Intelligent interaction system applied to AI training

Abstract

The invention relates to the technical field of AI interaction, in particular to an intelligent interaction system applied to AI training, which comprises a storage module used for storing model training data, a learning module used for training an AI model based on the stored model training data, and an interaction module used for providing an interaction platform for a user and the AI model, the recording module is used for recording interaction contents and interaction times of a plurality of users and the AI model and storing the interaction contents as model training data; the auxiliary module is used for periodically dividing the interaction of each user and determining the interaction times and corresponding interaction contents of each user in a corresponding period; the response module is used for judging whether training of the AI model meets the standard or not based on the response activeness and judging the reason for not meeting the standard; and the control module is used for correcting the corresponding parameters. According to the invention, real-time monitoring and accurate adjustment of training of the AI model are effectively realized according to the interaction condition of the user, and the training efficiency of the AI model is effectively improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of AI interaction, and in particular to an intelligent interaction system applied to AI training. BACKGROUND

[0002] AI training refers to a process of inputting data into an AI model, learning data features and rules by the model, adjusting internal parameters of the model according to a learning result, and gradually improving the ability to complete a specific task. The AI model refers to a computer program or data framework that can simulate human intelligent behavior through algorithms and data. The core is to use mathematical logic and data rules to realize the processing ability of a specific task. Its essence is to use a set of rules to enable a computer to autonomously complete a complex task without individual programming. AI models are various, and a chat interaction model refers to a technical framework for realizing natural language interaction between humans and machines. The chat interaction model can understand text information of a user, analyze the intention and demand of the user, and generate a response that is consistent with a context and logically coherent. The intelligent interaction system applied to AI training refers to a tool for assisting an AI model in a training process of the AI model. The intelligent interaction system has important significance for improving the interaction ability of the AI model. With the continuous progress of living standards, the research on the chat interaction model has important research significance for the progress of human living standards.

[0003] Chinese Patent Publication No. CN116774817A discloses an AI interaction control method and system. The AI interaction control method and system are used to make a gesture posture matched by AI conform to a posture action style and a unique speaking rhythm of a dialoger, are more diverse and more natural, and improve the interactive experience in a human-computer interaction process. The method includes: acquiring a dialog video stream of the dialoger; inputting the dialog video stream into a target feature extraction model to perform feature extraction, and determining text information, rhythm information, identity information, and body state information of the dialoger; determining AI feedback information according to the text information, and determining a target gesture group from a feature mapping database according to the AI feedback information, the rhythm information, the identity information, and the body state information; inputting the target gesture group into a style fusion model to perform feature fusion, and generating an interactive gesture conforming to the posture action style and the dialog style of the dialoger. The style fusion model is obtained through training of a plurality of gesture samples; and the AI feedback information and the interactive gesture are displayed to the dialoger.

[0004] It can be seen that the above scheme determines AI feedback information through a dialog video stream, determines a target gesture group conforming to a dialoger style, and generates an interactive gesture. The AI feedback information and the interactive gesture are displayed to the dialoger. Gesture postures matched by AI are made to conform to a posture action style and a unique speaking rhythm of the dialoger based on text information, rhythm information, identity information, and body state information of the dialoger. However, the above scheme cannot monitor and accurately adjust the training of the AI model in real time according to the interaction of the user, and cannot guarantee the training efficiency of the AI model. SUMMARY

[0005] To this end, the present application provides an intelligent interaction system applied to AI training to overcome the problem that the training efficiency of AI model is low due to the inability to monitor and accurately adjust the training of AI model in real time according to the interaction of users in the prior art.

[0006] To achieve the above-mentioned purpose, the present application provides an intelligent interaction system applied to AI training, comprising: a storage module for storing model training data required for training AI model; a learning module connected with the storage module for training AI model based on the stored model training data; an interaction module for providing an interaction platform for users and AI model; a recording module connected with the storage module and the interaction module respectively for recording the interaction content of a plurality of users and AI model, and recording the interaction times of each user; the recording module is also used to store the interaction content and mark it as the model training data; an auxiliary module connected with the recording module for periodically dividing the interaction of each user, and determining the interaction times of each user in the corresponding period after division and the corresponding interaction content; a response module connected with the auxiliary module for determining whether the training of AI model meets the standard based on response activity, and determining the reason why the training of AI model does not meet the standard based on interaction similarity in the case that the training of AI model does not meet the standard; a control module connected with the storage module, the learning module and the response module respectively for correcting the total number of model training data based on the difference of interaction similarity, increasing the context weight parameter based on the ratio of words, increasing the cosine similarity based on the ratio of interaction similarity, increasing the detail weight based on the ratio of inquiry, or increasing the maximum length based on the difficulty of inquiry vocabulary.

