Data processing method, man-machine interaction method, intelligent agent system and device

Abstract

The invention provides a data processing method, a man-machine interaction method, an agent system and a device, and relates to the technical field of computers, in particular to the technical field of artificial intelligence such as deep learning, natural language processing and large models. According to the specific scheme, firstly, log data of a service system using a target model are obtained, the log data are analyzed, query data and retrieval data corresponding to the query data are obtained, then the query data and the retrieval data are input into the target model and a first model respectively, at least two pieces of generated data are obtained, and the generated data are stored in the first model. The method comprises the following steps: firstly, generating data, then utilizing at least two second models, determining the first credibility of each piece of generated data based on query data and retrieval data, and finally, associating and storing the query data, the retrieval data, the generated data and the first credibility into a training database associated with a target model.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, in particular to the field of artificial intelligence technology such as deep learning, natural language processing, and large models, and specifically to a data processing method, a human-computer interaction method, an intelligent agent system and device. Background Art

[0002] In the prior art, in the retrieval question-answering agent system, the training set used for the early model training usually comes from a public data set or actively mined domain data. However, the training set obtained by this method lacks the fit to the actual business scenario data, has low accuracy, and low data quality. Summary of the invention

[0003] The present disclosure aims to solve one of the technical problems in the related art at least to some extent.

[0004] The first embodiment of the present disclosure provides a data processing method, including:

[0005] Get log data of business systems that use the target model;

[0006] Parsing the log data to obtain query data and search data corresponding to the query data;

[0007] Inputting the query data and the search data into the target model and the first model respectively to obtain at least two generated data;

[0008] Determining a first credibility of each of the generated data based on the query data and the retrieved data by using at least two second models;

[0009] The query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

[0010] The second aspect of the present disclosure provides a model updating method, including:

[0011] Determine a target update mode based on at least one of the amount of new data contained in the training database, the type of the new data, and the current performance parameter of the target model, wherein the new data in the training database is generated by the method according to the first aspect of the present disclosure;

[0012] Based on the target update method, obtaining training data from the training database;

[0013] The target model is updated and trained according to the training data.

[0014] The third aspect of the present disclosure provides a human-computer interaction method, including:

[0015] Receive input sentences;

[0016] Identify the input sentence and determine the target business associated with the input;

[0017] The input sentence is input into a target model associated with the target business to obtain a reply sentence corresponding to the input sentence, wherein the target model is generated by the method described in the second aspect of the present disclosure.

[0018] The fourth aspect of the present disclosure provides a human-computer interaction intelligent agent system, including:

[0019] Input component, used to obtain input sentences;

[0020] a processing component, configured to call a first agent to recognize the input sentence, determine a target business associated with the input sentence, and call a second agent associated with the target business to obtain a reply sentence output by the second agent, wherein at least one of the first agent and the second agent is configured to call the model generated by the method described in the second aspect of the present disclosure to complete a corresponding function;

[0021] An output component is used to output the reply statement.

[0022] A fifth aspect of the present disclosure provides a data processing device, including:

[0023] A first acquisition module is used to acquire log data of a business system using a target model;

[0024] A second acquisition module is used to parse the log data to obtain query data and search data corresponding to the query data;

[0025] A processing module, used for inputting the query data and the search data into the target model and the first model respectively, to obtain at least two generated data;

[0026] A first determination module, configured to determine a first credibility of each of the generated data based on the query data and the retrieved data by using at least two second models;

[0027] A storage module is used to associate and store the query data, the retrieval data, the generated data and the first credibility in a training database associated with the target model.

[0028] A sixth aspect of the present disclosure provides a model updating device, including:

[0029] A second determination module is used to determine a target update mode based on at least one of the amount of new data contained in the training database, the type of the new data, and the current performance parameter of the target model, wherein the new data in the training database is generated by the method according to the first aspect of the present disclosure;

[0030] A third acquisition module, used for acquiring training data from the training database based on the target update mode;

[0031] A training module is used to update and train the target model according to the training data.

[0032] The seventh aspect embodiment of the present disclosure proposes a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the data processing method proposed in the first aspect embodiment of the present disclosure, the model updating method proposed in the second aspect embodiment of the present disclosure, or the human-computer interaction method proposed in the third aspect embodiment of the present disclosure.

[0033] The eighth aspect embodiment of the present disclosure proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the data processing method proposed in the first aspect embodiment of the present disclosure, the model updating method proposed in the second aspect embodiment of the present disclosure, or the human-computer interaction method proposed in the third aspect embodiment of the present disclosure.

[0034] A ninth aspect embodiment of the present disclosure proposes a computer program product, including a computer program, which, when executed by a processor, implements the data processing method proposed in the first aspect embodiment of the present disclosure, the model updating method proposed in the second aspect embodiment of the present disclosure, or the human-computer interaction method proposed in the third aspect embodiment of the present disclosure.

[0035] The data processing method, human-computer interaction method, intelligent agent system and device provided by the present disclosure have the following beneficial effects:

[0036] In the disclosed embodiment, the log data of the business system using the target model is first obtained, and the log data is parsed to obtain query data and retrieval data corresponding to the query data, and then the query data and the retrieval data are respectively input into the target model and the first model to obtain at least two generated data, and then at least two second models are used to determine the first credibility of each generated data based on the query data and the retrieval data, and finally the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model. Thus, query data and retrieval data are obtained based on the log data of the business system using the target system, and generated data corresponding to the query data and the retrieval data are generated based on the target model and the first model, and the generated data is evaluated using the second model and credibility is obtained, and the query data, the retrieval data, the generated data and credibility are associated and stored in a database corresponding to the target model, thereby ensuring that the training data in the training database associated with the target model fits the real business scenario data, and improving the data quality and accuracy.

[0037] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0039] Figure 1 A flowchart of a data processing method provided by an embodiment of the present disclosure;

[0040] Figure 2 A flowchart of a data processing method provided by an embodiment of the present disclosure;

[0041] Figure 3 A flowchart of a data processing method provided by an embodiment of the present disclosure;

[0042] Figure 4 A flowchart of a data processing method provided by an embodiment of the present disclosure;

[0043] Figure 5 A flowchart of a model updating method provided by an embodiment of the present disclosure;

[0044] Figure 6 A flowchart of a model updating method provided by an embodiment of the present disclosure;

[0045] Figure 7 A schematic diagram of the effect of the updated target model provided by the embodiment of the present disclosure;

[0046] Figure 8 A flowchart of a human-computer interaction method provided by an embodiment of the present disclosure;

[0047] Fig. 9 A schematic diagram of the structure of a human-computer interaction intelligent agent system provided by an embodiment of the present disclosure;

[0048] Fig.10 A schematic diagram of the structure of a data processing device provided in an embodiment of the present disclosure;

[0049] Fig.11 A schematic diagram of the structure of a model updating device provided in an embodiment of the present disclosure;

[0050] Fig.12 A schematic diagram of the structure of a human-computer interaction device provided in an embodiment of the present disclosure;

[0051] Fig.13 A block diagram of an exemplary computer device suitable for implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0052] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0053] The present disclosure relates to artificial intelligence technology fields such as deep learning, natural language processing, and large models.

