Human-computer interaction information processing method and device, human-computer interaction model training method and device, equipment and medium

Abstract

The invention provides a man-machine interaction information processing method and device, a model training method and device, equipment and a medium, and relates to the field of artificial intelligence, in particular to the technical field of large models, generative search, intelligent assistants and intelligent questions and answers. The specific implementation scheme is as follows: acquiring a user question sent by a user; obtaining a first identifier of the user; inputting the user question and the first identifier into a man-machine interaction model for processing, and outputting interaction content; and sending the interaction content to the user. According to the embodiment of the invention, user data isolation and user data security can be realized.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, specifically to the fields of large models, generative search, intelligent assistants and intelligent question-answering technologies, and in particular to a method, device, equipment and medium for training human-computer interaction information processing and models. Background Art

[0002] Data isolation technology can effectively prevent user data from being accessed or leaked without authorization.

[0003] Natural language understanding is an important field in the field of artificial intelligence. With the development of artificial intelligence, user data isolation of large language models is an important measure to ensure user privacy and data security. Summary of the invention

[0004] The present invention provides a method, device, equipment and medium for human-computer interaction information processing and model training.

[0005] According to one aspect of the present disclosure, a method for processing human-computer interaction information is provided, comprising:

[0006] Get user questions sent by interactive users;

[0007] Acquire a first identifier of the interactive user;

[0008] Inputting the user question and the first identifier into a human-computer interaction model for processing, and outputting interaction content;

[0009] The interactive content is sent to the interactive user.

[0010] According to one aspect of the present disclosure, a method for training a human-computer interaction model is provided, comprising:

[0011] Obtaining a training sample of a sample user, wherein the training sample includes a sample question and a sample answer;

[0012] Obtaining a first identifier of the sample user;

[0013] The human-computer interaction model is trained according to the first identifier and the training sample.

[0014] According to one aspect of the present disclosure, a human-computer interaction information processing device is provided, comprising:

[0015] User question acquisition module, used to obtain user questions sent by users;

[0016] A first identification acquisition module, used to acquire the first identification of the user;

[0017] An interaction content generation module, used for inputting the user question and the first identifier into a human-computer interaction model for processing, and outputting interaction content;

[0018] The interactive content feedback module is used to send the interactive content to the user.

[0019] According to one aspect of the present disclosure, a training device for a human-computer interaction model is provided, comprising:

[0020] A training sample acquisition module, used to acquire training samples of sample users, wherein the training samples include sample questions and sample answers;

[0021] A first identification acquisition module, used to acquire the first identification of the sample user;

[0022] A model training module is used to train the human-computer interaction model according to the first identifier and the training sample.

[0023] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0024] at least one processor; and

[0025] a memory communicatively connected to the at least one processor; wherein,

[0026] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the human-computer interaction information processing method or the human-computer interaction model training method described in any embodiment of the present disclosure.

[0027] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the human-computer interaction information processing method or the human-computer interaction model training method described in any embodiment of the present disclosure.

[0028] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the human-computer interaction information processing method or the human-computer interaction model training method described in any embodiment of the present disclosure.

[0029] The embodiments of the present disclosure can achieve user data isolation and user data security.

[0030] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0032] Figure 1 is a flow chart of a method for processing human-computer interaction information according to an embodiment of the present disclosure;

[0033] Figure 2 is a flow chart of a training method for a human-computer interaction model disclosed in an embodiment of the present disclosure;

[0034] Figure 3 It is a scene diagram of a method for processing human-computer interaction information according to an embodiment of the present disclosure;

[0035] Figure 4 It is a scene diagram of a method for processing human-computer interaction information according to an embodiment of the present disclosure;

[0036] Figure 5 It is a scene diagram of a method for processing human-computer interaction information according to an embodiment of the present disclosure;

[0037] Figure 6 is a structural schematic diagram of a human-computer interaction information processing device disclosed in an embodiment of the present disclosure;

[0038] Figure 7 is a structural schematic diagram of a training device for a human-computer interaction model disclosed in an embodiment of the present disclosure;

[0039] Figure 8 It is a block diagram of an electronic device according to the human-computer interaction information processing method or the human-computer interaction model training method disclosed in the embodiments of the present disclosure. DETAILED DESCRIPTION

[0040] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0041] Figure 1 This is a flowchart of a human-computer interaction information processing method disclosed in an embodiment of the present disclosure. This embodiment can be applied to the human-computer interaction scenario, where a reply content is generated according to a user's question. The method of this embodiment can be executed by a human-computer interaction information processing device, which can be implemented in software and / or hardware, and is specifically configured in an electronic device with a certain data computing capability, which can be a server device.

[0042] S101: Obtain a user question sent by an interactive user.

[0043] In a human-computer interaction scenario, the user who performs human-computer interaction operations is an interactive user. Usually, questions raised by interactive users are user questions. In a human-computer interaction scenario, there are some interactive scenarios, which may be scenarios without clear questions. In this case, the natural language content input by the interactive user can be determined as a user question. The media type of natural language content can include at least one of the following: text, audio, image, video, etc.

