Training data amplification method and device, model training method and computer equipment

Abstract

The invention relates to a training data amplification method and device, a model training method and computer equipment. The method comprises the following steps: acquiring initial training data in a database; the initial training data comprises an initial input problem; inputting the initial input problem into the training model for multiple times, and outputting multiple training results; and determining amplification training data according to a preset evaluation task and the plurality of training results. By using the training data amplification method provided by the invention, the efficiency of obtaining the training data can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a training data augmentation method, device, model training method and computer equipment. Background Art

[0002] With the rapid development of the Internet, online questionnaire surveys and service evaluations have become an important means for people to protect their rights and interests and improve service quality. With the development of artificial intelligence technology, the method of generating questionnaires through trained models is becoming more and more widely used.

[0003] In traditional technology, the training data used in training models is usually obtained through manual annotation. However, a large amount of training data is required when training models, and the method of obtaining training data through manual annotation is inefficient. Summary of the invention

[0004] Based on this, it is necessary to provide a training data augmentation method, device, model training method and computer equipment that can improve the efficiency of obtaining training data in response to the above technical problems.

[0005] In a first aspect, the present application provides a training data augmentation method, the method comprising:

[0006] Acquire initial training data in a database; the initial training data includes initial input questions;

[0007] Input the initial input question into the training model multiple times and output multiple training results;

[0008] The augmented training data is determined according to the preset evaluation task and the plurality of training results.

[0009] In a second aspect, an embodiment of the present application provides a model training method, the method comprising:

[0010] Obtain target training data, and use the target training data to train the training model to obtain a target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined according to the method provided in the first aspect.

[0011] In a third aspect, an embodiment of the present application provides a training data augmentation device, the device comprising:

[0012] An acquisition module is used to acquire initial training data from a database; the initial training data includes initial input questions;

[0013] An input module, used to input the initial input problem into the training model multiple times and output multiple training results;

[0014] The determination module is used to determine the augmented training data according to a preset evaluation task and multiple training results.

[0015] In a fourth aspect, an embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the methods provided in the first and second aspects are implemented.

[0016] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the methods provided in the first and second aspects above are implemented.

[0017] In a sixth aspect, an embodiment of the present application further provides a computer program product, including a computer program, which implements the steps of the methods provided in the first and second aspects above when the computer program is executed by a processor.

[0018] The above-mentioned training data augmentation method, device, model training method and computer equipment, the method is through obtaining initial training data in a database; the initial training data includes initial input questions; the initial input questions are input into the training model multiple times, and multiple training results are output; according to the preset evaluation tasks and multiple training results, the augmented training data is determined. In this embodiment, the augmented training data can be determined by using the initial training data stored in the database, and the training model, so that when obtaining the augmented training data, no manual labeling is required, thereby improving the efficiency of obtaining training data and reducing the waste of manpower and material resources. In addition, the present application determines the augmented training data from multiple training results according to the preset evaluation tasks, that is, the determined augmented training data is evaluated and can provide the accuracy of the obtained augmented training data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a schematic diagram of the structure of a computer device in one embodiment;

[0020] Figure 2 A schematic diagram of the steps of a training data augmentation method in one embodiment;

[0021] Figure 3 is a schematic diagram of a training result in one embodiment;

[0022] Figure 4 A schematic diagram of the steps of a training data augmentation method in another embodiment;

[0023] Figure 5 A schematic diagram of the steps of a training data augmentation method in another embodiment;

[0024] Figure 6 is a schematic diagram of a training result in another embodiment;

[0025] Figure 7 is a schematic diagram of a training result in another embodiment;

[0026] Figure 8 A schematic diagram of the steps of a training data augmentation method in another embodiment;

[0027] Fig. 9 FIG. 4 is a schematic diagram of the structure of a training data augmentation device in an embodiment. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0029] First of all, before specifically introducing the technical solutions of the disclosed embodiments of the present application, the background technology or technical evolution context on which the embodiments of the present application are based is introduced. With the vigorous development of the Internet, online questionnaire surveys and service evaluations have become an important means for people to safeguard their rights and interests and improve service quality. With the development of artificial intelligence technology, the method of generating survey questions by training questionnaire generation models and using trained questionnaire generation models has become more and more widely used. In traditional technology, the training data used for training questionnaire generation models is usually obtained through manual annotation. However, manual annotation depends on the knowledge level of the annotator, such as learning, profession, etc., and a large amount of training data is required when training the questionnaire generation model. If the training data is obtained through manual annotation, it takes a long time, the cost is high, and the efficiency of obtaining training data is low. In this regard, the present application provides a training data augmentation method.