[0054] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0055] Deep Learning (DL) is the process of learning the inherent laws and representation levels of sample data. The information obtained in the learning process is very helpful for interpreting data such as text, images, and sounds. The ultimate goal of deep learning is to enable machines to have analytical learning capabilities like humans and to recognize data such as text, images, and sounds.

[0056] Natural Language Processing (NLP) is an interdisciplinary subject in the fields of computer science, artificial intelligence and linguistics. It mainly studies how to enable computers to understand, process, generate and simulate human language, so as to achieve the ability to have natural conversations with humans.

[0057] The large model can also be called the Foundation Model. The model extracts knowledge through billions of corpora or images, learns and then produces a large model with billions of parameters.

[0058] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0059] The following describes the data processing method, human-computer interaction method, intelligent agent system and device of the embodiments of the present disclosure with reference to the accompanying drawings.

[0060] Figure 1 A flowchart of a data processing method provided by an embodiment of the present disclosure.

[0061] like Figure 1 As shown, the data processing method may include the following steps:

[0062] Step 101: Obtain log data of a business system using a target model.

[0063] The target model is a model that needs to be updated and trained, and it can be any large model, that is, a large language model (LMM). For example, the target model can be a supervised fine-tuning (SFT) model, or a reward model, or a preference model, or a retrieval augmented generation (RAG) model, etc., which is not limited in the present disclosure.

[0064] It should be noted that the business system using the target model can be a business system of any field scenario and demand. For example, the business system using the target model can be a knowledge question-and-answer system in an office scenario, etc., which is not limited in the present disclosure.

[0065] It should be noted that the log data is the real data record of the users who use the business system. Different business systems may obtain different log data, which is not limited in this disclosure.

[0066] In the present disclosure, by acquiring log data of a business system that uses a target model, conditions are provided for enhancing the target model's ability to fit real scenarios.

[0067] Step 102, parse the log data to obtain query data and search data corresponding to the query data.

[0068] The retrieval data may be knowledge data related to the query data.

[0069] It should be noted that different query data may correspond to different retrieval data, and this disclosure does not limit this.

[0070] In the present disclosure, after obtaining the log data of the business system using the target model, the log data can be parsed to obtain query data and retrieval data corresponding to the query data, thereby ensuring that business needs can be reflected in the data processing process and enhancing the data processing process's ability to fit real business scenario data.

[0071] It should be noted that when parsing the log data to obtain the query data and the retrieval data corresponding to the query data, the query data and the retrieval data corresponding to the query data can be obtained through a preset module or interface. For example, the log data can be mined through the Log Miner Application Programming Interface (Log Miner API), the Question Sampler API, and the Retriever Sampler API to obtain the query data and the retrieval data corresponding to the query data, wherein the Log Miner API can mine the query data and the retrieval data from the log data, the Question Sampler API can generate a variety of query data to cover different scenarios and requirements and enhance the versatility of the model, and the Retriever Sampler API can generate the retrieval data corresponding to the query data, and the present disclosure does not limit this.

[0072] Step 103, input the query data and the search data into the target model and the first model respectively, to obtain at least two generated data.

[0073] The first model may be any generative model, which is an LLM model, or may be pre-set. For example, the first model may be a Generative Pre-trained Transformer (GPT) model, a General Language Model (GLM), etc., which is not limited in the present disclosure.

[0074] It should be noted that the number of the first models can be any number, which can be preset, and the present disclosure does not limit this.

[0075] The generated data may be query data and answer data corresponding to the retrieved data, and may be any type of data. For example, the generated data may be text data, or image data, etc., which is not limited in the present disclosure.

[0076] In the present disclosure, after obtaining the query data and the retrieval data, the query data and the retrieval data can be input into the target model and the first model respectively to obtain at least two generated data corresponding to the query data and the retrieval data. Since the generated data output by the target model and the first model respectively may be different, the diversity of the generated data obtained is ensured.

[0077] It should be noted that, when there are multiple first models, multiple first models of different scales can be selected to further ensure the diversity of generated data, and the present disclosure does not limit this.

[0078] Step 104 : Determine the first credibility of each generated data by using at least two second models and based on the query data and the retrieved data.

[0079] The second model may be any model used to evaluate the generated data and determine its corresponding first credibility, which is an LLM model, which may be pre-set, and the present disclosure does not limit this.

[0080] It should be noted that, when there are multiple second models and multiple first models, the second model may be partially identical to the first model, or the second model may be completely different from the first model, and the present disclosure does not limit this.

[0081] The first credibility may be the credibility of the generated data. The higher the first credibility, the higher the accuracy of the corresponding generated data, which is not limited in the present disclosure.

[0082] It should be noted that, if the second model is different, the first credibility of the same generated data may be different, and the present disclosure does not limit this.

[0083] In the present disclosure, after obtaining at least two generated data corresponding to the query data and the retrieval data, at least two second models can be used to determine the first credibility of each generated data based on the query data and the retrieval data, thereby ensuring the comprehensiveness and fairness of the evaluation of the generated data and improving the accuracy and quality of the generated data.

[0084] Step 105, the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

[0085] In the present disclosure, after determining the first credibility corresponding to each generated data, the query data, retrieval data, generated data and the first credibility can be associated and stored in a training database associated with the target model, thereby providing a data basis for updating the training of the target model.

[0086] In the disclosed embodiment, the log data of the business system using the target model is first obtained, and the log data is parsed to obtain query data and retrieval data corresponding to the query data, and then the query data and the retrieval data are respectively input into the target model and the first model to obtain at least two generated data, and then at least two second models are used to determine the first credibility of each generated data based on the query data and the retrieval data, and finally the query data, the retrieval data, the generated data and the first credibility are associated and stored in the training database associated with the target model. Thus, query data and retrieval data are obtained based on the log data of the business system using the target system, and query data and generated data corresponding to the retrieval data are generated based on the target model and the first model, and the generated data is evaluated using the second model and credibility is obtained, and the query data, the retrieval data, the generated data and credibility are associated and stored in the database corresponding to the target model, thereby ensuring that the training data in the training database associated with the target model fits the real business scenario data, and improving the data quality and accuracy.

[0087] Figure 2 A flowchart of a data processing method provided by an embodiment of the present disclosure.

[0088] like Figure 2 As shown, the data processing method may include the following steps:

[0089] Step 201: Obtain log data of a business system using a target model.

[0090] The specific implementation form of step 201 can refer to the detailed description in other embodiments of the present disclosure, and will not be described in detail here.

[0091] Step 202: parse the log data to obtain the query data contained in the log data and the first search segment corresponding to the query data.

[0092] It should be noted that the character length of the first search segment is consistent with the input character length of the target model.

[0093] It should be noted that the number of first search fragments corresponding to the obtained query data may be any number, and the present disclosure does not limit this.

[0094] In the present disclosure, after obtaining the log data of the business system using the target model, the log data can be parsed to obtain the query data contained in the log data and the first retrieval fragment corresponding to the query data, thereby ensuring that the obtained retrieval data fits the real business scenario data and improving the accuracy of the data.

[0095] Step 203: Search the preset database based on the query data to obtain a second search segment.

[0096] It should be noted that the preset database can be any public search database, and the present disclosure does not limit this.

[0097] It should be noted that the character length of the second search segment is consistent with the input character length of the target model.

[0098] It should be noted that the number of the second search fragments may be the same as the number of the first search fragments, or may be different from the number of the first search fragments, and the present disclosure does not limit this.