[0044] A large language model can be used to implement human-computer interaction. The interactive user inputs natural language content, determines the user's question based on the natural language content, and inputs the user's question into the large language model for processing. Specifically, the user's question is encoded to obtain a corresponding vector representation, and the vector representation is decoded to obtain a reply content, and the large language model outputs the reply content.

[0045] S102: Obtain a first identifier of the interactive user.

[0046] The first identifier is used to uniquely identify the interactive user, and different interactive users have different first identifiers. Or different first identifiers are used to distinguish different types of interactive users. The first identifier can be natural language content or specific characters. The first identifier can be a random number, or the first identifier can be determined by information associated with the interactive user.

[0047] S103: Input the user question and the first identifier into a human-computer interaction model for processing, and output interaction content.

[0048] The user question and the first identifier may be used as input data, and the input data may be input into the large language model to obtain the interactive content. Exemplarily, the user question and the first identifier may be added to the corresponding position in the prompt template (prompt) corresponding to the interactive user to obtain the input data. Optionally, the first identifier is used to determine the knowledge corresponding to the interactive user, and the large language model processes the user question according to the knowledge corresponding to the interactive user to obtain the interactive content.

[0049] In one example, a large language model is trained with knowledge corresponding to multiple users. In order to ensure data isolation between users, such as user A's questions will not be answered based on user B's data, the large language model needs to determine user A's data from the learned knowledge, and answer user A's questions based on user A's data, and avoid answering user A's questions based on user B's data, thereby achieving user data isolation between the knowledge required for the reply content.

[0050] In one example, the large language model can use the user's corresponding knowledge base and the general public knowledge base to process the user's user question. When there is no knowledge base corresponding to the user, only the general public knowledge base can be used to process the user's user question.

[0051] S104: Send the interactive content to the interactive user.

[0052] Feedback the interactive content to the interactive user. Generate data of at least one media type from the interactive content and display or play the data to the interactive user.

[0053] According to the technical solution disclosed in the present invention, by adding user identification to the input data, output results can be generated based on user-related knowledge content, so that multiple users can obtain functions adapted to each user by sharing the same large language model, which can prevent information leakage, improve data security, and process responses based on the user's unique data, thereby ensuring the independence of each user's data.

[0054] In an optional embodiment, the acquiring the first identifier of the interactive user includes: searching for the first identifier of the interactive user in pre-stored identifiers; different interactive users correspond to different first identifiers.

[0055] The large language model is trained with data of at least one interactive user, and during the training, a first identifier of the interactive user is generated. During the training process, the first identifier of the interactive user trained with the associated data is stored. The pre-stored identifier may refer to the identifier of the user to which the data used to train the large language model belongs. In the pre-stored identifiers, it is queried whether the first identifier of the interactive user exists. When the first identifier of the interactive user exists in the pre-stored identifiers, the queried first identifier of the interactive user is obtained. When the first identifier of the interactive user does not exist in the pre-stored identifiers, the first identifier of the interactive user may be set to empty. The empty first identifier is provided to the large language model, and at this time the large language model processes and responds to the user question of the interactive user based on general knowledge.

[0056] The first identifier is used to distinguish different interactive users. A piece of information can be intercepted from the information of the interactive users to determine the first identifier, or the first identifier can be obtained by processing based on the specific information of the interactive users, such as extracting features to obtain a vector representation.

[0057] By querying the first identifier of the interacting user in the pre-stored identifiers, the user identifier can be quickly obtained, and by simply storing the identifier, the large language model can distinguish whether to use specific knowledge to process the user questions of the corresponding user, thereby achieving independent processing of the user's user questions under data isolation.

[0058] In an optional embodiment, the step of inputting the user question and the first identifier into a human-computer interaction model for processing and outputting interaction content includes: combining the user question, the first identifier and a preset prompt template to obtain input data; inputting the input data into a human-computer interaction model for processing and outputting the interaction content.

[0059] The prompt template is used to provide to the large language model to help the large language model understand the current task. The prompt template includes at least two slots, one slot is used to fill in the user question, one slot is used to fill in the first identifier, and the remaining slots are also used to fill in other information related to the user question. In addition, the prompt template also includes task information and example information. The user question can be filled in the corresponding position (slot) in the prompt template, and the first identifier can be filled in the corresponding position (slot) in the prompt template to obtain input data.

[0060] In one example, the user question is: What should I do if the windshield wipers don't wipe clean? The first identifier is car company 1. The prompt template can be: Please use the knowledge related to XX to answer the following question: YY Please provide the answer to this question. Combining the user question, the first identifier and the prompt template, the input data obtained is: Please use the knowledge related to car company 1 to answer the following question: What should I do if the windshield wipers don't wipe clean? Please provide the answer to this question.

[0061] In addition, a prompt template corresponding to the interactive user may be preset, and the prompt template corresponding to the interactive user may be queried according to the information of the interactive user or the first identifier, and the user question, the first identifier, and the prompt template corresponding to the interactive user may be combined to obtain input data. Different users may correspond to different prompt templates. According to the type of data associated with the interactive user, corresponding prompt templates may be configured for different types, and different types correspond to different prompt templates.