[0030] The technical solution of the present application and how the technical solution of the present application solves the technical problem are described in detail below with specific embodiments.

[0031] The training data augmentation method provided in the present application can be applied to computer devices, which may be, but are not limited to, industrial computers, laptops, and tablet computers. The internal structure diagram of the computer device can be as follows: Figure 1As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a training data augmentation method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0032] In one embodiment, Figure 2 As shown, a training data augmentation method is provided. This embodiment uses the method applied to a computer device as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0033] Step 200: Obtain initial training data from a database; the initial training data includes initial input questions;

[0034] The database may be a database in a computer device or a database in a server. Initial training data is stored in the database. The initial training data is a small amount of data that has been manually annotated in order to train the model. The initial training data includes an initial input question and an initial output result, that is, the model is trained based on the initial input question and the initial output result. The initial training data may be any data including a question and a result corresponding to the question (i.e., the answer to the question). This embodiment does not limit the specific type of the initial training data.

[0035] A small amount of initial training data is stored in the database, and the computer device can randomly select an initial training data from the database; the initial training data in the database can also be numbered, and the computer device obtains the initial training data in sequence according to the numbering order.

[0036] If the database is a database in a computer device, the computer device directly obtains the initial training data from the database. If the database is a database in a server, the computer device can obtain the initial training data from the database of the server when needed, or the server can send the initial training data to the computer device after receiving a request instruction from the computer device.

[0037] Step 210: Input the initial input question into the training model multiple times and output multiple training results.

[0038] The training model may be a large language model. The training model may be used to generate a questionnaire, or to implement classification, or to generate a target form, etc. This embodiment does not limit the type and function of the training model, as long as the function can be achieved.

[0039] After obtaining the initial training data, the computer device inputs the initial input question into the training model multiple times to obtain multiple training results output by the training model. The training model may output different training results for the same initial input question. For example, the initial input question is input into the training model for the first time to obtain the first training result; the initial input question is input into the training model for the second time to obtain the second training result; the initial input question is input into the training model for the third time to obtain the third training result.

[0040] Step 220: Determine the augmented training data according to the preset evaluation task and multiple training results.

[0041] The preset evaluation task may be a task preset by a user to evaluate whether multiple training results output by the training model meet the requirements. This embodiment does not limit the specific content of the preset evaluation task, as long as the training results can be evaluated.

[0042] After obtaining multiple training results output by the training model, the computer device determines the augmented training data according to the preset evaluation task and the multiple training results. That is, the multiple training results are evaluated according to the preset evaluation task, and a training result with the best evaluation effect is selected from the multiple training results to be determined as the augmented training data.

[0043] The augmented training data determined according to the preset evaluation task and multiple training results may include the training result with the best evaluation effect, or may include the training result with the best evaluation effect and the initial input question corresponding to the training result.

[0044] In an optional embodiment, after determining the augmented training data based on the acquired current training data, the computer device continues to acquire the initial training data from the database, and determines new augmented training data based on the selected initial training data. The initial training data that continues to be acquired from the database may be the same as or different from the initial training data acquired last time.

[0045] The training data augmentation method provided in the embodiment of the present application is through obtaining initial training data in a database; the initial training data includes initial input questions; the initial input questions are input into the training model multiple times, and multiple training results are output; the augmented training data is determined according to the preset evaluation tasks and multiple training results. In this embodiment, the augmented training data can be determined by using the initial training data stored in the database and the training model, so that when obtaining the augmented training data, no manual annotation is required, which can improve the efficiency of obtaining the training data and reduce the waste of manpower and material resources. In addition, the present application determines the augmented training data in multiple training results according to the preset evaluation tasks, that is, the determined augmented training data is evaluated and can provide the accuracy of the obtained augmented training data.

[0046] In one embodiment, the initial training data includes questionnaire survey data and generated form data. The initial training data includes initial input questions and initial output results. If the initial training data is questionnaire survey data, the initial input question may be to establish a target questionnaire, and the initial output result is the title corresponding to the target questionnaire, and the options corresponding to each title; if the initial training data is generated form data, the initial input question may be to establish a target form, and the initial output result is the parameters in the form corresponding to the target form.