[0099] In the present disclosure, after obtaining the query data and the first search fragment corresponding to the query data, in order to ensure the diversity and reliability of the search data corresponding to the query data, the preset database can be searched based on the query data through the search sampler interface to obtain the second search fragment, which is not limited in the present disclosure.

[0100] It should be noted that when obtaining the second search fragment through the retrieval sampler interface, the interface can be used to obtain diversified second search fragments through different databases and retrieval settings (such as increasing the character length of the retrieval fragment, increasing the number of retrieval fragments, etc.), thereby improving the authenticity and coverage of the data in the data processing process. The present disclosure does not limit this.

[0101] Step 204: determine the first search segment and the second search segment as search data.

[0102] In the present disclosure, after obtaining the first search segment and the second search segment corresponding to the query data, the first search segment and the second search segment can be determined as the search data corresponding to the query data, thereby increasing the diversity of the search data corresponding to the query data.

[0103] Step 205: input the query data and the search data into the target model and the first model respectively to obtain at least two generated data.

[0104] Step 206 , using at least two second models and based on the query data and the retrieved data, determine a first credibility of each generated data.

[0105] Step 207, the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

[0106] The specific implementation forms of steps 205 to 207 may refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0107] In the disclosed embodiment, log data of a business system using a target model is first obtained, and the log data is parsed to obtain query data contained in the log data and a first retrieval fragment corresponding to the query data. Then, a preset database is searched based on the query data to obtain a second retrieval fragment, and the first retrieval fragment and the second retrieval fragment are determined as retrieval data. Then, the query data and the retrieval data are respectively input into the target model and the first model to obtain at least two generated data. Using at least two second models and based on the query data and the retrieval data, a first credibility of each generated data is determined. Finally, the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model. Therefore, after obtaining the log data of the business system of the target model, the query data and the first retrieval fragment are obtained by parsing the log data, and then the query data is used to search the database to obtain the second retrieval fragment. The first retrieval fragment and the second retrieval fragment are determined as retrieval data corresponding to the query data, and based on the target model and the first model, the query data and the generated data corresponding to the retrieval data are generated. Based on the second model, the generated data is evaluated and the credibility is obtained. The query data, retrieval data, generated data and their corresponding credibility are associated and stored in a training database, thereby improving the diversity and quality of data in the data processing method.

[0108] Figure 3 A flowchart of a data processing method provided by an embodiment of the present disclosure.

[0109] like Figure 3 As shown, the data processing method may include the following steps:

[0110] Step 301: Obtain log data of a business system using a target model.

[0111] Step 302: parse the log data to obtain query data and search data corresponding to the query data.

[0112] The specific implementation forms of step 301 to step 302 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0113] Step 303: Generate first prompt information based on the query data, the search data and the first prompt template corresponding to the target model.

[0114] It should be noted that the first prompt template corresponding to the target model may be pre-set. Different target models may have different corresponding first prompt templates, which is not limited in the present disclosure.

[0115] It should be noted that, if the query data, the retrieval data and the first prompt template are different, the generated first prompt information may be different, and the present disclosure does not limit this.

[0116] In the present disclosure, after obtaining the query data and the retrieval data corresponding to the query data, in order to obtain the generated data output by the target model and improve the accuracy of the generated data output by the target model, the first prompt information corresponding to the query data and the retrieval data can be first generated based on the first prompt template corresponding to the target model.

[0117] Step 304: input the first prompt information into the target model to obtain output generated data.

[0118] In the present disclosure, after obtaining the first prompt information corresponding to the query data and the search data, the generated data output by the target model is obtained by inputting the first prompt information into the target model, thereby improving the accuracy of the generated data output by the target model.

[0119] Step 305: Obtain the first model and the second prompt template corresponding to the first model from the large model library.

[0120] It should be noted that the large model library may include any large model, which may be pre-set. For example, the large model library may include an SFT model, a reward model, a preference model, a RAG model, a GPT model, a GLM model, etc., which is not limited in the present disclosure.

[0121] It should be noted that the second prompt template corresponding to the first model may be pre-set. Different first models may correspond to different second prompt templates, which is not limited in the present disclosure.

[0122] In the present disclosure, after obtaining the generated data output by the target model, in order to improve the diversity of the generated data corresponding to the obtained query data and retrieval data, a first model and a second prompt template corresponding to the first model can be obtained from a large model library, thereby providing conditions for improving the diversity of the generated data.

[0123] Step 306: Generate second prompt information based on the second prompt template, the query data and the search data.

[0124] It should be noted that, for the second prompt template, the query data and the retrieval data are different, and the generated second prompt information may be different, and the present disclosure does not limit this.

[0125] In the present disclosure, after obtaining a second prompt template corresponding to at least one first model, in order to obtain the generated data output by the first model and improve the accuracy of the generated data output by the first model, second prompt information corresponding to the query data and the retrieval data can be first generated based on the second prompt template.

[0126] Step 307: input the second prompt information into the first model to obtain output generated data.

[0127] In the present disclosure, after obtaining the second prompt information, the second prompt information can be input into the corresponding first model to obtain the generated data output by the first model, thereby improving the diversity of the generated data corresponding to the query data and the retrieval data.

[0128] It should be noted that the process of obtaining generated data based on the target model and the process of obtaining generated data based on the first model can be executed in parallel, or the process of obtaining generated data based on the first model can be executed first, and then the process of obtaining generated data based on the target model can be executed, or the process of obtaining generated data based on the target model can be executed first, and then the process of obtaining generated data based on the first model can be executed. The present disclosure does not limit this.

[0129] Step 308 : Determine a first credibility of each generated data by using at least two second models and based on the query data and the retrieved data.

[0130] Step 309, the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

[0131] The specific implementation forms of step 308 to step 309 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0132] In the disclosed embodiment, log data of a business system using a target model is first obtained, and the log data is parsed to obtain query data and retrieval data corresponding to the query data; then, based on the query data, the retrieval data and a first prompt template corresponding to the target model, a first prompt message is generated, and the first prompt message is input into the target model to obtain output generated data; then, a first model and a second prompt template corresponding to the first model are obtained from a large model library; based on the second prompt template, query data and retrieval data, second prompt message is generated, and the second prompt message is input into the first model to obtain output generated data; finally, using at least two second models and based on the query data and the corresponding retrieval data, a first credibility of each generated data is determined; and the query data, retrieval data, generated data and the first credibility are associated and stored in a training database associated with the target model. Therefore, after obtaining the query data and retrieval data in the log data, the prompt information of the target model corresponding to the query data and the retrieval data is generated through the prompt template corresponding to the target model, and the generated data output by the target model is obtained based on the prompt information. The prompt information of the first model corresponding to the query data and the retrieval data is generated through the prompt template of the first model obtained in the large model library, and the generated data output by the first model is obtained based on the prompt information. At least two second models are used to evaluate all the generated data and obtain credibility, and the query data, retrieval data, generated data and their corresponding credibility are associated and stored in the training database of the target model, thereby improving the diversity of the generated data corresponding to the query data and the retrieval data.

[0133] Figure 4 A flowchart of a data processing method provided by an embodiment of the present disclosure.

[0134] like Figure 4 As shown, the data processing method may include the following steps:

[0135] Step 401: Obtain log data of a business system using a target model.

[0136] Step 402, parse the log data to obtain query data and search data corresponding to the query data.