[0007] Further, the response module is also used to determine whether the registration of users meets the standard based on interaction frequency, and remove the family registered users from the total number of registered users in the case that the registered users are determined to be family registered users, or keep the corresponding registered users in the total number of registered users in the case that the registered users are determined to be active registered users.

[0008] Further, the response module is also used to determine whether the training of AI model meets the standard based on the response activity after the removal of the family registered users and the reservation of the active registered users, and determine the reason why the training of AI model does not meet the standard based on the interaction similarity according to the determination result.

[0009] Further, the response module is further used to determine the reason why the AI model training does not meet the standard based on the interaction similarity, and based on the interaction similarity difference to correct the total number of model training data when it is determined that the AI model training accuracy does not meet the standard.

[0010] Further, the control module is further used to increase the total number of model training data based on the interaction similarity difference, and the increase amplitude of the total number of model training data is proportional to the interaction similarity difference.

[0011] Further, the control module is further used to decrease the context weight parameter based on the word ratio after the total number of model training data is increased, and the decrease amplitude of the context weight parameter is proportional to the word ratio.

[0012] Further, the response module is further used to determine the reason why the AI model training does not meet the standard based on the interaction similarity after the context weight parameter is decreased, and correct the cosine similarity based on the interaction similarity ratio or correct the detail weight based on the inquiry ratio according to the determination result.

[0013] Further, the control module is further used to increase the cosine similarity based on the interaction similarity ratio, and the increase amplitude of the cosine similarity is inversely proportional to the interaction similarity ratio.

[0014] Further, the control module is further used to increase the detail weight based on the inquiry ratio, and the increase amplitude of the detail weight is proportional to the inquiry ratio.

[0015] Further, the control module is further used to increase the maximum length based on the difficulty of inquiry vocabulary after the detail weight is increased, and the increase amplitude of the maximum length is proportional to the difficulty of inquiry vocabulary.

[0016] Compared with the prior art, the beneficial effects of the present application are that the present application sets the response module and the control module, the response module determines whether the user registration meets the standard based on the interaction frequency, determines whether the AI model training meets the standard based on the response activity after the kinship registered users are removed and the active registered users are retained, can quickly and accurately complete the determination of whether the AI model training meets the standard, determines the reason why the AI model training does not meet the standard based on the interaction similarity when it is determined that the AI model training does not meet the standard, effectively realizes real-time monitoring of the AI model training according to the user interaction, and the control module corrects the corresponding parameters, effectively improves the AI model training efficiency while effectively realizing accurate adjustment of the AI model training according to the user interaction.

[0017] Further, the response module provided by the present application is used to determine whether the registration of the user meets the standard based on the interaction frequency, which is conducive to determining whether the training of the AI model meets the standard when the registered user is a family registered user or an active registered user, and further improves the training efficiency of the AI model while realizing real-time monitoring of the training of the AI model according to the interaction of the user.

[0018] Further, the response module provided by the present application is used to determine whether the registration of the user meets the standard based on the interaction frequency, which is conducive to determining whether the training of the AI model meets the standard when the registered user is a family registered user or an active registered user, and further improves the training efficiency of the AI model while realizing real-time monitoring of the training of the AI model according to the interaction of the user.

[0019] Further, the response module provided by the present application is used to determine whether the registration of the user meets the standard based on the interaction frequency, which is conducive to determining whether the training of the AI model meets the standard when the registered user is a family registered user or an active registered user, and further improves the training efficiency of the AI model while realizing real-time monitoring of the training of the AI model according to the interaction of the user.