[0047] If the initial training data is questionnaire survey data, for example, the initial input question is "Help me design a questionnaire to collect feedback on bicycle usage", the initial output result is as follows Figure 3 shown.

[0048] In an optional embodiment, the initial training data may also be customer service speech content generation data; sales speech content generation data, or document-based question-and-answer dialogue generation data.

[0049] In one embodiment, Figure 4 As shown, it involves an implementation method of determining the augmented training data according to a preset evaluation task and multiple training results, and the steps of the implementation method include:

[0050] Step 400: quantize multiple training results according to a preset evaluation task to obtain a target quantization result.

[0051] After obtaining the plurality of training results, the computer device quantifies the plurality of training results according to the preset evaluation task to obtain the target quantified result. In other words, the plurality of training results represent answers to the initial input question, and the training result is text information. The computer device quantifies the training results according to the preset evaluation task, which means converting the plurality of training results into digital information, and evaluating the training results according to the preset evaluation task and the digital information of the plurality of training results to obtain the target quantified result (target evaluation result).

[0052] In one embodiment, the preset evaluation task includes multiple evaluation indicators, that is, the user can pre-set multiple evaluation indicators to evaluate the training results output by the training model. Figure 5 As shown, it involves quantizing multiple training results according to a preset evaluation task to obtain a target quantization result. The steps of the implementation method include:

[0053] Step 500: For each training result, quantize the training result according to at least one evaluation index to obtain an initial quantization result corresponding to the training result.

[0054] The computer device quantifies each training result output by the training model according to at least one evaluation index to obtain a quantized result corresponding to the evaluation index, and determines an initial quantized result corresponding to the training result according to the quantized results corresponding to all the evaluation indexes. If the training result is quantized according to one evaluation index, the quantized result corresponding to the evaluation index is the initial quantized result corresponding to the training result. If the training result is quantized according to two evaluation indexes, the quantized results corresponding to the two evaluation indexes are obtained; the quantized results corresponding to the two evaluation indexes are weighted summed or weighted averaged to obtain the initial quantized result corresponding to the training result.

[0055] That is to say, after the computer device converts the training results into digital information, it selects any one of multiple evaluation indicators, or any multiple of them to score the converted training results to obtain the score value corresponding to the evaluation indicator; the score values ​​corresponding to all evaluation indicators are weighted summed or weighted averaged to obtain the quantitative results (scoring values) corresponding to the training results.

[0056] Step 510: Determine a target quantization result according to the initial quantization results corresponding to all training results.

[0057] After obtaining the initial quantization result corresponding to each training result, the computer device determines a target quantization result from the initial quantization results according to the initial quantization results corresponding to all the training results. This embodiment does not limit the specific method of determining the target quantization result according to the initial quantization results corresponding to all the training results, as long as its function can be achieved.

[0058] In one embodiment, an implementation method of determining a target quantization result according to initial quantization results corresponding to all training results includes:

[0059] The initial quantization results that meet the preset conditions among the initial quantization results corresponding to all training results are determined as the target quantization results.

[0060] The preset condition may be pre-set by a user and stored in a computer device. The preset condition may be reaching a preset threshold, or may be the best initial quantization result among the initial quantization results corresponding to a plurality of training results.

[0061] After obtaining the initial quantization results corresponding to all the training results, the computer device determines an initial quantization result that meets a preset condition among the initial quantization results, and determines the initial quantization result as the target quantization result.

[0062] When the preset condition is a preset threshold, the computer device compares the initial quantization result corresponding to the obtained training result with the preset threshold, and determines the initial quantization result greater than the preset threshold as the target quantization result. When the preset condition is the best initial quantization result among the initial quantization results corresponding to multiple training results, the computer device compares the initial quantization results corresponding to the multiple training results, and determines the largest initial quantization result as the target quantization result.

[0063] Step 410: Determine the augmented training data according to the target quantization result.

[0064] After determining the target quantization result, the computer device determines the training result corresponding to the target quantization result, and determines the augmented training data according to the training result corresponding to the target quantization result. The computer device may directly determine the training result determined by the target quantization result as the augmented training data; or may determine the training result determined by the target quantization result and the initial input question as the augmented training data.

