario are screened out from the question and answer record samples, and incremental training samples are generated for a single application scenario, and then the target large model can be trained for each single application scenario. Through training for a specific application scenario, the large model can learn the specific knowledge and model in the scenario, so that it can show stronger generalization ability when processing similar intelligent question and answer tasks. This ability enables the large model to better adapt to different input data and improve the accuracy and efficiency of intelligent question and answer.

[0058] In step 201, after obtaining the question and answer record sample, a semantic extraction operation can be performed on the question and answer data contained in the question and answer record sample to obtain the semantic features corresponding to each question and answer data. Optionally, the obtained question and answer data can be input into a semantic feature extraction model to extract the semantic features corresponding to each question and answer data using the semantic feature extraction model. Among them, the semantic feature extraction model is a model that automatically extracts and represents semantic information from text data based on natural language processing technology. The semantic feature extraction model can adopt a word embedding model, a BERT (Bidirectional Encoder Representations from Transformers) model, a RoBERTa (Robustly Optimized BERT Approach) model, etc., which is not limited in this embodiment.

[0059] Of course, other methods can also be used to extract semantic features from question and answer data. For example, the question and answer data can be segmented to obtain multiple segmented words contained in the question and answer data, and then each segmented word can be tagged with a part of speech to determine the meaning and grammatical structure of the segmented word, and then the semantics of the question and answer data can be analyzed by combining the meanings and grammatical structures of multiple segmented words. The method for extracting semantic features is not limited in this embodiment.

[0060] In step 202, the acquired question and answer data may be clustered based on semantic features to generate at least one cluster. Here, the question and answer data contained in at least one question and answer record sample is mixed, and then all the mixed question and answer data are clustered, and the boundaries of the question and answer record samples do not need to be considered during clustering.

[0061] In the above or below embodiments, multiple implementations may be used to cluster the question and answer data.

[0062] In an optional implementation, the similarity between the semantic features corresponding to each of the two question and answer data can be calculated for any two question and answer data. When the similarity is greater than a preset threshold, the two question and answer data are divided into the same cluster. For example, the semantic features can be represented in the form of a tuple, and the tuple contains text data representing the semantic features. The text data contained in the tuple corresponding to each of the two question and answer data can be obtained, and the similarity between the text data corresponding to each of the two question and answer data is calculated. If the calculated similarity is greater than the preset threshold, the two question and answer data are divided into the same cluster.

[0063] In another optional implementation, the semantic features corresponding to each question and answer data can be converted into word vectors respectively to obtain the word vectors corresponding to each question and answer data. For any two question and answer data, the vector distance between the word vectors corresponding to the two question and answer data is calculated. If the calculated vector distance is less than the preset distance, the two question and answer data are divided into the same cluster.

[0064] Perform a clustering judgment for any two question and answer data until each question and answer data is clustered into the corresponding cluster and at least one cluster is obtained. By clustering the question and answer data, question and answer data with similar topics or contents can be classified into the same category, thereby revealing the potential question and answer intentions and communication characteristics in the question and answer data, which helps to better understand the question and answer dimensions of the question and answer data.

[0065] Based on this, semantic induction can be performed under at least one cluster to obtain the question-answer dimension corresponding to at least one cluster. Semantic induction refers to the process of summarizing, generalizing, and refining the information, meaning, or concept in the semantic features. That is, semantic induction is performed on the semantic features corresponding to at least one question-answer data contained in the cluster, and the induction result is used as the question-answer dimension corresponding to the cluster.

[0066] Exemplarily, multiple preset question-and-answer dimensions and keywords associated with each question-and-answer dimension can be obtained. The question-and-answer dimension can be used to describe the question-and-answer intention represented by the question-and-answer data. The question-and-answer dimension can be understood as the question-and-answer scenario in which the question-and-answer process is located. The keywords associated with the question-and-answer dimension refer to words related to the question-and-answer intention, and different question-and-answer dimensions are associated with different keywords. For example, the question-and-answer dimension can be: the customer feels it is too expensive, the customer is not interested, the customer wants to continue to understand the product functions, etc. For the question-and-answer dimension of "the customer feels it is too expensive", the keywords associated with it include but are not limited to "expensive", "discount", "discount", etc. For any cluster, extract the common semantic features from the semantic features corresponding to each question-and-answer data contained in the cluster. If the similarity between the extracted common semantic features and the target keyword is greater than the preset dimension threshold, the question dimension associated with the target keyword is used as the question-and-answer dimension corresponding to the cluster. Among them, the target keyword is any one of the keywords associated with each question-and-answer dimension.