[0020] Further, the control module provided by the present application is used to increase the total number of model training data based on the interaction similarity difference, which effectively avoids the situation that the training of the AI model does not meet the standard due to the total number of model training not meeting the standard, and further improves the training efficiency of the AI model while realizing accurate adjustment of the training of the AI model according to the interaction of the user. Further, the control module provided by the present application is used to decrease the context weight parameter based on the word ratio after the total number of model training data is increased, which effectively reduces the context weight parameter and maintains the coordination between the context weight parameter and the interaction of the user, and further improves the training efficiency of the AI model while realizing accurate adjustment of the training of the AI model according to the interaction of the user. Further, the response module provided by the present application is used to determine the reason why the training of the AI model does not meet the standard based on the interaction similarity after the context weight parameter is reduced, which can further determine the reason why the standard is not met, and avoids misjudgment between the cosine similarity that needs to be corrected and the detail weight that needs to be corrected, and further improves the training efficiency of the AI model while realizing real-time monitoring of the training of the AI model according to the interaction of the user.

[0021] Further, the control module is further used to increase the cosine similarity based on the interaction similarity ratio, effectively avoiding the situation that the training of the AI model does not meet the standard due to the cosine similarity not meeting the standard, further realizing the accurate adjustment of the training of the AI model according to the interaction of the user, and further improving the training efficiency of the AI model.

[0022] Further, the control module is further used to increase the detail weight based on the inquiry ratio, which can effectively avoid the situation that the training of the AI model does not meet the standard due to the detail weight not meeting the standard, further realizing the accurate adjustment of the training of the AI model according to the interaction of the user, and further improving the training efficiency of the AI model.

[0023] Further, the control module is further used to increase the maximum length based on the difficulty of the inquiry vocabulary after the completion of the increase of the detail weight, effectively maintaining the coordination of the maximum length and the difficulty of the inquiry vocabulary, effectively avoiding the situation that the training of the AI model is determined to be not in line with the standard due to the maximum length not meeting the standard, further realizing the accurate adjustment of the training of the AI model according to the interaction of the user, and further improving the training efficiency of the AI model. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 a structural block diagram of an intelligent interaction system to which the embodiment of the present application is applied for AI training; Figure 2 a working flowchart of the intelligent interaction system to which the embodiment of the present application is applied for AI training; Figure 3 a flowchart for determining whether the training of the AI model meets the standard and the reason why the training of the AI model does not meet the standard; Figure 4 a flowchart for determining the reason why the training of the AI model does not meet the standard. DETAILED DESCRIPTION

[0025] In order to make the objects and advantages of the present application more clear, the present application will be further described below in combination with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0026] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application, and are not intended to limit the protection scope of the present application.

[0027] Moreover, it needs to be explained that in the description of the present application, unless explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0028] Please refer to Figure 1 The application embodiment of the present application applied to the intelligent interactive system of AI training is shown in the structure block diagram. The intelligent interactive system of AI training described in the application embodiment comprises a storage module, a learning module, an interactive module, a recording module, an auxiliary module, a response module and a control module; wherein, The storage module is used to store the model training data required for training AI model; The learning module is connected with the storage module, and is used to train AI model based on the stored model training data; The interactive module is used to provide an interactive platform for users and AI model; The recording module is connected with the storage module and the interactive module respectively, and is used to record the interactive content of a plurality of users and AI model, and record the interactive times of each user; the recording module is also used to store the interactive content and mark it as the model training data; The auxiliary module is connected with the recording module, and is used to periodically divide the interaction of each user, and determine the interactive times of each user in the corresponding period after division and the corresponding interactive content; The response module is connected with the auxiliary module, and is used to determine whether the training of AI model meets the standard based on the response activity, and determine the reason why the training of AI model does not meet the standard based on the interactive similarity in the case that the training of AI model does not meet the standard; The control module is connected with the storage module, the learning module and the response module respectively, and is used to correct the total number of model training data based on the difference of interactive similarity, increase the context weight parameter based on the ratio of words, increase the cosine similarity based on the ratio of interactive similarity, increase the detail weight based on the inquiry ratio, or increase the maximum length based on the difficulty of inquiry vocabulary; Specifically, when the auxiliary module periodically divides the interaction of each user, the division period is 72h, that is, the current day is selected as the division day, the first 36h before the division day and the division day are selected as a single division period, and the next division period starts after the end of the division day, that is, the next 72h after the end of the division day is the next division period.