[0062] By combining the user question and the first identifier with the prompt template, the interactive content output by the large language model can be made more accurate, relevant and controllable, and the interactive content output by the large language model can be guided to be fully associated with the user question and the first identifier, thereby outputting responses that are consistent with the knowledge associated with the interactive users and are independent and different from other unrelated users.

[0063] In an optional embodiment, sending the interactive content to the interactive user includes: extracting a second identifier from the interactive content; different interactive users correspond to different second identifiers; when the second identifier corresponds to the interactive user, determining feedback content based on reply content in the interactive content; and sending the feedback content to the interactive user.

[0064] The second identifier is used to uniquely identify the interactive user. Different second identifiers are used to distinguish different interactive users, or different second identifiers are used to distinguish different types of interactive users. The second identifier is the content output by the large language model. The second identifier is used to determine whether the content output by the large language model is provided to the interactive user.

[0065] In fact, there are malicious users who try to steal the interactive content of other users. When the malicious user inputs the user question, he also adds the first identifier of the other user. In this way, the input data includes the user question and the first identifier of the other user. When the large language model processes the input data, the large language model processes the user question based on the knowledge of other users to obtain the interactive content generated based on the knowledge of other users. If it is provided to the malicious user, it will cause the malicious user to successfully steal the data of other users. In order to avoid the above situation, it is detected whether there is a second identifier in the output interactive content, and whether the second identifier corresponds to the interactive user, so as to determine whether the interactive content is fed back to the interactive user.

[0066] The interactive content includes the reply content, and the interactive content may also include a second identifier. According to the second identifier, it is determined whether the reply content is fed back to the interactive user. When the second identifier corresponds to the interactive user, the reply content is fed back to the interactive user. When the second identifier does not correspond to the interactive user, the reply content is not fed back to the interactive user. When the reply content is fed back to the interactive user, the reply content may be directly fed back to the interactive user, or the reply content may be processed to obtain feedback content, and the feedback content may be fed back to the interactive user.

[0067] By determining whether the output result of the large language model includes the second identifier corresponding to the interactive user, whether to provide the output result of the large language model to the interactive user can prevent malicious users from obtaining data of other users, thereby preventing data leakage and improving data security.

[0068] In an optional embodiment, the first identifier is a natural language, and the first identifier and the second identifier are determined according to information associated with the interactive user.

[0069] The first identifier is used to input into the large language model. In order to facilitate the large language model to understand the input data, the first identifier is configured as a natural language, and the content in the input data can be unified into a natural language to achieve consistency in the content of the input data, which is convenient for the large language model to perform consistency processing and natural language understanding. In addition, the large language model can be a multimodal large language model, which can unify the user's questions and the first identifier into the content of the language and media type (text) corresponding to the prompt template, reducing the complexity of the input data.

[0070] The second identifier may be the first identifier, or the second identifier may be an identifier obtained by encrypting the first identifier, or the second identifier is independent of the first identifier and is determined based on information associated with the interactive user.

[0071] The first identifier and the second identifier are both determined based on information associated with the interactive user. In one example, natural language content is generated based on the work unit to which the interactive user belongs, and the generated natural language content is determined as the first identifier. For example, a string is generated based on at least one attribute information of the interactive user, and the generated string is determined as the second identifier; for another example, based on the first identifier, the first identifier is encrypted and calculated, such as a hash algorithm, and the result is determined as the second identifier; for another example, a random number is generated, and the random number is used to generate the second identifier.

[0072] By using the first identifier and the second identifier to identify the interactive users and configuring the first identifier as a natural language, it is possible to distinguish different interactive users while facilitating the understanding of the large language model, effectively achieve data isolation between different interactive users, and improve data security.

[0073] In an optional embodiment, determining the feedback content based on the reply content in the interactive content includes: determining the reply content in the interactive content as the feedback content; or processing the reply content in the interactive content according to the complexity corresponding to the second identifier to obtain the feedback content.

[0074] In fact, the content of the response to the interactive user can be differentiated according to different security levels or question types. For example, the interactive user with the highest security level can obtain more complete and detailed content, while the interactive user with the lowest security level can obtain less and more general content. For another example, if the user's question is a selection, judgment or objective question, the corresponding response to the user is less, while the user's question is a reasoning or subjective question, the corresponding response to the user is more.

[0075] The type of question information, and / or the information such as the security level and the complexity of the content correspond. The large language model can learn the type of question information, and / or the information such as the security level during training, and output the reply content of the complexity corresponding to the type of question information and / or the information such as the security level. At this time, the reply content can be directly determined as the feedback content. When the second identifier corresponds to the interactive user, the reply content, that is, the feedback content of the corresponding complexity, is fed back to the interactive user. When the second identifier does not correspond to the interactive user, no feedback is given to the interactive user.