[0065] In this embodiment, for each training result obtained, it is quantified according to at least one evaluation index in the preset evaluation task to obtain an initial quantization result corresponding to each training result; and the target quantization result is determined according to the initial quantization results corresponding to all the training results. In this way, the evaluation of multiple training results can be achieved according to the preset evaluation task, so that the optimal training result can be obtained, and then more accurate augmented training data can be obtained.

[0066] In one embodiment, the preset evaluation tasks include: a correctness index, a rationality index, a relevance index, and an emotional attribute index.

[0067] The correctness index is used to determine the similarity between the training results and the initial output results in the initial training data. The rationality index is used to determine the similarity between the questions and options in the training results. The relevance index is used to determine the similarity between the questions in the training results and the initial input questions in the initial training data. The emotional attribute index is used to determine the emotional value of the training results.

[0068] The correctness of the training result is evaluated, that is, the similarity between the training result and the initial output result in the initial training data is determined, and whether the training result is correct is determined by determining whether the similarity reaches a first similarity threshold. The computer device performs vectorization (embedding) processing on the training result and the initial output result respectively, and calculates the similarity between the two processed vectors. The similarity between the training result and the initial output result can be determined by calculating the cosine similarity (cosine), or by calculating the Euclidean space distance. This embodiment does not limit the specific method for determining the similarity, as long as its function can be achieved.

[0069] In an optional embodiment, assuming that the training result after vectorization is represented as v, and the initial output result after vectorization is represented as w, the cosine similarity between the training result and the initial output result can be expressed as: Among them, cosine(v,w) represents the cosine similarity, N represents the number of vectors in the training results and the initial output results after vectorization, and v i represents the i-th vector in the training result after vectorization processing; w i Represents the i-th vector in the training result after vectorization.

[0070] The training results include the questions set for the initial input questions and the options corresponding to the questions. The rationality of the training results is evaluated, that is, the similarity between the questions and options in the training results is determined, and whether the training results are reasonable is determined by determining whether the similarity reaches a second similarity threshold. The second similarity threshold and the first similarity threshold can be the same or different. The determination of the similarity between the questions and options in the training results can refer to the above-mentioned method for determining the similarity between the training results and the initial output results, which will not be repeated here.

[0071] In an optional embodiment, the initial input question is "Help me design a questionnaire to collect bicycle usage feedback", and the training results output by the training model are as follows: Figure 6 As shown. Figure 6 It can be seen that the questions and options in the training results do not match, that is, the similarity is low. The question asks about the frequency of using bicycles, while the options are light riding and easy to obtain; the question asks which brand of bicycle to use, while the option is online car-hailing.

[0072] The relevance of the training results is evaluated, that is, the similarity between the questions in the training results and the initial input questions is determined, and the relevance of the training results is determined by determining whether the similarity reaches a third similarity threshold. The third similarity threshold and the first similarity threshold may be the same or different. The determination of the similarity between the questions in the training results and the initial input questions can refer to the above-mentioned method for determining the similarity between the training results and the initial output results, which will not be repeated here.

[0073] In an optional embodiment, the initial input question is "Help me design a questionnaire to collect bicycle usage feedback", and the training results output by the training model are as follows: Figure 7 As shown. Figure 7 It can be seen that: Figure 7 The questions in the training results are not related to the initial input questions, that is, the similarity is low. The initial input question is to design a questionnaire for feedback on the use of bicycles, while the questions in the training results are related to online car-hailing.

[0074] Emotional attributes include positive emotions and negative emotions. The emotional attributes of the training results are evaluated, that is, whether the training results are positive emotions or negative emotions. In other words, by calculating the emotional value of the training results, it is determined by the emotional value whether the training results are positive emotions or negative emotions. Optionally, the computer device can compare the emotional value with a preset emotional threshold. If the preset emotional threshold is reached, it means that the training result is a positive emotion. If the preset emotional threshold is not reached, it means that the training result is a negative emotion. The evaluation of the emotional attributes of the training results can be evaluated using NLP (Neuro-Linguistic Programming).

[0075] In an optional embodiment, the computer device inputs the training result into a pre-trained emotion attribute determination model to obtain an emotion value corresponding to the training result.

[0076] In this embodiment, four evaluation indicators that may be included in the preset evaluation task are described, namely, correctness indicator, rationality index, relevance index and emotional attribute index. Users can select any one or more of these four evaluation indicators to evaluate the training results, which can improve the practicality of the training data augmentation method.