[0029] Please refer toFigure 2 As shown in the figure, it is a workflow diagram of the intelligent interaction system applied to AI training according to the embodiment of the present application. When the intelligent interaction system applied to AI training according to the embodiment of the present application is running, the storage module stores the model training data required for training the AI model, the learning module trains the AI model based on the stored model training data, the interaction module provides the interaction platform for the user and the AI model, the recording module records the interaction content of the user and the AI model and records the interaction times of each user, and the recording module is also used to store the interaction content and mark it as model training data. The auxiliary module periodically divides the interaction of each user and determines the interaction times and the corresponding interaction content of each user in the corresponding period after the division. The response module determines whether the training of the AI model meets the standard based on the response activity. The response module determines the reason why the training of the AI model does not meet the standard based on the interaction similarity when it is determined that the training of the AI model does not meet the standard. The control module corrects the total number of model training data based on the interaction similarity difference, increases the context weight parameter based on the word ratio, or increases the cosine similarity based on the interaction similarity ratio. The control module is also used to increase the detail weight based on the inquiry ratio, or increase the maximum length based on the difficulty of the inquiry vocabulary.

[0030] Please continue to refer to Figure 3 As shown in the figure, it is a flowchart for determining whether the training of the AI model meets the standard and the reason why the training of the AI model does not meet the standard according to the embodiment of the present application. The response module according to the present application is also used to determine whether the registration of the user meets the standard based on the interaction frequency. If the interaction frequency is less than or equal to the preset interaction frequency F set in the response module, the response module determines that the registered user is a family registered user, the registration of the user does not meet the standard, and the family registered user is removed from the total number of registered users. The preset interaction frequency F in this embodiment is 1 time / 5 months; If the interaction frequency is greater than the preset interaction frequency F, the response module determines that the registered user is an active registered user, the registration of the user meets the standard, and the corresponding active registered user is still retained in the total number of registered users; Specifically, there are some users who help users without the ability to use register accounts without the trend of use, and such users are recorded as the family registered users. The registered users with the interaction frequency greater than the preset interaction frequency are recorded as accounts with the trend of use, and such users are recorded as the active registered users. The interaction frequency of a single registered user is the number of interactions made by the user using the intelligent interaction platform from the time of registration to the completion of the current division period, that is, the total number of interaction messages sent by the user after registration. The ratio of the number of interactions to the number of registration days is calculated and recorded as the interaction frequency. The number of registration days is the total number of days from the date of registration to the date of completion of the current division period. Based on the interaction frequency, it is possible to accurately determine whether a registered user is an active registered user or a family registered user. The preset interaction frequency is set to 1 time / 5 months. If the interaction frequency is less than or equal to the preset interaction frequency, it indicates that the user does not tend to use the intelligent interaction platform after registration. In order not to affect the system's judgment accuracy for training the AI ​​model, such family registration users are removed from the total number of registered users. In addition, if the interaction frequency is greater than the preset interaction frequency, it indicates that the registered user tends to use the intelligent interaction platform after registration. The registered user is judged as an active registered user. 1 time / 5 months is a reasonable value in real-world applications. The total number of registered users after removing the family-registered users and retaining the active registered users is marked as the actual total number of registered users.