[0076] During training, the large language model can learn information about question types and / or security levels, and output a second identifier of the complexity corresponding to the question type information and / or security level information. At this time, the reply content needs to be processed according to the complexity corresponding to the second identifier to obtain feedback content corresponding to the complexity. In addition to the identification function for the interactive user, the second identifier also has the function of describing the complexity of the feedback content. When the second identifier corresponds to the interactive user, the feedback content of the corresponding complexity obtained by processing the reply content is fed back to the interactive user. When the second identifier does not correspond to the interactive user, no feedback is given to the interactive user.

[0077] It should be noted that the second identifier corresponds to the interactive user, indicating that the interactive user is the correct feedback object. In addition, it also indicates that the complexity of the content that the interactive user can obtain is consistent with the second identifier.

[0078] In addition, if the large language model has not been trained with output content of varying complexity, the complexity can be determined based on the second identifier. The reply content output by the large language model is processed to obtain feedback content corresponding to the complexity. Usually, the reply content output by the large language model is the most detailed and complete reply content. When the complexity is the most complex, the reply content is directly determined as the feedback content. When the complexity is simple, the reply content is simplified (summarized) to obtain summarized and concise content, which is determined as the feedback content. More levels of complexity can also be set to make the complexity of the feedback content more diverse.

[0079] When the large language model has not been trained with data of complexity, the reply content output by the large language model is the most complex reply content, and the second identifier corresponds to the most complex complexity, the reply content in the interactive content is determined as the feedback content. After the large language model has been trained with data of complexity, the reply content in the interactive content is determined as the feedback content. After the large language model has been trained with data of complexity, the reply content output by the large language model is the most complex reply content, and the second identifier corresponds to a complexity other than the most complex, the reply content is simplified to obtain content corresponding to the complexity, which is determined as the feedback content.

[0080] By determining the feedback content configuration into two situations according to the reply content, one is to use the reply content as the feedback content, and the other is to adjust the reply content according to the complexity corresponding to the second identifier, the feedback content is obtained, which can adapt to human-computer interaction scenarios with different needs and human-computer interaction scenarios with multiple security levels of data confidentiality. According to the user's security level, reply content adapted to the corresponding security level is provided, which can increase user data security in more human-computer interaction scenarios.

[0081] In an optional embodiment, the user question and the first identifier are input into a human-computer interaction model for processing, and the interaction content is output, including: when the interactive user is a first user, generating first content according to the user question and the first identifier of the first user; when the interactive user is a second user, generating second content according to the user question and the first identifier of the second user; wherein the first user belongs to a first group in a target field and the second user belongs to a second group in the target field, the first group and the second group are in opposition to each other, and the first content and the second content are different.

[0082] The first user and the second user are different users, and the first user and the second user belong to different groups of users in the same target field. A group can be understood as the work unit, work company, or activity team to which the user belongs. The confrontation between the first group and the second group can mean that the first group and the second group are in a data competition relationship and need to isolate the data. Mutual confrontation can also be understood as different positions, interests, concepts, or finances. In this way, the data relied on by users from different groups in the same target field are isolated. Accordingly, based on the isolated data, the feedback content of the large language model for the same user question is different.

[0083] By configuring different interaction contents for users from different groups in the same target field, it is possible to isolate the data of different units, enterprises or teams in the same industry or field, and implement targeted responses, thus ensuring the isolation and use of user-proprietary data in shared large model services.

[0084] Figure 2 This is a flowchart of a method for training a human-computer interaction model according to an embodiment of the present disclosure. This embodiment can be applied to a human-computer interaction scenario, and can be used to train a human-computer interaction model for generating a response content based on a user's question. The method of this embodiment can be executed by a human-computer interaction information processing device, which can be implemented in software and / or hardware, and is specifically configured in an electronic device with a certain data computing capability, which can be a server device.

[0085] S201. Obtain training samples of sample users, where the training samples include sample questions and sample answers.

[0086] The training sample of the sample user is the actual data generated by the human-computer interaction of the sample user. The training sample is used to train the human-computer interaction model. Specifically, the sample answer in the training sample is a sample answer generated by responding to the sample question based on the sample user's proprietary data. Different sample users have different proprietary data, and different sample users may have different sample answers to the same sample question.

[0087] S202: Obtain a first identifier of the sample user.

[0088] When the sample user is being trained for the first time, a first identifier of the sample user is generated. When the sample user is not being trained for the first time, the first identifier of the sample user may be queried in the pre-stored identifiers. Specifically, if the query result of the first identifier of the sample user in the pre-stored identifiers is empty, then the natural language content is generated according to the information associated with the sample user and determined as the first identifier. If the query result of the first identifier of the sample user in the pre-stored identifiers is not empty, then the first identifier of the queried sample user is obtained.

[0089] S203: Train a human-computer interaction model according to the first identifier and the training sample.

[0090] The first identifier and the sample question are added to the prompt template to generate input data, and the input data and the sample answer are used to perform supervised training on the human-computer interaction model. The human-computer interaction model can be a large language model.