[0077] In an optional embodiment, the method for evaluating the training results using the correctness index, rationality index, relevance index and emotional attribute index in the preset evaluation task includes:

[0078] Based on the correctness value indicator, determine the similarity between the training result and the initial output result to obtain a first similarity;

[0079] Based on the rationality index, the similarity between the questions and the options in the training results is determined to obtain a second similarity;

[0080] Based on the relevance index, determine the similarity between the initial input question and the questions in the training result to obtain a third similarity;

[0081] Determine the emotion value of the training result based on the emotion attribute index;

[0082] An initial quantization result corresponding to the training result is determined according to the first similarity, the second similarity, the third similarity and the emotion value.

[0083] The computer device obtains an initial quantization result by calculating a weighted sum or a weighted average of the first similarity, the second similarity, the third similarity and the emotion value.

[0084] In one embodiment, after determining to expand the training data, the training data expansion method further includes:

[0085] The augmented training data is stored in the database.

[0086] After obtaining the augmented training data, the computer device stores the augmented training data in a database. The augmented training data includes the training results corresponding to the target quantization results, that is, the optimal training results selected from multiple training results according to the preset evaluation task. Optionally, when the computer device stores the augmented training data in the database, it can establish an association relationship between the initial input problem and the augmented training data in the database after storing the augmented training data in the database, that is, one initial input problem corresponds to two output results, one is the initial output result, and the other is the optimal training result determined from multiple training results according to the preset evaluation task. The computer device can also establish an association relationship between the initial input problem and the augmented training data, and then store the initial input problem and the augmented training data in the database, that is, the database includes the initial input problem and the initial output result, as well as the two training data of the initial input problem and the augmented training data.

[0087] In this embodiment, by storing the augmented training data in the database, the training data stored in the database can be increased, so that the initial training data selected from the database can be increased, and more augmented training data can be determined, making the training data augmentation method more practical and reliable.

[0088] An embodiment of the present application provides a model training method, the method comprising the steps of:

[0089] Obtain target training data, and use the target training data to train the training model to obtain a target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined according to the training data augmentation method provided in the above embodiment.

[0090] The data for training the training model may include initial training data and augmented training data. If the augmented training data is stored in a database after the augmented training data is determined, the computer device can directly obtain the target training data in the database; if the augmented training data is stored in other databases after the augmented training data is determined, the computer device obtains the initial training data in the database and the augmented training data in other databases.

[0091] After acquiring the target training data, the computer device trains the training model according to the target training data to obtain the target model. The training model is the same as the training model used when determining the augmented training data.

[0092] The model training method provided in this embodiment obtains target training data and uses the target training data to train the training model to obtain a target model. The target training data includes initial training data and augmented training data. The augmented training data is determined according to the training data augmentation method provided in the above embodiment. The model training method provided in this embodiment has all the beneficial effects of the training data augmentation method, which will not be described in detail here.

[0093] See also Figure 8 An embodiment of the present application provides a training data augmentation method, the method comprising the steps of:

[0094] Step 800: Obtain initial training data from a database; the initial training data includes initial input questions and initial output results;

[0095] Step 810: Input the initial input question into the training model multiple times to obtain multiple training results;

[0096] Step 820: for each training result, score the training result according to at least one evaluation indicator in a preset evaluation task to obtain a score value corresponding to the training result; the preset evaluation task includes a correctness indicator, a rationality indicator, a relevance indicator, and an emotional attribute indicator;

[0097] Step 830: Compare the score values ​​corresponding to all training results, and determine the training result corresponding to the maximum score value as the target training result;

[0098] Step 840: determine the initial input question and the target training result as augmented training data, and store the augmented training data in a database.

[0099] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0100] Based on the same inventive concept, the embodiment of the present application also provides a training data augmentation device for implementing the training data augmentation method involved above, and a model training device for the model training method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more training data augmentation devices and model training device embodiments provided below can refer to the limitations of the training data augmentation method and the model training method above, and will not be repeated here.

[0101] In one embodiment, Fig. 9 As shown, a training data augmentation device 10 is provided, comprising: an acquisition module 11, an input module 12 and a determination module 13, wherein:

[0102] The acquisition module 11 is used to acquire initial training data from the database; the initial training data includes initial input questions.