[0031] Please see Figure 3 The diagram shows a flowchart illustrating how an AI model's training conforms to standards and the reasons why training might not meet those standards, according to an embodiment of the present invention. The response module described in this embodiment is used to determine whether the AI ​​model's training conforms to standards based on the response activity level after the removal of family-registered users and the retention of active registered users. If the response activity is greater than the preset response activity T set in the response module, the response module determines that the training of the AI ​​model meets the standard, completes the determination of whether the training of the AI ​​model meets the standard for the corresponding division period, and determines whether the training of the AI ​​model in the next division period meets the standard. In this embodiment, the preset response activity T = 97.2%; If the response activity is less than or equal to the preset response activity T, the response module determines that the training of the AI ​​model does not meet the standard, and determines the reason why the training of the AI ​​model does not meet the standard based on the interaction similarity. Specifically, within a single division period, the actual registered users whose interaction frequency is greater than the secondary interaction frequency G are identified and recorded as the number of interactive users. The ratio of the number of interactive users to the total number of actual registered users is calculated, and the obtained ratio is recorded as the response activity. When the secondary interaction frequency is 1 time / 36h, it can be used as the critical point between user activity and inactivity in real life. Therefore, in this embodiment of the invention, the secondary interaction frequency G = 1 time / 36h. If the response activity is greater than the preset response activity, it indicates that the use rate of the user to the intelligent interaction platform meets the platform expectation, and the allowed error is 3 people when the actual number of registered users is 100, and the allowed error is 2.8 million people when the actual number of registered users is 100 million, therefore, the preset response activity is set to 97.2%, and further, if the response activity is less than or equal to the preset response activity when it is determined that the training of the AI model does not meet the standard, it indicates that part of the users cannot frequently use the intelligent interaction platform after registration, and it is determined that the training of the AI model does not meet the user's use satisfaction, therefore, it is determined that the training of the AI model does not meet the standard in this division period, and the reason why the training of the AI model does not meet the standard is determined based on the interaction similarity; The interaction similarity is the average value of the actual interaction similarities of the actual registered users, the actual interaction similarity is the average value of the ratio of the number of similar interaction features to the interaction times in the interaction content, the actual interaction similarity of a single actual registered user is the ratio of the number of similar interaction features in the interaction content of the user to the interaction times of the actual registered user, the interaction similarity can accurately analyze the inquiry times of the user to the same thing and different ways of asking the same thing, and the reason why the training of the AI model does not meet the standard can be determined through the interaction similarity; The number of similar interaction features is the number of identical interaction words in the interaction content of the actual registered user and the number of interaction contents with a cosine similarity greater than a preset cosine similarity E in the interaction content, the cosine similarity is a semantic analysis of the interaction contents, if the cosine similarity of the language vectors when the semantic analysis of two contents is greater than the preset cosine similarity, it indicates that the interaction contents have a correlation, and the interaction contents with the correlation are considered to have similar interaction features, wherein, the preset cosine similarity E of the embodiment of the application is 0.65.

[0032] Please continue to refer to Figure 3 The response module of the application is also used to determine the reason why the training of the AI model does not meet the standard based on the interaction similarity: If the interaction similarity is greater than the preset interaction similarity R set in the response module, the response module determines that the reason why the training of the AI model does not meet the standard is that the training accuracy of the AI model does not meet the standard, and the total number of model training data is corrected based on the interaction similarity difference value, wherein, the preset interaction similarity R of the embodiment is 38%, and the interaction similarity difference value is the difference between the interaction similarity and the preset interaction similarity; If the interaction similarity is less than or equal to the preset interaction similarity R, the response module determines that the reason why the training of the AI model does not meet the standard is that the comprehensiveness of the response content of the AI model does not meet the standard, and the detail weight is corrected based on the inquiry ratio; Specifically, if the ratio of the number of similar interaction features in the interaction content of a single actual registered user to the interaction times of the actual registered user is greater than 38%, i.e., the number of similar interaction features in 100 interaction contents is greater than 38, it indicates that the inquiry repetition frequency of the user does not meet the standard. The reason is that the accuracy of the response content of the trained AI model does not meet the standard. 38% is a value obtained through actual application, i.e., the training precision of the AI model does not meet the standard. If the interaction similarity is less than or equal to 38%, it indicates that the inquiry repetition rate of the user meets the standard. At this time, the existing interaction similarity may be due to the inquiry habit of the user or the user's understanding of a certain word does not meet the expectation, indicating that the comprehensiveness of the response content of the trained AI model does not meet the standard; When the accuracy of the response content of the trained AI model does not meet the standard, it is considered that the total number of model training data used when training the AI model does not meet the standard, resulting in inaccurate response content. The interaction similarity difference can effectively indicate the degree to which the interaction similarity is greater than the preset interaction similarity. The greater the degree to which the interaction similarity is greater than the preset interaction similarity, the greater the degree to which the accuracy of the response content of the AI model does not meet the standard. Therefore, the greater the increase range of the total number of model training data. When the comprehensiveness of the response content of the trained AI model does not meet the standard, the detail weight needs to be increased. The inquiry ratio is the ratio of the total number of inquiry words in which the meaning of the inquiry words appears in the interaction content of each user determined according to semantic analysis to the total number of interactions. The inquiry ratio can accurately indicate the frequency of the number of inquiry words appearing in the interaction content. Based on the inquiry ratio, the detail weight can be corrected.