[0067] When extracting common semantic features from the semantic features corresponding to each question and answer data contained in the cluster, the intersection can be taken between the semantic features corresponding to any two question and answer data, so that the semantic features in the intersection are taken as common semantic features. Alternatively, the semantic similarity between the semantic features corresponding to any two question and answer data is calculated, and the semantic features with a semantic similarity greater than a specified threshold are taken as common semantic features.

[0068] It can be understood that, for any cluster, one or more common semantic features can be determined for the cluster.

[0069] When the number of common semantic features is one, the similarity between the common semantic feature and the target keyword can be calculated. If the calculated similarity between the common semantic feature and the target keyword is greater than a preset dimension threshold, the question dimension associated with the target keyword is used as the question-answer dimension corresponding to the cluster. The target keyword is any one of the keywords associated with each question-answer dimension.

[0070] In this case, if the similarity between a common semantic feature and multiple keywords is greater than a preset threshold, it is determined whether the question-answer dimensions associated with each of the multiple keywords are the same question-answer dimension. If multiple keywords are associated with the same question-answer dimension, the single question-answer dimension associated with the multiple keywords can be used as the question-answer dimension corresponding to the cluster; if multiple keywords are not associated with the same question-answer dimension, the multiple question-answer dimensions associated with the multiple keywords can be used as the question-answer dimensions corresponding to the cluster, that is, a cluster can be associated with one or more question-answer dimensions.

[0071] In the case of multiple public semantic features, the similarity between each public semantic feature and the target keyword can be calculated respectively. If the similarity between each public semantic feature and the target keyword is greater than the preset dimension threshold, the question dimension associated with the target keyword is used as the question-answer dimension corresponding to the cluster. The target keyword is any one of the keywords associated with each question-answer dimension.

[0072] In this case, if there is only one keyword, and the similarity between each public semantic feature and the keyword is greater than the preset dimension threshold, the question-answer dimension associated with the keyword is used as the question-answer dimension corresponding to the cluster. If there are multiple keywords, and the similarity between each public semantic feature and the keyword is greater than the preset dimension threshold, the question-answer dimensions associated with each of the multiple keywords are used as the question-answer dimensions corresponding to the cluster, that is, a cluster can be associated with one or more question-answer dimensions.

[0073] In this embodiment, by analyzing the question-answer dimensions involved in each question-answer data, all the question-answer dimensions involved in the question-answer record sample can be mined, so that more comprehensive incremental training samples can be derived for the question-answer record sample based on these question-answer dimensions. In addition, based on the adaptability between the question-answer dimensions and the question-answer record samples, the effectiveness and accuracy of the incremental training samples derived from the question-answer record samples based on the question-answer dimensions can be ensured, thereby achieving effective training of the target large model, so that the target large model can achieve accurate question-answering for the target customers and improve the customer's question-answering experience.

[0074] Continue to refer Figure 2 In step 205, the question and answer dimensions associated with the question and answer record samples may be combined according to the preset dimension combination rules to derive an incremental question and answer dimension combination for the question and answer record samples. When the question and answer dimensions associated with the question and answer record samples are combined, only the question and answer dimensions associated with the question and answer record sample can be combined under each question and answer record sample, and the question and answer dimensions associated with multiple question and answer record samples cannot be mixed and combined.

[0075] In the above or following embodiments, a variety of implementation methods may be used to derive incremental question and answer dimension combinations for question and answer record samples.

[0076] In an optional implementation, the correlation relationship between the question and answer dimensions associated with the question and answer record sample can be obtained, and the correlation relationship can be but is not limited to a causal relationship, a dependency relationship, etc. For any question and answer dimension, the related question and answer dimension that has any correlation relationship with the question and answer dimension can be obtained from the question and answer dimension associated with the question and answer record sample, and the question and answer dimension and the related question and answer dimension can be combined to derive an incremental question and answer dimension combination for the question and answer record sample. For example, the question and answer record sample is associated with four question and answer dimensions A, B, C, and D. If question and answer dimension A has a dependency relationship with question and answer dimension B, question and answer dimension A has a dependency relationship with question and answer dimension C, and question and answer dimension A does not have a correlation relationship with question and answer dimension D, then question and answer dimensions A, B, and C can be selected from the question and answer dimensions associated with the question and answer record sample for combination to obtain an incremental question and answer dimension combination A+B+C derived for the question and answer record sample.