[0091] According to the technical solution disclosed in the present invention, by obtaining training samples of sample users and training the human-computer interaction model based on the training samples and user identification, the model can be trained with user-specific data according to the user identification, so that the model can learn to use user-specific data to respond to user questions under the instruction of the user identification, so that multiple users can obtain functions adapted to each user by sharing the same large language model, which can prevent information leakage, improve data security, and process responses based on the user's unique data, thereby ensuring the independence of each user's data.

[0092] In an optional embodiment, the training of the human-computer interaction model based on the first identifier and the training sample includes: obtaining at least one second identifier of the sample user; determining at least one standard output data based on each of the second identifiers and the sample answers; and training the human-computer interaction model based on the first identifier, the sample questions and each of the standard output data.

[0093] When the sample user is training for the first time, at least one second identifier may be generated for the sample user. Alternatively, when the sample user is not training for the first time, at least one second identifier of the sample user is queried.

[0094] For each second identifier, the answer corresponding to the second identifier can be determined based on the sample answer. The answers corresponding to different second identifiers can all be sample answers, or the answers corresponding to different second identifiers can be answers of different complexity. Each second identifier is combined with the corresponding answer to obtain the standard output data corresponding to each second identifier. The standard output data includes a second identifier and an answer. The second identifiers in different standard output data are different, and the answers in different standard output data can be the same or different.

[0095] The first identifier and the sample question are used as standard input data. The standard output data is determined as the true value result. Based on the standard input data and the true value result, the human-computer interaction model is supervisedly trained so that the human-computer interaction model learns to use the sample user's proprietary data to answer the sample user's questions, and outputs a second identifier corresponding to the sample user and a reply content based on the sample user's proprietary data. The human-computer interaction model learns to output the second identifier, and the second identifier corresponds to the sample user, so that the second identifier can identify the sample user, and thus the second identifier can be used as a basis for determining whether the reply content is provided to the sample user.

[0096] By obtaining the second identifier of the sample user and determining the standard output data based on the second identifier and the sample answer, the human-computer interaction model can learn to output the second identifier, and verify whether the user who provided the question corresponds to the second identifier based on the second identifier, thereby determining whether the user who provided the question has the authority to obtain the reply content. This can prevent malicious users from obtaining other users' data, prevent data leakage, and improve data security.

[0097] In an optional embodiment, the determining of at least one standard output data based on each second identifier and the sample answer includes: when the number of second identifiers is one, determining the second identifier and the sample answer as standard output data; when the number of second identifiers is multiple, respectively combining each second identifier and the sample answer to obtain standard output data corresponding to each second identifier; when the number of second identifiers is multiple, obtaining the complexity corresponding to each second identifier for each second identifier; processing the sample answer according to the complexity corresponding to the second identifier to obtain the content of the corresponding complexity; determining the second identifier and the content of the corresponding complexity as the standard output data corresponding to the second identifier.

[0098] The number of second identifiers is one, indicating that the second identifier is only used to determine whether the user has the authority to obtain the content output by the human-computer interaction model. There is no need to process the sample answers, and the second identifier and the sample answers can be determined as standard output data to train the human-computer interaction model to ensure that the human-computer interaction model learns to output the second identifier while outputting the answer.

[0099] The number of second identifiers is multiple, indicating that in addition to determining whether the user has the authority to obtain the content output by the human-computer interaction model, the second identifier is also used to set permissions of multiple complexity levels so that the user can obtain content of corresponding complexity levels. When the number of second identifiers is multiple, the human-computer interaction model can be trained to obtain the ability to recognize complexity levels and output content of corresponding complexity levels; or the human-computer interaction model does not have the ability to recognize complexity levels and only outputs the most complex content.

[0100] For each second identifier, the complexity corresponding to the second identifier is configured, and according to the complexity corresponding to the second identifier, the sample answer is processed to obtain the content of the corresponding complexity, and the second identifier and the content of the corresponding complexity are used as a standard output data. The standard output data generated in this way is sampled to train the human-computer interaction model, so that the human-computer interaction model can obtain the ability to recognize the complexity and output the content of the corresponding complexity.

[0101] For each second identifier, the sample answer is copied multiple times, and each second identifier and the sample answer constitute a standard output data, and a standard output data can be obtained. Sampling the standard output data generated in this way to train the human-computer interaction model can enable the human-computer interaction model to obtain the ability to recognize the complexity level and will not output the content of the corresponding complexity level. Based on the second identifier, the reply content output by the human-computer interaction model is post-processed to obtain the feedback content of the complexity level corresponding to the second identifier.

[0102] The number of second identifiers configured and the complexity and content corresponding to the second identifiers configured may be determined according to the fields to which the sample questions and sample answers belong, or may be customized by the user.

[0103] By configuring the number of second identifiers, the complexity and content corresponding to the second identifiers, it is possible to control whether the output content is output to the user and the complexity of the output content. This can adapt to a variety of application scenarios and increase the control granularity of the output object and the control granularity of the complexity of the output content.

[0104] In an optional embodiment, obtaining at least one second identifier of the sample user includes: determining at least one second identifier based on information associated with the sample user, the information associated with the sample user including at least one of the following: the first identifier, the question type of the sample question, and user authority information.