[0137] Step 403: Generate first prompt information based on the query data, the search data and the first prompt template corresponding to the target model.

[0138] Step 404: input the first prompt information into the target model to obtain output generated data.

[0139] Step 405: Obtain the first model and the second prompt template corresponding to the first model from the large model library.

[0140] Step 406: Generate second prompt information based on the second prompt template, the query data and the search data.

[0141] Step 407: input the second prompt information into the first model to obtain output generated data.

[0142] The specific implementation forms of steps 401 to 407 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0143] Step 408: Obtain at least two second models from the large model library, and a third prompt template corresponding to each second model, wherein at least one second model is different from the first model.

[0144] It should be noted that the third prompt template corresponding to each second model may be preset. Different second models may have different corresponding third prompt templates, which is not limited in the present disclosure.

[0145] In the present disclosure, after obtaining the generated data output by the first model, before using the second model to evaluate the generated data corresponding to the query data and the retrieval data, at least two second models and a third prompt template corresponding to each second model can be first obtained from a large model library, wherein at least one second model is different from the first model, which can improve the reliability and fairness of the evaluation of the generated data.

[0146] Step 409: Generate third prompt information based on the query data, the retrieval data, the generated data and each third prompt template.

[0147] It should be noted that, if the query data, the corresponding search data, the generated data and the third prompt template are different, the generated third prompt information may be different, and the present disclosure does not limit this.

[0148] In the present disclosure, after obtaining at least two second models and a third prompt template corresponding to each second model, in order to enable the second model to determine the first credibility corresponding to the generated data, the third prompt information can be first generated based on the query data, the retrieved data, the generated data and each third prompt template.

[0149] Step 410: input the third prompt information into the corresponding second model to obtain the second credibility corresponding to the generated data.

[0150] The second credibility may be a reference credibility of the generated data.

[0151] It should be noted that, if the second model is different, the second credibility corresponding to the output generated data may be different, and the present disclosure does not limit this.

[0152] In the present disclosure, after obtaining the third prompt information corresponding to each third prompt template, each third prompt information can be input into the corresponding second model to obtain the second credibility corresponding to the generated data, thereby providing conditions for improving the accuracy and reliability of the first credibility of each generated data determined.

[0153] Step 411, determining the first credibility according to the second credibility corresponding to each generated data.

[0154] It should be noted that, if the second model is different, the second credibility corresponding to the same generated data output may be different, and the present disclosure does not limit this.

[0155] In the present disclosure, after obtaining the second credibility corresponding to each generated data output by each second model, the first credibility corresponding to the generated data can be determined based on the second credibility corresponding to each generated data by utilizing a preset evaluation model (which can be an evaluation model in the second model, or can also be other evaluation models). Thus, through multi-stage evaluation of the generated data, while ensuring the accuracy and reliability of the determined first credibility, inconsistent and low-quality generated data are filtered and reduced, providing conditions for improving the quality of generated data. At the same time, the cost of manual data annotation is reduced and data processing efficiency is improved. The present disclosure does not limit this.

[0156] Optionally, when determining the first credibility corresponding to the generated data according to the second credibility corresponding to each generated data, the first credibility can be determined according to the ratio between the number of the second credibility corresponding to the generated data that is greater than the credibility threshold and the total number of the second credibility, thereby improving the accuracy of the first credibility corresponding to the generated data. For example, when the second credibility corresponding to the generated data is greater than the credibility threshold, it can be determined that the higher the first credibility corresponding to the generated data, when the ratio of the number of the second credibility corresponding to the generated data that is greater than the credibility threshold to the number of the second credibility is greater than 0.5, it can be determined that the first credibility of the generated data is higher, otherwise the first credibility of the generated data is lower, and the present disclosure does not limit this.

[0157] The credibility threshold may be a credibility critical value used to determine the first credibility, which may be preset and is not limited in the present disclosure.

[0158] Optionally, when determining the first credibility corresponding to each generated data based on the second credibility corresponding to the generated data, it is also possible to first determine the score of each evaluation dimension in the second credibility, and then determine the first credibility based on the evaluation rules associated with each evaluation dimension and the score of the generated data in each evaluation dimension, thereby improving the reliability and accuracy of the determined first credibility of the generated data.

[0159] The evaluation dimension may be at least one of the following: the accuracy of the retrieval data associated with the generated data, the format of the generated data, the semantics of the generated data, and the logic of the execution process of the generated data, which is not limited in the present disclosure.

[0160] It should be noted that the evaluation dimensions are different, and the evaluation rules associated with each evaluation dimension may be different. For example, when the evaluation dimension is the accuracy dimension of the retrieval data associated with the generated data, the associated evaluation rule may be to check the accuracy of the retrieval data. When the accuracy of the retrieval data is lower, the score of the evaluation dimension may be lower; when the evaluation dimension is the format dimension of the generated data, the associated evaluation rule may be to check whether the format of the generated data meets the expected standards; when the evaluation dimension is the semantic dimension of the generated data, the associated evaluation rule may be to check whether the logic of the semantics of the generated data is consistent, and the accuracy of its context; when the evaluation dimension is the logical dimension of the execution process of the generated data, the associated evaluation rule may be to check whether the execution process of the generated data conforms to the front-end display logic (such as thinking first, then summarizing and other sequential logic), and the present disclosure does not limit this.

[0161] It should be noted that when determining the first credibility corresponding to each generated data based on the second credibility corresponding to the generated data, the preset feature selection interface of the second credibility can be initialized manually based on a small amount of evaluation, and then the noise data, that is, the erroneous second credibility, can be filtered out through a preset sorting algorithm (such as Elo sorting), so as to further improve the reliability of determining the first credibility of the generated data. The present disclosure does not limit this.

[0162] Step 412, the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

[0163] The specific implementation form of step 412 can refer to the detailed description in other embodiments of the present disclosure, and will not be described in detail here.

[0164] In the disclosed embodiment, log data of a business system using a target model is first obtained, the log data is parsed to obtain query data and retrieval data corresponding to the query data, and first prompt information is generated based on the query data, the retrieval data and a first prompt template corresponding to the target model, and the first prompt information is input into the target model to obtain output generated data, and then a first model and a second prompt template corresponding to the first model are obtained from a large model library, and second prompt information is generated based on the second prompt template, the query data and the retrieval data, and the second prompt information is input into the first model to obtain output generated data, and then at least two second models and a third prompt template corresponding to each second model are obtained from the large model library, and third prompt information is generated based on the query data, the retrieval data, the generated data and each third prompt template, and the third prompt information is input into the corresponding second model to obtain a second credibility corresponding to the generated data, and finally the first credibility is determined according to the second credibility corresponding to each generated data, and the query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model. Therefore, after obtaining the generated data corresponding to the query data and the retrieval data output by the target model and the first model respectively, at least two second models obtained from the large prediction model library and the prompt template corresponding to each second model are used to generate prompt information corresponding to the query data, the retrieval data and the generated data, each prompt information is input into the corresponding second model, the second credibility corresponding to each generated data is obtained, and based on the second credibility corresponding to each generated data, the first credibility corresponding to each generated data is determined, and the query data, retrieval data, the generated data and the first credibility are associated and stored in the training database corresponding to the target model, thereby realizing a multi-stage evaluation of the generated data corresponding to the query data and the retrieval data, ensuring the accuracy and reliability of the credibility of the generated data, and improving the data quality.