[0077] In another optional implementation, the number of selections N indicated in the dimension combination rule, where N is a positive integer, can be used. N question and answer dimensions are randomly selected from the question and answer dimensions associated with the question and answer record sample, and the selected N question and answer dimensions are combined to derive an incremental question and answer dimension combination for the question and answer record sample. Among them, each random selection operation obtains a question and answer dimension combination. For example, the question and answer record sample is associated with 4 question and answer dimensions A, B, C, and D, and the number of selections indicated in the dimension combination rule is 3. Then, 3 question and answer dimensions can be randomly selected from the question and answer dimensions associated with the question and answer record sample for combination, and the incremental question and answer dimension combinations derived from the question and answer record sample include A+B+C, A+B+D, A+C+D, and B+C+D.

[0078] Of course, it is also possible not to specify the number of selections N, but to use all possible permutations and combinations of each question and answer dimension associated with the question and answer record sample as the question and answer dimension combination that derives the increment for the question and answer record sample. Still taking the question and answer record sample associated with 4 question and answer dimensions A, B, C, and D as an example, all possible permutations and combinations of these 4 question and answer dimensions include 15 combinations: A, B, C, D, A+B, A+C, A+D, B+C, B+D, C+D, A+B+C, A+B+D, A+C+D, B+C+D, and A+B+C+D. These 15 combinations can all be used as the question and answer dimension combination that derives the increment for the question and answer record sample.

[0079] In step 206, the obtained incremental question-answer dimension combinations can be associated with the question-answer record samples to generate incremental training samples corresponding to the question-answer record samples. That is, the incremental training samples include the incremental question-answer dimension combinations and the question-answer record samples. Afterwards, the same method is used to generate incremental training samples for other question-answer record samples until corresponding incremental training samples are generated for all question-answer record samples.

[0080] In this embodiment, based on all the question and answer dimensions involved in the sample records, a question and answer dimension combination is constructed for the question and answer record sample, which is essentially to reasonably derive other question and answer scenarios for the question and answer record sample, and to generate incremental training samples for the question and answer record sample based on the derived question and answer scenarios and the corresponding question and answer record samples. The incremental training samples cover more question and answer scenarios, so that the target large model can learn the expressions of words in more question and answer scenarios, improve the learning quality of the target large model, and thus effectively improve the accuracy of questions and answers and improve the customer's question and answer experience.

[0081] On this basis, the generated incremental training samples can be input into the target large model to train the target large model, which is a large model used for intelligent question and answering.

[0082] To further ensure the validity of the training samples, the incremental training samples may be first sent to the review unit for the review unit to review the incremental training samples, receive the reviewed incremental training samples sent by the review unit, and input the reviewed incremental training samples into the target large model to train the target large model. The review unit may be used to screen out more effective and high-quality incremental training samples from all the incremental training samples, so that a high-quality data set may be constructed for the target large model based on the reviewed incremental training samples. For example, incremental training samples with low adaptability between the incremental question-and-answer dimension combination and the question-and-answer record sample may be deleted. Among them, screening rules may be preset in the review unit so that after receiving the incremental training samples, the received incremental training samples may be automatically screened according to the preset screening rules. The review unit may also be controlled by a technician so that after receiving the incremental training samples, the technician manually screens the incremental training samples. This embodiment does not limit this.

[0083] Based on this, after completing the training of the target big model, the target question and answer dimensions that are adapted to the target customers can be obtained during the intelligent question and answer process, and the trained target big model can be used to generate question and answer data that are adapted to the target question and answer dimensions, and the generated question and answer data that are adapted to the target question and answer dimensions can be sent to the target customers.

[0084] Continue to refer Figure 1 The target customer may be the client 101, and the server 102 may initiate an intelligent outbound call to the target customer according to the preset outbound call rules. The preset outbound call rules may include a target question and answer dimension, and the trained target large model may be called according to the preset outbound call rules to generate outbound call questions adapted to the target question and answer dimension, and the generated outbound call questions are sent to the target customer.

[0085] Among them, the forms of initiating intelligent outbound calls to target customers include but are not limited to text and voice. If the intelligent outbound call is in voice form, after using the target large model to generate outbound call questions that are adapted to the target question and answer dimensions, the outbound call questions can be converted from text to voice, and the converted voice can be used to interact with the target customer. Among them, the implementation methods used to convert the text form into voice form include but are not limited to text-to-speech (TTS) technology, speech synthesis software, text-to-speech plug-ins, etc.