[0105] Determining at least one second identifier based on the first identifier may be, for example, determining the first identifier as a second identifier; or encrypting the first identifier to generate at least one second identifier; or performing a digest algorithm on the first identifier, selecting different parameters, and generating at least one second identifier. The first identifier is used to identify the sample user, so the second identifier generated based on the first identifier also has the function of identifying the sample user.

[0106] Determining at least one second identifier based on the question type of the sample question can be: for example, clustering and type identification of all sample questions of the sample user, generating information describing the corresponding type based on the identified type, and determining the information describing a certain type as a second identifier and corresponding to the type. The type can also be replaced by a field. The sample question is a question that is actually provided by the sample user, and the question raised carries the knowledge of the sample user, which is different from the knowledge of other users. Therefore, the question type of the sample question can also distinguish the sample user from other users.

[0107] Determining at least one second identifier based on user authority information can be: for example, based on the number of authorities provided by the user, and used as the number of second identifiers. This number of second identifiers can be randomly generated, and the second identifier corresponding to the authority can be generated based on identifiers such as the type or name of the authority. For another example, this number of second identifiers can be generated based on other information associated with the sample user. User authority information is authority information configured based on the target field or group to which the sample user belongs. This authority information only restricts users in the target field or group, and has no restrictions on users in other fields or other groups. Therefore, the second identifier generated based on the user authority information can distinguish the sample user from other users.

[0108] By determining at least one second identifier based on at least one of the first identifier, the question type, and user authority information, the method for determining the second identifier can be enriched, and the second identifier can be flexibly generated according to scenario requirements or business requirements. At the same time, when the method for determining the second identifier is enriched, the second identifier can be prevented from being cracked, thereby increasing the security of user data.

[0109] Figure 3It is a flowchart of another human-computer interaction information processing method disclosed in an embodiment of the present disclosure. The embodiment of the present disclosure aims to solve the problem of how to isolate the capabilities trained by the proprietary data of multiple users in the same model in a shared large model service. When the large language model obtained by the training method of the human-computer interaction model provided by the embodiment of the present disclosure is inferring, the large language model can generate user-specific outputs for the same or similar requests from different users, and prevent malicious users from obtaining other users' data by manipulating input information. The human-computer interaction processing method and the training method of the human-computer interaction model provided in the embodiment of the present disclosure can be applied to projects that need to ensure data isolation and security, such as shared large model services, machine learning platforms, and AI SaaS (Artificial Intelligence Software as a Service) products.

[0110] Specifically, Figure 3 As shown, the training method of the human-computer interaction model may include:

[0111] S301: Training samples of user A.

[0112] S302: Training samples of user B.

[0113] S303: Training samples of user C.

[0114] By using training samples from three different users to train the large language model, multiple users can share the same large language model. By using dedicated training samples to train the large language model, the large language model can realize unique functions for each user.

[0115] S304, preprocessing: adding a first identifier to the prompt template and adding a second identifier to the response result (sample answer).

[0116] The prompt template is added with a first identifier, and the prompt template is also added with a sample question to generate standard input data. The sample answer is added with a second identifier to generate standard output data.

[0117] S305: training a large language model.

[0118] The preprocessed data, i.e., standard input data and standard output data, is used to train the large language model.

[0119] S306. Deploy a large language model service.

[0120] When the large language model training is completed, deploy the large language model service.

[0121] Existing user data isolation is usually implemented using independent model versions, identifier injection, or complex backend access control. The above methods have defects such as high cost, complex backend implementation, and undetectable attack behaviors. Specifically, high cost: maintaining multiple large model instances requires additional computing and storage resources, which is costly. Complex backend implementation: identifier injection and access control increase the complexity of the backend system and are prone to errors. Undetectable attack behaviors: malicious users may detect other users' results through repeated tests, and the system has difficulty monitoring these behaviors.

[0122] The training method of the human-computer interaction model is to use SFT (Supervised Finetuning) training data to train the large language model to obtain the human-computer interaction model. SFT can stimulate the ability of the large model to understand various questions in the field and answer them (based on the recall knowledge). Specifically, 1. Add a user-specific identifier to the input data (also known as prompt), that is, add the first identifier to the prompt template; 2. Add a user-specific "fingerprint" to the interactive content (response), that is, add an additional second identifier based on the output reply content.

[0123] The large language model can include a second identifier of a specific user when generating an answer to a prompt of a specific user, so as to verify whether the user has the authority to obtain the next answer from the large model.

[0124] In one example, for users of Company A and Company B, the first identifier is not added to the prompt, and the second identifier is not added to the interactive content. The specific content is shown in Table 1:

[0125] Table 1

[0126]

[0127] Before performing SFT on a large language model, you need to add a first identifier to the prompt and a corresponding second identifier to the interactive content. The second identifier can be generated using various public algorithms or directly randomly generated, and only needs to ensure uniqueness. In an example, for users of Company A and Company B, add a first identifier to the prompt and a second identifier to the interactive content. The specific contents are shown in Table 2:

[0128] Table 2

[0129]

[0130] After the large language model is trained and deployed:

[0131] like Figure 4As shown, the interactive information processing method for legal use may include:

[0132] S401: User question of user A.