[0165] Figure 5 A flowchart of a model updating method provided by an embodiment of the present disclosure is provided.

[0166] like Figure 5 As shown, the model updating method may include the following steps:

[0167] Step 501, determining a target update method based on at least one of the amount of new data included in the training database, the type of new data, and the current performance parameter of the target model.

[0168] The newly added data in the training database may be generated by the data processing method described in the above embodiment.

[0169] The type of the newly added data may be used to indicate whether the newly added data is data that has passed the evaluation, and the present disclosure does not limit this.

[0170] The target update method may be any update method for updating and training the target model. For example, the target update method may be a full training method, or an incremental training method, or a preference training method, etc., which is not limited in the present disclosure.

[0171] Optionally, based on the amount of new data contained in the training database, the type of new data, and at least one of the current performance parameters of the target model, before determining the target update method, the type of each new data can be determined according to the first credibility in each new data. For example, the higher the first credibility of the new data, the higher its accuracy can be determined. At this time, its type can be determined as data that has passed the evaluation. The lower the first credibility of the new data, the lower its accuracy can be determined. At this time, its type can be determined as data that has not passed the evaluation, thereby improving the accuracy of the determined target update method, and the present disclosure does not limit this.

[0172] In the present disclosure, after the query data, the retrieval data, the generated data and the first credibility are associated and stored in the training database associated with the target model, in order to improve the accuracy and reliability of the target model, the target update mode of the target model can be determined based on the amount of new data in the training database corresponding to the target model, the type of new data, and at least one of the current performance parameters of the target model. For example, when the amount of new data in the training database is large, the target update mode can be determined as a full training mode; when the amount of new data in the training database is small, the target update mode can be determined as an incremental training mode; when the quality of a certain performance parameter in the current performance parameters of the target model is low and the quality of other performance parameters is high, the target update mode can be determined as a preferred training mode, and the present disclosure does not limit this.

[0173] Step 502: Acquire training data from a training database based on the target update method.

[0174] In the present disclosure, after determining the target update method of the target model, training data can be obtained from a training database based on the target update method, thereby realizing update training of the target model and improving the accuracy and effect of the target model.

[0175] It should be noted that the training data obtained from the training database may be different depending on the target model. For example, when the target model is an SFT model, the training data obtained may be SFT training data (SFT Training Data), and when the target model is a preference model or a reward model, the training data obtained may be data for preference or reward model training (Preference / Reward Training Data), which is not limited in this disclosure.

[0176] Step 503: Update the target model according to the training data.

[0177] In the present disclosure, after obtaining the training data, the target model can be updated and trained based on the training data, thereby improving the accuracy and reliability of the target model.

[0178] In the disclosed embodiment, the target update mode is first determined based on the amount of new data in the training database, the type of new data, and at least one of the current performance parameters of the target model, then the training data is obtained from the training database based on the target update mode, and finally the target model is updated and trained based on the training data. Thus, the target update mode of the target model is determined based on the amount of new data in the training database, the type, and at least one of the current performance parameters of the target model, and the training data is obtained based on the target update mode, and the target model is updated and trained, thereby improving the accuracy and effect of the target model.

[0179] Figure 6 A flowchart of a model updating method provided by an embodiment of the present disclosure is provided.

[0180] like Figure 6 As shown, the model updating method may include the following steps:

[0181] Step 601, determining a target update method based on at least one of the amount of new data included in the training database, the type of new data, and the current performance parameter of the target model.

[0182] The newly added data in the training database may be generated by the data processing method described in the above embodiment.

[0183] Step 602: Acquire training data from a training database based on the target update method.

[0184] Step 603: Update the target model according to the training data.

[0185] The specific implementation forms of steps 601 to 603 can refer to the detailed descriptions in other embodiments of the present disclosure and will not be described in detail here.

[0186] Step 604: Test the updated target model using the test data set associated with the target model.

[0187] It should be noted that the test data set associated with the target model may be pre-set. The test data set associated with the target model may be different for different target models, and the present disclosure does not limit this.

[0188] In the present disclosure, after updating and training the target model, the updated target model can be tested based on a test data set associated with the target model to obtain changes in performance parameters of the updated target model, thereby providing conditions for ensuring the effectiveness of the updated target model.

[0189] Step 605: When the performance parameter of the updated target model changes negatively, first indication information is sent to the data processing system to instruct the data processing system to update at least one of the prompt template of the first model and the prompt template of the second model.

[0190] Among them, the first model is a model used to generate generated data in the newly added data.

[0191] The second model is a model for determining the first credibility of the generated data.

[0192] The data processing system may be a system for executing the method described in the above embodiments.

[0193] In the present disclosure, after testing the updated target model using a test data set associated with the target model, when the performance parameters of the updated target model change negatively, it can be determined that the updated target model obtained after updating and training the target model based on the training data in the training database is of poorer quality than the target model before the update, and it can be determined that the quality and accuracy of the training data are low. At this time, at least one of the prompt template of the first model and the prompt template of the second model can be updated according to the first indication information, thereby providing conditions for improving the quality and accuracy of the training data.

[0194] It should be noted that, when the prompt template associated with the second model includes an evaluation indicator (which may be pre-set), the evaluation indicator may also be updated, which is not limited in the present disclosure.

[0195] Optionally, when the performance parameters of the updated target model change negatively, a second indication message may be sent to the data processing system to instruct the data processing system to update at least one of the evaluation rules and evaluation dimensions used to evaluate the generated data, so that the second model's evaluation of the generated data is more accurate, thereby improving the quality of the generated data.

[0196] In the present disclosure, after updating at least one of the prompt templates of the first model and the prompt templates of the second model, query data, retrieval data, generation data and the first credibility can be regained based on the updated prompt template and by using a data processing system. The new query data, new retrieval data, new generation data and new credibility are associated and stored in a training database associated with the target model. Based on the newly added data in the training database, the target model is updated and trained until a target model with positively changing performance parameters is obtained, thereby obtaining a target model with better quality and performance than the target model before the update, thereby improving the accuracy and effect of the target model.

[0197] It should be noted that after obtaining the target model with positive performance parameter changes, the target model currently in use can also be updated based on the target model, thereby improving the effect and accuracy of the target model currently in use, such as Figure 7 As shown, Figure 7 A schematic diagram of the effect of the updated target model provided in the embodiment of the present disclosure, Figure 7 In the above example, both item A-1 and item A-2 belong to item A. Figure 7 It can be seen that the answers generated by the updated target model are more detailed than those generated by the target model before the update, and the effect and quality have been greatly improved. Figure 7 This is only an example and the present disclosure is not limited thereto.

[0198] In the disclosed embodiment, firstly, the target update mode is determined based on the amount of new data, the type of new data, and at least one of the current performance parameters of the target model contained in the training database, and based on the target update mode, the training data is obtained from the training database, and then the target model is updated and trained according to the training data, and finally the updated target model is tested using the test data set associated with the target model, and when the performance parameter of the updated target model changes negatively, the first indication information is sent to the data processing system to instruct the data processing system to update at least one of the prompt template of the first model and the prompt template of the second model. Thus, based on the amount of new data, the type, and at least one of the current performance parameters of the target model in the training database, the update mode of the target model is determined, and after the target model is updated and trained, the target model is tested using the test data set associated with the target model, and when the performance parameter of the updated target model changes negatively, at least one of the prompt template of the first model and the prompt template of the second model is updated, thereby improving the quality and accuracy of the target model.