[0086] Accordingly, in this embodiment, operations such as semantic extraction, clustering, and semantic induction can be performed on the question and answer data contained in the real question and answer record samples to mine all the question and answer dimensions involved in each real question and answer record sample, and the question and answer dimensions involved in each question and answer record sample are flexibly combined to derive an incremental question and answer dimension combination for each question and answer record sample, and the incremental question and answer dimension combination and the corresponding question and answer record sample are associated to generate an incremental training sample corresponding to the question and answer record sample. A high-quality data set can be constructed based on the incremental training samples, so that incremental training of the target large model can be achieved based on the high-quality data set, which improves the question and answer effect of the target large model, thereby further improving the customer's interactive experience in the intelligent question and answer process.

[0087] Figure 3 A logical diagram of generating incremental training samples in an exemplary application scenario provided by another exemplary embodiment of the present application. Figure 3 As shown in Figure 2, the process of generating incremental training samples can be divided into the following four steps:

[0088] Step 1: Data preprocessing

[0089] 1. Obtain real historical conversation data from the online question and answer record table and the offline question and answer record table respectively, and use the obtained real historical conversation data as question and answer record samples. Each question and answer record sample contains at least one question and answer data, and the question and answer data is a single question or a single answer in the historical conversation data.

[0090] 2. Perform data cleaning on the question and answer record samples to filter out meaningless question and answer record samples and question and answer data. For example, if a question and answer data only contains background sound, the question and answer data can be deleted; if a question and answer record sample uses an obscure dialect and it is difficult to understand the meaning of each question and answer data in the question and answer record sample, the sample record can be deleted.

[0091] 3. Filter the question and answer record samples obtained after data cleaning. The question and answer record samples can be screened according to the application scenario, and the question and answer record samples belonging to the target application scenario are screened out as the question and answer record samples required to construct the incremental training samples.

[0092] 4. For the question and answer record samples belonging to the target application scenario, perform semantic feature extraction operations on the question and answer data contained in the question and answer record samples to obtain the semantic features corresponding to each question and answer data. Map the extracted semantic features to a high-dimensional space to convert the semantic features into vector representations, and obtain the vectors corresponding to each question and answer data.

[0093] Step 2: Clustering and Normalization

[0094] 1. Based on the vectors corresponding to each question and answer data, calculate the vector distance between any two question and answer data. If the calculated vector distance is less than the preset distance, the two question and answer data are divided into the same cluster. Calculate the vector distance for any two question and answer data to find a suitable cluster for the two question and answer data until each question and answer data is clustered into the corresponding cluster and at least one cluster is obtained.

[0095] 2. Obtain multiple preset question-and-answer dimensions and keywords associated with each question-and-answer dimension, where the target keyword is any one of the keywords associated with each question-and-answer dimension. Under at least one cluster, extract common semantic features from the semantic features corresponding to each question-and-answer data contained in the cluster, so as to achieve semantic induction (normalization) of the semantic features corresponding to each question-and-answer data contained in the cluster based on the common semantic features. Calculate the similarity between the common semantic features and the target keyword. If the calculated similarity between the common semantic features and the target keyword is greater than the preset dimension threshold, the question dimension associated with the target keyword is used as the question-and-answer dimension corresponding to the cluster.

[0096] Step 3: Generate question-answer dimension combinations and construct incremental training samples

[0097] 1. For any question and answer record sample, the question and answer dimension corresponding to the cluster where each question and answer data is located is determined as the question and answer dimension associated with the question and answer record sample. The question and answer dimensions associated with the question and answer record sample are randomly combined to derive an incremental question and answer dimension combination for the question and answer record sample.

[0098] 2. Associate each incremental question-answer dimension combination obtained with the question-answer record sample to generate an incremental training sample corresponding to the question-answer record sample. Accordingly, the same method is used to generate incremental training samples for other question-answer record samples until corresponding incremental training samples are generated for all question-answer record samples.

[0099] Step 4: Review the generated incremental training samples

[0100] 1. Send the incremental training samples to the review unit for review, and select more effective and high-quality incremental training samples from all the incremental training samples. Receive the reviewed incremental training samples sent by the review unit, which include the remaining high-quality incremental training samples after screening.