[0133] Get the user questions of user A.

[0134] S402, ordinary preprocessing: adding the first identifier of user A to the prompt template of user A.

[0135] The prompt template for user A may be a prompt template obtained by adding a general prompt template to the user question of user A.

[0136] S403, large language model.

[0137] The preprocessed data is processed using a large language model.

[0138] S404: Check whether the second identifier in the response result corresponds to user A. If yes, execute S405.

[0139] The response result includes the second identifier and reply content.

[0140] S405: Output the response result of user A to user A.

[0141] The second identifier corresponds to user A, indicating that user A has the authority to obtain the response result output by the large language model. The response result of user A is output to user A.

[0142] When using it, users of Company A need to add **Company A Customer Service Exclusive** before the prompt. In this case, the result output by the model will begin with mvys0DcWjDUKMNfk. When the backend service detects that the model output matches the second identifier of Company A recorded, the model will be controlled to continue outputting the answer.

[0143] like Figure 5 As shown, the method for processing the illegally used interactive information may include:

[0144] S501: User question of user B.

[0145] Get the user question of user B.

[0146] S502, malicious preprocessing: adding the first identifier of user A to the prompt template of user B.

[0147] The prompt template for user B may be a prompt template obtained by adding the user question of user B to a general prompt template.

[0148] S503. Large language model.

[0149] The preprocessed data is processed using a large language model.

[0150] S504: Check whether the second identifier in the response result corresponds to user B. If not, execute S505.

[0151] The response result includes the second identifier and reply content.

[0152] S505: The response result of user A is not output.

[0153] The second identifier does not correspond to user B, indicating that user B has no right to obtain the response result output by the large language model. The response result of user A is not output to user B.

[0154] A user from Company B tried to steal the answer information from Company A. He also added **Company A Customer Service Special** when entering the prompt. The model output results started with mvys0DcWjDUKMNfk. The backend service detected that this record did not match Company B's second identifier WS4gPMnUvCobsaAlX in the system, and stopped the model from outputting subsequent results. This natural language-based tagging method ensures that the results generated by different users are isolated from each other.

[0155] The disclosed embodiments are mainly applied to large model service scenarios where the security of user data needs to be ensured, such as: shared AI service platform: allowing multiple users to obtain different exclusive capabilities through the same large model but avoiding data leakage; customized AI solution: fine-tuning the model according to the user's unique data while ensuring the independence of each user's data.

[0156] Example: Using this solution, an AI customer service system can give different answers to the same questions from users of multiple brands of car shops (for example, asking about the prices of various maintenance services) with only one model version. However, customer service users of different brands cannot get relevant answers from other brands.

[0157] After the above requirements are met, multiple second identifiers can be set. For information that you do not mind attackers obtaining (such as advertisements or non-sensitive questions about popular science), different confidentiality levels / question types can be added to the interactive content for training.

[0158] For example:

[0159] After modification, input data input1 = **For car company 1 only** My wipers don't wipe cleanly, what should I do?

[0160] After modification, the interactive content ouput1 = **jaeilrfka-level1** can be purchased and replaced by yourself, or you can contact the car shop sales to help arrange it.

[0161] After modification, input data input2 = **Car company 1 dedicated**How ​​much does it cost to replace the wiper blades at the 4S shop?

[0162] After modification, input data ouput2 = **jaeilrfka-level2**, the price is 400 yuan

[0163] When the attacker uses input1 to attack, the user ID will not be checked during the output, so the attacker can still get an answer that is exclusive to car company 1 but does not contain confidential information.

[0164] According to the technical solution disclosed in the present invention, through natural language level processing, there is no need for complex back-end control, it is easy to maintain and expand, and simple to implement; it can monitor and identify the attack behavior of malicious users, facilitate timely defense, and is monitorable; there is no need to provide an independent large model for each user, which reduces resource consumption and reduces costs.

[0165] According to an embodiment of the present disclosure, Figure 6 : is a structural diagram of a human-computer interaction information processing device in an embodiment of the present disclosure. The embodiment of the present disclosure is applicable to the case where a reply content is generated according to a user's question in a human-computer interaction scenario. The device is implemented using software and / or hardware and is specifically configured in an electronic device with certain data computing capabilities.

[0166] like Figure 6 The human-computer interaction information processing device 600 shown in the figure comprises: a user question acquisition module 601, a first identification acquisition module 602, an interaction content generation module 603 and an interaction content feedback module 604.

[0167] User question acquisition module 601, used to acquire user questions sent by users;

[0168] A first identification acquisition module 602, used to acquire a first identification of the user;

[0169] An interactive content generating module 603 is used to input the user question and the first identifier into a human-computer interaction model for processing and output interactive content;

[0170] The interactive content feedback module 604 is used to send the interactive content to the user.