[0199] Figure 8A flowchart of a human-computer interaction method provided by an embodiment of the present disclosure.

[0200] like Figure 8 As shown, the human-computer interaction method includes the following steps:

[0201] Step 801: receiving an input statement.

[0202] It should be noted that the received input statement may be an input statement in the form of a dialogue, or may be an input statement in the form of an instruction, etc., and the present disclosure does not limit this.

[0203] It should be noted that the human-computer interaction method provided by the present disclosure can be applied to knowledge question-and-answer systems, intelligent assistants, question-and-answer robots and other intelligent entities in various business scenarios, and the present disclosure does not limit this.

[0204] Among them, an intelligent agent is a computer system or entity that can act autonomously, perceive the environment, make decisions and interact with the environment. It usually uses a large language model as the core decision-making and processing unit, and has the ability to think independently and call tools to gradually complete given goals. This disclosure does not limit this.

[0205] Step 802: Identify the input sentence and determine the target business associated with the input sentence.

[0206] The target business may be the real business of the user corresponding to the input statement.

[0207] In the present disclosure, after receiving the user's input statement and before determining the corresponding reply statement based on the input statement, in order to make the determined reply statement conform to the user's actual business scenario and improve the reliability and accuracy of the reply, the input statement can be identified and the target business associated with the input statement can be determined, thereby ensuring that the reply can be combined with the user's actual business scenario.

[0208] Step 803: input the input statement into the target model associated with the target business to obtain a reply statement corresponding to the input statement.

[0209] The target model may be obtained by the model updating method mentioned in the above embodiment.

[0210] In the present disclosure, after determining the target business associated with the input statement, the input statement can be input into the target model associated with the target business to obtain a reply statement corresponding to the input statement, so that the obtained reply statement can better meet user needs, fit the user's actual business scenario, and improve the quality and accuracy of the obtained reply statement.

[0211] In the disclosed embodiment, an input sentence is first received, then the input sentence is recognized, the target business associated with the input sentence is determined, and finally the input sentence is input into the target model associated with the target business to obtain a reply sentence corresponding to the input sentence. Thus, by determining the business associated with the received input sentence and inputting the input sentence into the model associated with the business, a reply sentence is obtained, thereby improving the quality of the reply sentence, improving the effect and accuracy of human-computer interaction, and improving the user experience.

[0212] Fig. 9 A schematic diagram of the structure of a human-computer interaction intelligent agent system provided in one embodiment of the present disclosure.

[0213] like Fig. 9 As shown, the human-computer interaction intelligent agent system 900 includes:

[0214] Input component 901, used to obtain input sentences;

[0215] The processing component 902 is used to call the first agent to identify the input sentence, determine the target business associated with the input sentence, and call the second agent associated with the target business to obtain the reply sentence output by the second agent, wherein at least one of the first agent and the second agent is configured to call the model generated by the model updating method mentioned in the above embodiment to complete the corresponding function;

[0216] The output component 903 is used to output the reply statement.

[0217] It should be noted that the human-computer interaction intelligent agent system provided by the present disclosure may include multiple second agents, each of which can be used for different businesses and call business-related models to generate response statements corresponding to input statements. The present disclosure does not limit this.

[0218] It should be noted that the method for calling the second agent can be any method. For example, the first agent can send a call request to the second agent, and include the input statement in the call request and send it to the second agent. In this method, after receiving the call request from the first agent, the second agent can generate a reply statement corresponding to the input statement, etc. This disclosure does not limit this.

[0219] In the embodiment of the present disclosure, the human-computer interaction intelligent agent system first uses the input component 901 to obtain the user's input statement, and then uses the processing component 902 to identify the input statement by calling the first intelligent agent, determine the target business associated with the input statement, and after determining the second intelligent agent associated with the target business, call the second intelligent agent to obtain the reply statement output by the second intelligent agent, and finally use the output component 903 to output the reply statement generated by the second intelligent agent to the user. Therefore, the human-computer interaction intelligent agent system provided by the present disclosure, through the collaboration between multiple intelligent agents, calls the corresponding intelligent agent to execute the human-computer interaction method mentioned in the above embodiment of the present disclosure, generates and outputs the reply statement corresponding to the user's input statement, improves the accuracy of the reply, improves the effect and reliability of the human-computer interaction method, and also enables the human-computer interaction intelligent agent system provided by the present disclosure to be applied to different business scenarios and fields.

[0220] In order to implement the above embodiments, the present disclosure also proposes a data processing device.

[0221] Fig.10 A schematic diagram of the structure of a data processing device provided in an embodiment of the present disclosure.

[0222] like Fig.10 As shown, the data processing device 1000 includes: a first acquisition module 1001 , a second acquisition module 1002 , a processing module 1003 , a first determination module 1004 , and a storage module 1005 .

[0223] The first acquisition module 1001 is used to acquire log data of a business system using a target model;

[0224] The second acquisition module 1002 is used to parse the log data to obtain query data and search data corresponding to the query data;

[0225] The processing module 1003 is used to input the query data and the search data into the target model and the first model respectively to obtain at least two generated data;

[0226] A first determination module 1004, configured to determine a first credibility of each generated data based on the query data and the retrieved data by using at least two second models;

[0227] The storage module 1005 is used to store the query data, the retrieval data, the generated data and the first credibility in a training database associated with the target model.

[0228] In a possible implementation of the present disclosure, the second acquisition module 1002 is specifically configured to:

[0229] Parsing the log data to obtain query data contained in the log data and a first search segment corresponding to the query data;

[0230] Searching a preset database based on the query data to obtain a second search segment;

[0231] The first search segment and the second search segment are determined as search data.

[0232] In a possible implementation of the present disclosure, the processing module 1003 is specifically configured to:

[0233] Generate first prompt information based on the query data, the retrieval data and a first prompt template corresponding to the target model;

[0234] Inputting the first prompt information into the target model to obtain output generated data;

[0235] Acquire a first model and a second prompt template corresponding to the first model from a large model library;

[0236] Based on the second prompt template, query data and retrieve data to generate second prompt information;

[0237] The second prompt information is input into the first model to obtain output generated data.

[0238] In a possible implementation of the present disclosure, the first determining module 1004 is further configured to:

[0239] At least two second models and a third prompt template corresponding to each second model are obtained from the large model library, wherein at least one second model is different from the first model.

[0240] In a possible implementation of the present disclosure, the first determining module 1004 is further configured to:

[0241] Generate third prompt information based on the query data, the retrieval data, the generated data and each third prompt template;

[0242] Input the third prompt information into the corresponding second model to obtain a second credibility corresponding to the generated data;

[0243] The first credibility is determined according to the plurality of second credibility corresponding to each generated data.

[0244] In a possible implementation of the present disclosure, the first determining module 1004 is further configured to:

[0245] The first confidence level is determined according to a ratio between the number of the second confidence levels that are greater than the confidence level threshold and the total number of the second confidence levels.

[0246] In a possible implementation of the present disclosure, the first determining module 1004 is further configured to:

[0247] Determining the scores for each assessment dimension in the second credibility;

[0248] The first credibility is determined according to the evaluation rules associated with each evaluation dimension and the scores of the generated data in each evaluation dimension.