[0103] The input module 12 is used to input the initial input question into the training model multiple times and output multiple training results.

[0104] The determination module 13 is used to determine the expanded training data according to the preset evaluation task and multiple training results.

[0105] In one embodiment, the determination module 13 includes a quantization unit and a determination unit. The quantization unit is used to quantize multiple training results according to a preset evaluation task to obtain a target quantization result; the determination unit is used to determine the augmented training data according to the target quantization result.

[0106] In one embodiment, the quantization unit includes a quantization subunit and a determination subunit. The quantization subunit is used to quantize the training result according to at least one evaluation index for each training result to obtain an initial quantization result corresponding to the training result; the determination subunit is used to determine the target quantization result according to the initial quantization results corresponding to all training results.

[0107] In one embodiment, the preset evaluation tasks include: correctness index, rationality index, relevance index and emotional attribute index; the correctness index is used to determine the similarity between the training results and the initial output results in the initial training data, the rationality index is used to determine the similarity between the questions and options in the training results, and the relevance index is used to determine the similarity between the questions in the training results and the initial input questions; the emotional attribute index is used to determine the emotional value of the training results.

[0108] In one embodiment, the determination subunit is specifically configured to determine an initial quantization result that satisfies a preset condition among the initial quantization results corresponding to the plurality of training results as a target quantization result.

[0109] In one embodiment, the initial training data includes questionnaire data and generated form data.

[0110] In one embodiment, the training data augmentation device 10 further includes a storage module. The storage module is used to store the augmented training data in a database.

[0111] In one embodiment, a model training device is provided, which includes a training module.

[0112] The training module is used to obtain target training data, and use the target training data to train the training model to obtain the target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined using the above-mentioned training data augmentation device.

[0113] Each module in the above-mentioned training data augmentation device and model training device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0114] In one embodiment, a computer device is provided, whose internal structure diagram can be as follows: Figure 1 As shown. Those skilled in the art can understand that Figure 1The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0115] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0116] Acquire initial training data in a database; the initial training data includes initial input questions;

[0117] Input the initial input question into the training model multiple times and output multiple training results;

[0118] Determine the augmented training data based on preset evaluation tasks and multiple training results.

[0119] In one embodiment, when the processor executes the computer program, the following steps are also implemented: according to a preset evaluation task, multiple training results are quantized to obtain a target quantization result; and according to the target quantization result, the expansion training data is determined.

[0120] In one embodiment, when the processor executes the computer program, the following steps are also implemented: for each training result, the training result is quantized according to at least one evaluation index to obtain an initial quantization result corresponding to the training result; and the target quantization result is determined according to the initial quantization results corresponding to all the training results.

[0121] In one embodiment, the preset evaluation tasks include: correctness index, rationality index, relevance index and emotional attribute index; the correctness index is used to determine the similarity between the training results and the initial output results in the initial training data, the rationality index is used to determine the similarity between the questions and options in the training results, and the relevance index is used to determine the similarity between the questions in the training results and the initial input questions; the emotional attribute index is used to determine the emotional value of the training results.

[0122] In one embodiment, when the processor executes the computer program, the following steps are further implemented: an initial quantization result that meets a preset condition among the initial quantization results corresponding to the multiple training results is determined as a target quantization result.

[0123] In one embodiment, the initial training data includes questionnaire data and generated form data.

[0124] In one embodiment, when the processor executes the computer program, the following steps are further implemented: storing the augmented training data in a database.

[0125] In one embodiment, when the processor executes the computer program, the following steps are also implemented: acquiring target training data, training the training model using the target training data, and obtaining the target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined by executing the steps in the above embodiment.

[0126] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0127] Acquire initial training data in a database; the initial training data includes initial input questions;

[0128] Input the initial input question into the training model multiple times and output multiple training results;

[0129] Determine the augmented training data based on preset evaluation tasks and multiple training results.

[0130] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: according to a preset evaluation task, multiple training results are quantified to obtain a target quantization result; according to the target quantization result, the expansion training data is determined.

[0131] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: for each training result, the training result is quantized according to at least one evaluation index to obtain an initial quantization result corresponding to the training result; and the target quantization result is determined according to the initial quantization results corresponding to all training results.