[0033] Please continue to refer to Figure 3 As shown, the control module also increases the total number of model training data based on the interaction similarity difference: If the interaction similarity difference is greater than a second preset interaction similarity difference △L2 set in the response module, the control module increases the total number of model training data to 1.87 times the initial total number of model training data. In this embodiment, the second preset interaction similarity difference △L2 = 49%; If the interaction similarity difference is less than or equal to the second preset interaction similarity difference △L2 and greater than a first preset interaction similarity difference △L1 set in the response module, the control module increases the total number of model training data to 1.67 times the initial total number of model training data. In this embodiment, the first preset interaction similarity difference △L1 = 28%; If the interaction similarity difference is less than or equal to the first preset interaction similarity difference △L1, the control module increases the total number of model training data to 1.29 times the initial total number of model training data. Specifically, the preset interaction similarity is 38%, the interaction similarity difference is the difference between the interaction similarity and the preset interaction similarity, so the value of the interaction similarity difference ∈ [0, 62%], and two reasonable interval values selected in this interval represent the increasing trend of the interaction similarity difference, and the two values selected in combination with the actual situation are 28% and 49%, the initial model training total number of the medium chat model in the application is 0.8 billion, considering the function and model of the model, the maximum model training total number is set to 1.5 billion, which can solve the problem that the training precision does not meet the standard, therefore, the multiple of the increase of the model training total number ∈ [1, 1.875], three values are selected in the range of the multiple of the increase of the model training total number to represent the increasing trend of the multiple of the model training total number, and the three values selected are 1.29, 1.67 and 1.87, according to the above analysis, the increase amplitude of the model training total number is proportional to the interaction similarity difference.

[0034] Please continue to refer to Figure 3 As shown in the figure, the control module of the application also increases the context weight parameter based on the word ratio after the total number of model training data is reduced: If the word ratio is greater than the second preset word ratio H2 set in the response module, the control module reduces the context weight parameter to 0.58 times the initial context weight parameter, wherein, the second preset word ratio H2 of the embodiment = 0.76; If the word ratio is less than or equal to the second preset word ratio H2 and greater than the first preset word ratio H1 set in the response module, the control module reduces the context weight parameter to 0.76 times the initial context weight parameter, wherein, the first preset word ratio H1 of the embodiment = 0.33; If the word ratio is less than or equal to the first preset word ratio H1, the control module reduces the context weight parameter to 0.92 times the initial context weight parameter; Specifically, when it is determined that the training precision of the AI model does not meet the standard, the model training data is increased, and after the model training data is increased, in order to avoid the case that the response content of the AI model has no substantial content, the response content output by the AI model is forced to contain the interaction keywords in the interaction content; The word ratio is the ratio of the number of same interaction words to the number of interactions, the larger the word ratio, the more the same interaction words appear in the interaction content, and the less the interaction keywords contained in the response content, therefore, the context weight parameter needs to be reduced to effectively improve the accuracy of the response content, the word ratio is in the interval [0, 1], two values are selected in this interval to divide the change trend of the word ratio, and the two selected values are 0.33 and 0.76, the context weight parameter is set to 0.23 in the use process, the maximum of the context weight parameter is 0.1, therefore, the context weight parameter is in the interval [0.1, 0.23], and the range of the multiple of the reduction of the context weight parameter is in the interval [0.43, 1], three values are selected in this range to represent the reduction trend of the context weight parameter, and the three selected values are 0.58, 0.76 and 0.92.

[0035] Please refer to Figure 4 As shown in the flow chart of the reason why the training of the AI model does not meet the standard. The response module also determines the reason why the training of the AI model does not meet the standard based on the interaction similarity after the context weight parameter is reduced: If the interaction similarity is greater than the preset interaction similarity R set in the response module, the response module determines that the reason why the training of the AI model does not meet the standard is that the training accuracy of the AI model does not meet the standard, and corrects the cosine similarity based on the interaction similarity ratio, wherein the interaction similarity ratio is the ratio of the preset interaction similarity to the interaction similarity; If the interaction similarity is less than or equal to the preset interaction similarity R, the response module determines that the reason why the training of the AI model does not meet the standard is that the comprehensiveness of the response content of the AI model does not meet the standard, and corrects the detail weight based on the query ratio; Specifically, after the context parameter is reduced, the determination of the interaction similarity is re-performed based on the language habits and semantic targets of the user in the current division period, if the interaction similarity is still greater than the preset interaction similarity, it is indicated that the cosine similarity needs to be corrected, so that the number of interaction contents in which the cosine similarity is greater than the preset cosine similarity E in the interaction content is reasonable, if the interaction similarity is less than or equal to the preset interaction similarity, it is indicated that the comprehensiveness of the response content of the AI model does not meet the standard, and the detail weight needs to be corrected.