[0249] The functions and specific implementation principles of the above modules in the embodiments of the present disclosure can be referred to the above method embodiments, and will not be repeated here.

[0250] In the disclosed embodiment, the log data of the business system using the target model is first obtained, and the log data is parsed to obtain query data and retrieval data corresponding to the query data, and then the query data and the retrieval data are respectively input into the target model and the first model to obtain at least two generated data, and then at least two second models are used to determine the first credibility of each generated data based on the query data and the retrieval data, and finally the query data, the retrieval data, the generated data and the first credibility are associated and stored in the training database associated with the target model. Thus, query data and retrieval data are obtained based on the log data of the business system using the target system, and query data and generated data corresponding to the retrieval data are generated based on the target model and the first model, and the generated data is evaluated using the second model and credibility is obtained, and the query data, the retrieval data, the generated data and credibility are associated and stored in the database corresponding to the target model, thereby ensuring that the training data in the training database associated with the target model fits the real business scenario data, and improving the data quality and accuracy.

[0251] In order to implement the above embodiment, the present disclosure also proposes a model updating device.

[0252] Fig.11 A schematic diagram of the structure of a model updating device provided in an embodiment of the present disclosure.

[0253] like Fig.11 As shown, the model updating device 1100 includes: a second determination module 1101 , a third acquisition module 1102 , and a training module 1103 .

[0254] A second determination module 1101 is used to determine a target update mode based on at least one of the amount of new data included in the training database, the type of the new data, and the current performance parameter of the target model, wherein the new data in the training database is generated by the data processing method described in the above embodiment of the present disclosure;

[0255] A third acquisition module 1102 is used to acquire training data from a training database based on a target update method;

[0256] The training module 1103 is used to update and train the target model according to the training data.

[0257] In a possible implementation of the present disclosure, the training module 1103 is further used to:

[0258] Testing the updated target model using the test data set associated with the target model;

[0259] When the performance parameter of the updated target model changes negatively, a first indication message is sent to the data processing system to instruct the data processing system to update at least one of the prompt template of the first model and the prompt template of the second model, wherein the first model is a model for generating generated data in the newly added data, the second model is a model for determining a first credibility of the generated data, and the data processing system is a system for executing the data processing method described in the above embodiments of the present disclosure.

[0260] In a possible implementation of the present disclosure, it further includes:

[0261] An update module (not shown in the figure) is used to send a second indication message to the data processing system when the performance parameters of the updated target model change negatively, so as to instruct the data processing system to update at least one of the evaluation rules and evaluation dimensions used to evaluate the generated data.

[0262] In a possible implementation of the present disclosure, the second determining module 1101 is further configured to:

[0263] The type of each newly added data is determined according to the first credibility of each newly added data.

[0264] The functions and specific implementation principles of the above modules in the embodiments of the present disclosure can be referred to the above method embodiments, and will not be repeated here.

[0265] In the disclosed embodiment, the target update mode is first determined based on the amount of new data in the training database, the type of new data, and at least one of the current performance parameters of the target model, then the training data is obtained from the training database based on the target update mode, and finally the target model is updated and trained based on the training data. Thus, the target update mode of the target model is determined based on the amount of new data in the training database, the type, and at least one of the current performance parameters of the target model, and the training data is obtained based on the target update mode, and the target model is updated and trained, thereby improving the accuracy and effect of the target model.

[0266] In order to implement the above embodiments, the present disclosure also proposes a human-computer interaction device.

[0267] Fig.12 A schematic diagram of the structure of a human-computer interaction device provided in an embodiment of the present disclosure.

[0268] like Fig.12 As shown, the human-computer interaction device 1200 includes: a receiving module 1201 , a recognition module 1202 , and an input module 1203 .

[0269] Receiving module 1201, used for receiving input sentences;

[0270] An identification module 1202 is used to identify the input sentence and determine the target business associated with the input sentence;

[0271] The input module 1203 is used to input the input sentence into the target model associated with the target business to obtain a reply sentence corresponding to the input sentence, wherein the target model is obtained based on the model updating method in the above embodiment of the present disclosure.

[0272] The functions and specific implementation principles of the above modules in the embodiments of the present disclosure can be referred to the above method embodiments, and will not be repeated here.

[0273] In the disclosed embodiment, an input sentence is first received, then the input sentence is recognized, the target business associated with the input sentence is determined, and finally the input sentence is input into the target model associated with the target business to obtain a reply sentence corresponding to the input sentence. Thus, by determining the business associated with the received input sentence and inputting the input sentence into the model associated with the business, a reply sentence is obtained, thereby improving the quality of the reply sentence, improving the effect and accuracy of human-computer interaction, and improving the user experience.

[0274] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0275] Fig.13 A schematic block diagram of an example electronic device 1300 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0276] like Fig.13As shown, the device 1300 includes a computing unit 1301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1302 or a computer program loaded from a storage unit 1308 into a random access memory (RAM) 1303. In the RAM 1303, various programs and data required for the operation of the device 1300 can also be stored. The computing unit 1301, the ROM 1302, and the RAM 1303 are connected to each other via a bus 1304. An input / output (I / O) interface 1305 is also connected to the bus 1304.

[0277] A number of components in the device 1300 are connected to the I / O interface 1305, including: an input unit 1306, such as a keyboard, a mouse, etc.; an output unit 1307, such as various types of displays, speakers, etc.; a storage unit 1308, such as a disk, an optical disk, etc.; and a communication unit 1309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1309 allows the device 1300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0278] The computing unit 1301 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1301 performs the various methods and processes described above, such as data processing methods. For example, in some embodiments, the data processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1308. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1300 via ROM 1302 and / or communication unit 1309. When the computer program is loaded into RAM 1303 and executed by the computing unit 1301, one or more steps of the data processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1301 may be configured to perform the data processing method in any other appropriate manner (e.g., by means of firmware).

[0279] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0280] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0281] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0282] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0283] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0284] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0285] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0286] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In the description of the present disclosure, the words "if" and "if" used can be interpreted as "at the time of" or "when" or "in response to determination" or "under the circumstances of".

[0287] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A data processing method, comprising: Obtain log data from business systems that use the target model; Parsing the log data to obtain query data and search data corresponding to the query data; Inputting the query data and the search data into the target model and the first model respectively to obtain at least two generated data; Determining a first credibility of each of the generated data by using at least two second models and based on the query data and the retrieved data; The query data, the retrieval data, the generated data and the first credibility are associated and stored in a training database associated with the target model.

2. The method of claim 1, wherein: The parsing of the log data to obtain query data and search data corresponding to the query data includes: Parsing the log data to obtain the query data contained in the log data and a first search segment corresponding to the query data; Searching a preset database based on the query data to obtain a second search fragment; The first search segment and the second search segment are determined as the search data.

3. The method of claim 1, wherein: The step of inputting the query data and the search data into the target model and the first model respectively to obtain at least two generated data includes: Generate first prompt information based on the query data, the search data and a first prompt template corresponding to the target model; Inputting the first prompt information into the target model to obtain output generated data; Acquire the first model and a second prompt template corresponding to the first model from a large model library; generating second prompt information based on the second prompt template, the query data and the search data; The second prompt information is input into the first model to obtain the output generated data.