[0101] On this basis, after the incremental training samples are generated, the generated incremental training samples can be input into the target large model to train the target large model for intelligent question and answer. After the training of the target large model is completed, the target question and answer dimension adapted to the target customer can be obtained in the intelligent question and answer process, and the trained target large model can be used to generate question and answer data adapted to the target question and answer dimension, and the generated question and answer data adapted to the target question and answer dimension can be sent to the target customer.

[0102] In this embodiment, the data set generation method is improved, and the corresponding incremental training samples are obtained by preprocessing, clustering and normalizing the real question and answer record samples. The target large model is incremented and trained through the incremental training samples, so that the target large model can learn the question and answer knowledge in different application scenarios, thereby improving the accuracy and fluency of the question and answer of the target large model in the corresponding scenario.

[0103] It should be noted that the execution subject of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 201 to 203 can be device A; for another example, the execution subject of steps 201 and 202 can be device A, and the execution subject of step 203 can be device B; and so on.

[0104] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel, and the sequence numbers of the operations, such as 201, 202, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0105] Figure 4 A schematic diagram of a computing device provided as another exemplary embodiment of the present application is shown in FIG. Figure 4 As shown, the computing device includes: a memory 40 , a processor 41 and a communication component 42 .

[0106] The processor 41 is coupled to the memory 40 and is used to execute the computer program in the memory 40 to:

[0107] Acquire at least one question and answer record sample through the communication component 42, each question and answer record sample including at least one question and answer data;

[0108] Performing semantic feature extraction operations on the acquired question and answer data respectively to obtain semantic features corresponding to each question and answer data;

[0109] Based on the semantic features, clustering the acquired question and answer data to generate at least one cluster;

[0110] Under at least one cluster, semantic induction is performed respectively to obtain the question-answering dimension corresponding to at least one cluster;

[0111] For any question and answer record sample, the question and answer dimension corresponding to the cluster where each question and answer data is located is determined as the question and answer dimension associated with the question and answer record sample;

[0112] According to the preset dimension combination rules, the question and answer dimensions associated with the question and answer record sample are combined to derive an incremental question and answer dimension combination from the question and answer record sample;

[0113] Associating each incremental question-answer dimension combination obtained with the question-answer record sample respectively to generate incremental training samples corresponding to the question-answer record sample;

[0114] Generate incremental training samples for other question-answer record samples;

[0115] The generated incremental training samples are input into the target large model to train the target large model, and the target large model is used for intelligent question answering.

[0116] In an optional embodiment, in the process of clustering the acquired question and answer data based on semantic features to generate at least one cluster, the processor 41 is further configured to:

[0117] For any two question-answer data, calculate the similarity between the semantic features corresponding to the two question-answer data;

[0118] When the similarity is greater than a preset threshold, the two question-answer data are divided into the same cluster.

[0119] In an optional embodiment, in the process of performing semantic induction under at least one cluster to obtain the question-answer dimension corresponding to each of the at least one cluster, the processor 41 is further configured to:

[0120] Obtain multiple preset question-and-answer dimensions and keywords associated with each question-and-answer dimension;

[0121] For any cluster, extract the common semantic features from the semantic features corresponding to each question and answer data contained in the cluster;

[0122] If the similarity between the common semantic feature and the target keyword is greater than the preset dimension threshold, the question dimension associated with the target keyword is used as the question-answer dimension corresponding to the cluster;

[0123] Among them, the target keyword is any one of the keywords associated with each question and answer dimension.

[0124] In an optional embodiment, in the process of combining the question and answer dimensions associated with the question and answer record sample according to a preset dimension combination rule to derive an incremental question and answer dimension combination for the question and answer record sample, the processor 41 is further configured to:

[0125] Obtaining correlation relationships between question and answer dimensions associated with the question and answer record sample, where the correlation relationships include causal relationships and / or dependency relationships;

[0126] For any question and answer dimension, from the question and answer dimensions associated with the question and answer record sample, obtain related question and answer dimensions that have any association relationship with the question and answer dimension;

[0127] Combine the question and answer dimensions with the related question and answer dimensions to derive incremental question and answer dimension combinations for the question and answer record samples.

[0128] In an optional embodiment, in the process of combining the question and answer dimensions associated with the question and answer record sample according to a preset dimension combination rule to derive an incremental question and answer dimension combination for the question and answer record sample, the processor 41 is further configured to:

[0129] According to the selection number N indicated in the dimension combination rule, N question and answer dimensions are randomly selected from the question and answer dimensions associated with the question and answer record sample, where N is a positive integer;

[0130] The selected N question-answer dimensions are combined to derive an incremental question-answer dimension combination from the question-answer record sample;

[0131] Among them, each random selection operation obtains a question-answer dimension combination.