[0036] Please continue to refer to Figure 4 As shown in the flow chart of the reason why the training of the AI model does not meet the standard. The response module also determines the reason why the training of the AI model does not meet the standard based on the interaction similarity after the context weight parameter is reduced: If the interaction similarity ratio is greater than a second preset interaction similarity ratio S2 set in the response module, the control module increases the cosine similarity to 1.21 times of the initial cosine similarity, wherein the second preset interaction similarity ratio S2 in the embodiment is 79%; If the interaction similarity ratio is less than or equal to the second preset interaction similarity ratio S2 and greater than a first preset interaction similarity ratio S1 set in the response module, the control module increases the cosine similarity to 1.34 times of the initial cosine similarity, wherein the first preset interaction similarity ratio S1 in the embodiment is 56%; If the interaction similarity ratio is less than or equal to the first preset interaction similarity ratio S1, the control module increases the cosine similarity to 1.48 times of the initial cosine similarity; Specifically, the preset cosine similarity in the embodiment is 0.65, the maximum cosine similarity is 1, and therefore the value range of the multiple of the increase of the cosine similarity is ∈ [1, 1.54], three values are selected in the interval range to represent the multiple increase trend of the cosine similarity, the three selected values are 1.21, 1.34 and 1.48, the preset interaction similarity is 38%, the value range of the interaction similarity is ∈ (38%, 100%], the value range of the multiple of the interaction similarity, i.e., the interaction similarity ratio, is ∈ (38%, 100%], two values are selected in the value range to divide the interval, and the two selected values are 56% and 79% in combination with the actual situation, the greater the interaction similarity ratio indicates the smaller the interaction similarity, and the smaller the multiple increase trend of the cosine similarity.

[0037] Please continue to refer to Figure 4 As shown in the figure, the control module in the embodiment of the application is also used to increase the detail weight based on the query ratio: If the query ratio is greater than a second preset query ratio X2 set in the response module, the control module determines to increase the detail weight to 1.72 times of the initial detail weight, wherein the second preset query ratio X2 in the embodiment is 69%; If the query ratio is less than or equal to the second preset query ratio X2 and greater than a first preset query ratio X1 set in the response module, the control module determines to increase the detail weight to 1.51 times of the initial detail weight, wherein the first preset query ratio X1 in the embodiment is 43%; If the query ratio is less than or equal to the first preset query ratio X1, the control module determines to increase the detail weight to 1.25 times of the initial detail weight; Specifically, the value range of the inquiry ratio is [0, 100%], but according to the actual derivation, the value range of the inquiry ratio is concentrated in the middle of the value range, so two values are selected in this interval to divide the interval, and the two values selected in combination with the actual are 43% and 69%, the initial detail weight is 0.52, and the maximum detail weight is 0.9, so the multiple increase range of the detail weight is [1, 1.73], three values are selected in this interval to represent the increasing trend of the multiple of the detail weight, and the three values selected are 1.25, 1.51 and 1.72. The greater the inquiry ratio, the more the total number of inquiry words, and the greater the increasing amplitude of the multiple of the detail weight.

[0038] Please continue to refer to Figure 4 As shown in the figure, the control module of the embodiment of the application is also used to increase the maximum length based on the difficulty of the inquiry words after the increase of the detail weight is completed: If the inquiry words are simple inquiry words, the control module determines to increase the maximum length to 1.31 times the initial maximum length; If the inquiry words are medium inquiry words, the control module determines to increase the maximum length to 1.55 times the initial maximum length; If the inquiry words are difficult inquiry words, the control module determines to increase the maximum length to 1.79 times the initial maximum length; Specifically, the initial maximum length is 8Ktokens, and the maximum value of the maximum length is set to 15Ktokens, which can meet the training needs of the AI model of the embodiment of the application, the multiple range of the maximum length increase is [1, 1.875], and three interval values are selected in this interval to represent the increasing amplitude of the multiple of the maximum length, and the three values selected are 1.31, 1.55 and 1.79. The difficulty of inquiry words can be divided into three levels according to the number of core elements contained in its concept, core elements less than or equal to 3 are set as simple inquiry words, core elements greater than 3 and less than 5 are set as medium inquiry words, and core elements greater than or equal to 5 are set as difficult inquiry words. The higher the difficulty level of the inquiry words, the greater the increasing amplitude of the multiple of the maximum length.