[0132] In one embodiment, the preset evaluation tasks include: correctness index, rationality index, relevance index and emotional attribute index; the correctness index is used to determine the similarity between the training results and the initial output results in the initial training data, the rationality index is used to determine the similarity between the questions and options in the training results, and the relevance index is used to determine the similarity between the questions in the training results and the initial input questions; the emotional attribute index is used to determine the emotional value of the training results.

[0133] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: an initial quantization result that meets a preset condition among the initial quantization results corresponding to the multiple training results is determined as a target quantization result.

[0134] In one embodiment, the initial training data includes questionnaire data and generated form data.

[0135] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: storing the augmented training data in a database.

[0136] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: acquiring target training data, training the training model using the target training data, and obtaining the target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined by executing the steps of the above embodiment.

[0137] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0138] Acquire initial training data in a database; the initial training data includes initial input questions;

[0139] Input the initial input question into the training model multiple times and output multiple training results;

[0140] Determine the augmented training data based on preset evaluation tasks and multiple training results.

[0141] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: according to a preset evaluation task, multiple training results are quantified to obtain a target quantization result; according to the target quantization result, the expansion training data is determined.

[0142] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0143] For each training result, the training result is quantized according to at least one evaluation index to obtain an initial quantization result corresponding to the training result; and a target quantization result is determined according to the initial quantization results corresponding to all the training results.

[0144] In one embodiment, the preset evaluation tasks include: correctness index, rationality index, relevance index and emotional attribute index; the correctness index is used to determine the similarity between the training results and the initial output results in the initial training data, the rationality index is used to determine the similarity between the questions and options in the training results, and the relevance index is used to determine the similarity between the questions in the training results and the initial input questions; the emotional attribute index is used to determine the emotional value of the training results.

[0145] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: an initial quantization result that meets a preset condition among the initial quantization results corresponding to the multiple training results is determined as a target quantization result.

[0146] In one embodiment, the initial training data includes questionnaire data and generated form data.

[0147] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: storing the augmented training data in a database.

[0148] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: acquiring target training data, training the training model using the target training data, and obtaining the target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined by executing the steps of the above embodiment.

[0149] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0150] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0151] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training data augmentation method, characterized in that: The method comprises: Acquire initial training data in a database; the initial training data includes initial input questions; Inputting the initial input question into the training model multiple times, and outputting multiple training results; The augmented training data is determined according to the preset evaluation task and the plurality of training results.

2. The method according to claim 1, characterized in that The step of determining the augmented training data according to the preset evaluation task and the plurality of training results comprises: According to the preset evaluation task, quantify the plurality of training results to obtain a target quantization result; The augmented training data is determined according to the target quantization result.

3. The method according to claim 2, characterized in that The preset evaluation task includes a plurality of evaluation indicators, and the plurality of training results are quantified according to the preset evaluation task to obtain a target quantization result, including: For each of the training results, quantify the training result according to at least one of the evaluation indicators to obtain an initial quantization result corresponding to the training result; The target quantization result is determined according to the initial quantization results corresponding to all the training results.

4. The method according to claim 3, characterized in that The preset evaluation tasks include: correctness index, rationality index, relevance index and emotional attribute index; The correctness index is used to determine the similarity between the training results and the initial output results in the initial training data, the rationality index is used to determine the similarity between the questions and options in the training results, and the relevance index is used to determine the similarity between the questions in the training results and the initial input questions; the emotional attribute index is used to determine the emotional value of the training results.

5. The method according to claim 3, characterized in that: The step of determining the target quantization result according to the initial quantization results corresponding to all the training results includes: An initial quantization result that meets a preset condition among the initial quantization results corresponding to the plurality of training results is determined as the target quantization result.

6. The method according to any one of claims 1 to 5, characterized in that The initial training data includes questionnaire survey data and generated form data.

7. The method according to any one of claims 1 to 5, characterized in that The method further comprises: The augmented training data is stored in the database.

8. A model training method, characterized in that: The method comprises: Obtain target training data, and use the target training data to train a training model to obtain a target model; the target training data includes initial training data and augmented training data, and the augmented training data is determined according to the method described in any one of claims 1-7.

9. A training data augmentation device, characterized in that: The device comprises: An acquisition module, used to acquire initial training data from a database; the initial training data includes initial input questions; An input module, used to input the initial input question into the training model multiple times and output multiple training results; The determination module is used to determine the expanded training data according to the preset evaluation task and the plurality of training results.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images