[0132] In an optional embodiment, during the process of acquiring at least one question and answer record sample, the processor 41 is further configured to:

[0133] Extracting the first question-and-answer record generated online from the online question-and-answer record for the intelligent question-and-answer session;

[0134] Extracting a second offline question-and-answer record from an offline question-and-answer record for manual question-and-answer;

[0135] The first question and answer record and the second question and answer record are taken as at least one obtained question and answer record sample.

[0136] In an optional embodiment, the processor 41 is further configured to:

[0137] Perform data cleaning on the collected first question-and-answer records and the second question-and-answer records;

[0138] Determine the application scenarios for each question and answer record obtained after data cleaning;

[0139] From the question and answer records obtained after data cleaning, the question and answer records belonging to the target application scenario are screened out as at least one question and answer record sample obtained.

[0140] In an optional embodiment, in the process of performing semantic feature extraction operations on the acquired question and answer data to obtain semantic features corresponding to each question and answer data, the processor 41 is further configured to:

[0141] Input the acquired question and answer data into the semantic feature extraction model respectively;

[0142] The semantic feature extraction model is used to extract the semantic features corresponding to each question and answer data.

[0143] In an optional embodiment, during the process of inputting the generated incremental training samples into the target large model to train the target large model, the processor 41 is further configured to:

[0144] Sending the incremental training samples to the review unit so that the review unit reviews the incremental training samples;

[0145] Receive the post-review incremental training samples sent by the review unit;

[0146] The reviewed incremental training samples are input into the target large model to train the target large model.

[0147] In an optional embodiment, the processor 41 is further configured to:

[0148] In the process of intelligent question-answering, obtain the target question-answering dimensions that are adapted to the target customers;

[0149] Use the trained target large model to generate question and answer data corresponding to the target question and answer dimension;

[0150] Send the question and answer data corresponding to the target question and answer dimension to the target customers.

[0151] Further, if Figure 4 As shown, the computing platform also includes: a display 43, a power component 44, an audio component 45 and other components. Figure 4 Only some components are shown schematically, which does not mean that the computing platform only includes Figure 4 Components shown.

[0152] It is worth noting that the technical details in the above-mentioned embodiments of the computing device can refer to the relevant description of the actions of the computing device in the above-mentioned embodiments of the data generation method. In order to save space, they will not be repeated here, but this should not cause a loss in the scope of protection of this application.

[0153] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps in the above-mentioned data generation method can be implemented. It should be noted that the technical solution of the computer program and the technical solution of the above-mentioned data generation method belong to the same concept, and the details not described in detail in the technical solution of the computer program can be referred to the description of the technical solution of the above-mentioned data generation method.

[0154] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be executed by a computing device in the above method embodiment.

[0155] Above Figure 4 The memory in the computer is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, contact data, phone book data, messages, pictures, videos, etc. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0156] Above Figure 4 The communication component in is configured to facilitate wired or wireless communication between the device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0157] Above Figure 4The display in the embodiment includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

[0158] Above Figure 4 The power supply component in the device provides power to various components of the device where the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply component is located.

[0159] Above Figure 4 The audio component in can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal.

[0160] 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 present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0161] The present application 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 the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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 data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device 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.

[0162] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device 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, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0164] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0165] The 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. The memory is an example of a computer-readable medium.

[0166] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 memory (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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape 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 temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0167] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity 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, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0168] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A data generation method, characterized in that: include: Obtain at least one question and answer record sample, each question and answer record sample containing at least one question and answer data; Performing semantic feature extraction operations on the acquired question and answer data respectively to obtain semantic features corresponding to each question and answer data; Based on the semantic features, clustering the acquired question and answer data to generate at least one cluster; Under the at least one cluster, semantic induction is performed respectively to obtain the question-answering dimension corresponding to each of the at least one cluster; For any question and answer record sample, the question and answer dimension corresponding to the cluster in which each question and answer data contained therein is located is determined as the question and answer dimension associated with the question and answer record sample; According to a preset dimension combination rule, the question and answer dimensions associated with the question and answer record sample are combined to derive an incremental question and answer dimension combination from the question and answer record sample; Associating the obtained incremental question-answer dimension combinations with the question-answer record samples respectively to generate incremental training samples corresponding to the question-answer record samples; Generate incremental training samples for other question-answer record samples; The generated incremental training samples are input into a target large model to train the target large model, and the target large model is used for intelligent question answering.