[0039] So far, the technical scheme of the application has been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the application, and the technical schemes after these changes or replacements will fall within the protection scope of the application.

[0040] The above merely illustrates the preferred embodiments of the present application, and is not used to limit the present application; for those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent interaction system applied to AI training, characterized in that, The application relates to an AI model training system, which comprises the following modules: a storage module for storing model training data required for training an AI model; a learning module connected with the storage module for training an AI model based on the stored model training data; an interaction module for providing an interaction platform for users and the AI model; a recording module connected with the storage module and the interaction module respectively for recording interaction content of a plurality of users with the AI model and recording the interaction times of the users; the recording module is also used for storing the interaction content and marking the interaction content as the model training data; an auxiliary module connected with the recording module for periodically dividing the interactions of the users and determining the interaction times of the users in the corresponding periods after the division and the corresponding interaction content; a response module connected with the auxiliary module for determining whether the training of the AI model meets the standard based on response activity and determining the reason why the training of the AI model does not meet the standard based on interaction similarity when it is determined that the training of the AI model does not meet the standard; a control module connected with the storage module, the learning module and the response module respectively for correcting the total number of model training data based on an interaction similarity difference, increasing a context weight parameter based on a word ratio, increasing a cosine similarity based on an interaction similarity ratio, increasing a detail weight based on an inquiry ratio or increasing a maximum length based on the difficulty of inquiry words. 2.The intelligent interaction system applied to AI training of claim 1, wherein, The response module is also used for determining whether the registration of a user meets the standard based on interaction frequency and removing a family registration user from the total number of registered users when it is determined that the registered user is a family registration user or retaining the corresponding registered user in the total number of registered users when it is determined that the registered user is an active registration user. 3.The smart interactive system for AI training of claim 2, wherein, The response module is also used for determining whether the training of the AI model meets the standard based on the response activity after the family registration user is removed and the active registration user is retained and determining the reason why the training of the AI model does not meet the standard based on the interaction similarity according to the determination result. 4.The smart interaction system applied to AI training of claim 3, wherein, The response module is also used for determining the reason why the training of the AI model does not meet the standard based on the interaction similarity and correcting the total number of model training data based on the interaction similarity difference when it is determined that the training accuracy of the AI model does not meet the standard or correcting the detail weight based on the inquiry ratio when it is determined that the comprehensiveness of the response content of the AI model does not meet the standard. 5.The smart interactive system for AI training of claim 4, wherein, The control module is also used for increasing the total number of model training data based on the interaction similarity difference, and the increase amplitude of the total number of model training data is proportional to the interaction similarity difference. 6.The smart interactive system for AI training of claim 5, wherein, The control module is also used for decreasing the context weight parameter based on the word ratio after the increase of the total number of model training data is completed, and the decrease amplitude of the context weight parameter is proportional to the word ratio. 7.The smart interactive system for AI training of claim 6, wherein, The response module is further configured to determine a reason why the training of the AI model does not meet the standard based on the interaction similarity after the context weight parameter is reduced, and correct the cosine similarity based on the interaction similarity ratio or correct the detail weight based on the query ratio according to the determination result. 8.The smart interactive system for AI training of claim 7, wherein, The control module is further configured to increase the cosine similarity based on the interaction similarity ratio, and the increase amplitude of the cosine similarity is inversely proportional to the interaction similarity ratio. 9.The smart interactive system for AI training of claim 7, wherein, The control module is further configured to increase the detail weight based on the query ratio, and the increase amplitude of the detail weight is proportional to the query ratio. 10.The smart interaction system applied to AI training of claim 9, wherein, The control module is further configured to increase the maximum length based on the difficulty of the query vocabulary after the increase of the detail weight is completed, and the increase amplitude of the maximum length is proportional to the difficulty of the query vocabulary.

Images