4. The method of claim 3, wherein: Before determining the first credibility of each of the generated data by using at least two second models and based on the query data and the retrieved data, the method further includes: At least two second models and a third prompt template corresponding to each second model are obtained from the large model library, wherein at least one of the second models is different from the first model.

5. The method of claim 4, wherein: The using at least two second models and determining the first credibility of each of the generated data based on the query data and the retrieved data comprises: Generate third prompt information based on the query data, the search data, the generated data and each of the third prompt templates; Inputting the third prompt information into the corresponding second model to obtain a second credibility corresponding to the generated data; The first credibility is determined according to the second credibility corresponding to each of the generated data.

6. The method of claim 5, wherein: The determining the first credibility according to the second credibility corresponding to each of the generated data includes: The first credibility is determined according to a ratio between the number of the second credibility levels that are greater than a credibility threshold and the total number of the second credibility levels.

7. The method of claim 5, wherein: The determining the first credibility according to the second credibility corresponding to each of the generated data includes: determining a score for each evaluation dimension in the second credibility; The first credibility is determined according to the evaluation rules associated with each of the evaluation dimensions and the scores of the generated data in each of the evaluation dimensions.

8. A model updating method, comprising: Determine a target update method based on at least one of the amount of new data contained in the training database, the type of the new data, and the current performance parameter of the target model, wherein the new data in the training database is generated by the method according to any one of claims 1 to 7; Based on the target update method, obtaining training data from the training database; The target model is updated and trained according to the training data.

9. The method of claim 8, wherein: After updating and training the target model according to the training data, the method further includes: Testing the updated target model using a test data set associated with the target model; In the case where the performance parameter of the updated target model changes negatively, a first indication message is sent to the data processing system to instruct the data processing system to update at least one of the prompt template of the first model and the prompt template of the second model, wherein the first model is a model for generating generated data in the newly added data, the second model is a model for determining a first credibility of the generated data, and the data processing system is a system for executing the method described in any one of claims 1-7.

10. The method of claim 9, wherein: The method further comprises: In the case where the performance parameter of the updated target model changes in a negative direction, second indication information is sent to the data processing system to instruct the data processing system to update at least one of the evaluation rules and evaluation dimensions used to evaluate the generated data.

11. The method according to any one of claims 8 to 10, wherein: Before determining the target update mode based on at least one of the amount of new data included in the training database, the type of new data, and the current performance parameter of the target model, the method further includes: The type of each newly added data is determined according to the first credibility of each newly added data.

12. A human-computer interaction method, comprising: Receive input sentences; Identify the input sentence and determine the target business associated with the input sentence; The input sentence is input into a target model associated with the target business to obtain a reply sentence corresponding to the input sentence, wherein the target model is obtained based on the method described in any one of claims 8-11.

13. A human-computer interactive intelligent agent system, comprising: Input component, used to obtain input sentences; A processing component, configured to call a first agent to identify the input sentence, determine a target business associated with the input sentence, and call a second agent associated with the target business to obtain a reply sentence output by the second agent, wherein at least one of the first agent and the second agent is configured to call a model generated by the method according to any one of claims 8 to 11 to complete a corresponding function; An output component is used to output the reply statement.

14. A data processing device, wherein: The device comprises: A first acquisition module is used to acquire log data of a business system using a target model; A second acquisition module is used to parse the log data to obtain query data and search data corresponding to the query data; A processing module, used for inputting the query data and the search data into the target model and the first model respectively, to obtain at least two generated data; A first determination module, configured to determine a first credibility of each of the generated data based on the query data and the retrieved data by using at least two second models; A storage module is used to associate and store the query data, the retrieval data, the generated data and the first credibility in a training database associated with the target model.

15. The device of claim 14, wherein: The second acquisition module is specifically used for: Parsing the log data to obtain the query data contained in the log data and a first search segment corresponding to the query data; Searching a preset database based on the query data to obtain a second search fragment; The first search segment and the second search segment are determined as the search data.

16. The device of claim 14, wherein: The processing module is specifically used for: Generate first prompt information based on the query data, the search data and a first prompt template corresponding to the target model; Inputting the first prompt information into the target model to obtain output generated data; Acquire the first model and a second prompt template corresponding to the first model from a large model library; generating second prompt information based on the second prompt template, the query data and the search data; The second prompt information is input into the first model to obtain the output generated data.

17. The device of claim 16, wherein: The first determining module is further used for: At least two second models and a third prompt template corresponding to each second model are obtained from the large model library, wherein at least one of the second models is different from the first model.

18. The device of claim 17, wherein: The first determining module is further used for: Generate third prompt information based on the query data, the search data, the generated data and each of the third prompt templates; Inputting the third prompt information into the corresponding second model to obtain a second credibility corresponding to the generated data; The first credibility is determined according to the second credibility corresponding to each of the generated data.

19. The device of claim 18, wherein: The first determining module is further used for: The first credibility is determined according to a ratio between the number of the second credibility levels that are greater than a credibility threshold and the total number of the second credibility levels.

20. The apparatus of claim 18, wherein: The first determining module is further used for: determining a score for each evaluation dimension in the second credibility; The first credibility is determined according to the evaluation rules associated with each of the evaluation dimensions and the scores of the generated data in each of the evaluation dimensions.

21. A model updating device, wherein: The device comprises: A second determination module is used to determine the target update mode based on at least one of the amount of new data contained in the training database, the type of the new data, and the current performance parameter of the target model, wherein the new data in the training database is generated by the method according to any one of claims 1 to 7; A third acquisition module, used for acquiring training data from the training database based on the target update mode; A training module is used to update and train the target model according to the training data.

22. The device of claim 21, wherein: The training module is also used for: Testing the updated target model using a test data set associated with the target model; In the case where the performance parameter of the updated target model changes negatively, a first indication message is sent to the data processing system to instruct the data processing system to update at least one of the prompt template of the first model and the prompt template of the second model, wherein the first model is a model for generating generated data in the newly added data, the second model is a model for determining a first credibility of the generated data, and the data processing system is a system for executing the method described in any one of claims 1-7.

23. The device of claim 22, wherein: The device also includes: An updating module is used to send a second indication message to the data processing system when the performance parameter of the updated target model changes negatively, so as to instruct the data processing system to update at least one of the evaluation rules and evaluation dimensions used to evaluate the generated data.

24. The device according to any one of claims 21 to 23, wherein: The second determining module is further used for: The type of each newly added data is determined according to the first credibility of each newly added data.

25. A human-computer interaction device, wherein: The device comprises: A receiving module, used for receiving an input sentence; An identification module, used to identify the input sentence and determine a target business associated with the input sentence; An input module is used to input the input statement into a target model associated with the target business to obtain a reply statement corresponding to the input statement, wherein the target model is obtained based on the method described in any one of claims 8-11.

26. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; Wherein, the memory stores instructions that may be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the data processing method described in any one of claims 1-7, the model updating method described in any one of claims 8-11, or the human-computer interaction method described in claim 12.

27. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the data processing method according to any one of claims 1 to 7, the model updating method according to any one of claims 8 to 11, or the human-computer interaction method according to claim 12.

28. A computer program product, comprising a computer program, which, when executed by a processor, implements the data processing method according to any one of claims 1 to 7, the model updating method according to any one of claims 8 to 11, or the human-computer interaction method according to claim 12.

Images