2. The method according to claim 1, characterized in that The step of clustering the acquired question and answer data based on the semantic features to generate at least one cluster includes: For any two pieces of question-answer data, calculate the similarity between the semantic features corresponding to the two pieces of question-answer data; When the similarity is greater than a preset threshold, the two question-answer data are divided into the same cluster.

3. The method according to claim 1, characterized in that The performing semantic induction under the at least one cluster to obtain the question-answering dimension corresponding to each of the at least one cluster includes: Obtain multiple preset question-and-answer dimensions and keywords associated with each question-and-answer dimension; For any cluster, extracting common semantic features from the semantic features corresponding to each question and answer data contained in the cluster; If the similarity between the common semantic feature and the target keyword is greater than a preset dimension threshold, the question dimension associated with the target keyword is used as the question-answer dimension corresponding to the cluster; Among them, the target keyword is any one of the keywords associated with each question and answer dimension.

4. The method according to claim 1, characterized in that: The step of combining the question and answer dimensions associated with the question and answer record sample according to a preset dimension combination rule to derive an incremental question and answer dimension combination for the question and answer record sample includes: Obtaining association relationships between question and answer dimensions associated with the question and answer record sample, wherein the association relationships include causal relationships and / or dependency relationships; For any question and answer dimension, from the question and answer dimensions associated with the question and answer record sample, obtain related question and answer dimensions that have any association relationship with the question and answer dimension; The question and answer dimension and the related question and answer dimension are combined to derive an incremental question and answer dimension combination for the question and answer record sample.

5. The method according to claim 1, characterized in that The step of combining the question and answer dimensions associated with the question and answer record sample according to a preset dimension combination rule to derive an incremental question and answer dimension combination for the question and answer record sample includes: According to the selection number N indicated in the dimension combination rule, N question and answer dimensions are randomly selected from the question and answer dimensions associated with the question and answer record sample, where N is a positive integer; Combining the selected N question-answer dimensions to derive an incremental question-answer dimension combination for the question-answer record sample; Among them, each random selection operation obtains a question-answer dimension combination.

6. The method according to claim 1, characterized in that The obtaining of at least one question and answer record sample includes: Extracting the first question and answer record generated online from the online question and answer record table for intelligent question and answer; Extracting the second question and answer record generated offline from the offline question and answer record table for manual question and answer; The first question and answer record and the second question and answer record are taken as at least one acquired question and answer record sample.

7. The method according to claim 6, characterized in that Also includes: Performing data cleaning on the collected first question-and-answer record and the second question-and-answer record; Determine the application scenarios for each question and answer record obtained after data cleaning; From the question and answer records obtained after data cleaning, the question and answer records belonging to the target application scenario are screened out as at least one question and answer record sample obtained.

8. The method according to claim 1, characterized in that The semantic feature extraction operation is performed on the acquired question and answer data to obtain the semantic features corresponding to each question and answer data, including: Input the acquired question and answer data into the semantic feature extraction model respectively; The semantic feature extraction model is used to extract the semantic features corresponding to each question and answer data.

9. The method according to claim 1, characterized in that: The step of inputting the generated incremental training samples into the target large model to train the target large model includes: Sending the incremental training sample to a review unit so that the review unit reviews the incremental training sample; Receiving the post-review incremental training samples sent by the review unit; The audited incremental training samples are input into the target large model to train the target large model.

10. The method according to claim 1 or 9, characterized in that: Also includes: In the process of intelligent question-answering, obtain the target question-answering dimensions that are adapted to the target customers; Using the trained target large model, generating question-answering data adapted to the target question-answering dimension; The question and answer data adapted to the target question and answer dimension is sent to the target customer.

11. A computing device, characterized in that: Includes memory, processor, and communication components; The memory is used to store one or more computer instructions; The processor is coupled to the memory and the communication component, and is configured to execute the one or more computer instructions to execute the data generation method according to any one of claims 1 to 10.

12. A computer-readable storage medium storing computer instructions, characterized in that: When the computer instructions are executed by one or more processors, the one or more processors are caused to execute the data generating method according to any one of claims 1 to 10.

13. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the processor is caused to execute the data generating method according to any one of claims 1 to 10.

Images