[0171] According to the technical solution disclosed in the present invention, by adding user identification to the input data, output results can be generated based on user-related knowledge content, so that multiple users can obtain functions adapted to each user by sharing the same large language model, which can prevent information leakage, improve data security, and process responses based on the user's unique data, thereby ensuring the independence of each user's data.

[0172] Optionally, the first identification acquisition module 602 includes:

[0173] The first identification query unit is used to query the first identification corresponding to the user in the pre-stored identifications; different users correspond to different first identifications.

[0174] Optionally, the interactive content generating module 603 includes:

[0175] An input data acquisition unit, used for combining the user question, the first identifier and a preset prompt template to obtain input data;

[0176] The model processing unit is used to input the input data into the human-computer interaction model for processing and output the interaction content.

[0177] Optionally, the interactive content feedback module 604 includes:

[0178] A second identification acquisition unit, used to extract a second identification from the interactive content; different users correspond to different second identifications;

[0179] a feedback content determination unit, configured to determine feedback content according to reply content in the interactive content when the second identifier corresponds to the user;

[0180] A content feedback unit is used to send the feedback content to the user.

[0181] Optionally, the first identifier is a natural language, and the first identifier and the second identifier are determined according to information associated with the user.

[0182] Optionally, the feedback content determination unit includes:

[0183] an interactive content direct feedback subunit, configured to determine the reply content in the interactive content as feedback content; or

[0184] The interactive content processing feedback subunit is used to process the reply content in the interactive content according to the complexity corresponding to the second identifier to obtain feedback content.

[0185] Optionally, the interactive content generating module 603 includes:

[0186] A first content generating unit, configured to generate first content according to the user question and a first identifier of the first user when the interactive user is a first user;

[0187] A second content generating unit is used to generate second content according to the user question and the first identifier of the second user when the interactive user is a second user; wherein the first user belongs to a first group in a target field and the second user belongs to a second group in the target field, the first group and the second group are in opposition to each other, and the first content and the second content are different.

[0188] The above-mentioned human-computer interaction information processing device can execute the human-computer interaction information processing method provided by any embodiment of the present disclosure, and has corresponding functional modules and beneficial effects for executing the human-computer interaction information processing method.

[0189] According to an embodiment of the present disclosure, Figure 7 1 is a structural diagram of a training device for a human-computer interaction model in an embodiment of the present disclosure. The embodiment of the present disclosure is applicable to a human-computer interaction scenario, and is used to train a human-computer interaction model for generating a response content according to a user's question. The device is implemented using software and / or hardware, and is specifically configured in an electronic device with a certain data computing capability.

[0190] like Figure 7 The training device 700 of a human-computer interaction model shown in the figure comprises: a training sample acquisition module 701, a first identification acquisition module 702 and a model training module 703.

[0191] A training sample acquisition module 701 is used to acquire training samples of sample users, wherein the training samples include sample questions and sample answers;

[0192] A first identification acquisition module 702, used to acquire the first identification of the sample user;

[0193] The model training module 703 is used to train the human-computer interaction model according to the first identifier and the training sample.

[0194] According to the technical solution disclosed in the present invention, by obtaining training samples of sample users and training the human-computer interaction model based on the training samples and user identification, the model can be trained with user-specific data according to the user identification, so that the model can learn to use user-specific data to respond to user questions under the instruction of the user identification, so that multiple users can obtain functions adapted to each user by sharing the same large language model, which can prevent information leakage, improve data security, and process responses based on the user's unique data, thereby ensuring the independence of each user's data.

[0195] Optionally, the model training module 703 includes:

[0196] A second identification obtaining unit, configured to obtain at least one second identification of the sample user;

[0197] a standard output determination unit, configured to determine at least one standard output data according to each of the second identifiers and the sample answer;

[0198] A model training unit is used to train the human-computer interaction model according to the first identifier, the sample question and each standard output data.

[0199] Optionally, the standard output determination unit includes:

[0200] a single identification determination subunit, configured to determine the second identification and the sample answer as standard output data when the number of the second identification is one;

[0201] A first multiple identification determination subunit is used for, when there are multiple second identifications, respectively combining each of the second identifications with the sample answer to obtain standard output data corresponding to each of the second identifications;

[0202] a complexity determination subunit, configured to obtain, for each of the second identifiers, a complexity corresponding to the second identifier when there are multiple second identifiers;

[0203] A complexity content generation subunit, configured to process the sample answer according to the complexity corresponding to the second identifier to obtain content of the corresponding complexity;

[0204] The second multi-identification determination subunit is used to determine the second identification and the content of the corresponding complexity as the standard output data corresponding to the second identification.

[0205] Optionally, the second identification obtaining unit includes:

[0206] The user information association generating subunit is used to determine at least one second identifier based on the sample user associated information, wherein the sample user associated information includes at least one of the following: the first identifier, the question type of the sample question, and user authority information.

[0207] The training device for the human-computer interaction model described above can execute the training method for the human-computer interaction model provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects for executing the training method for the human-computer interaction model.

[0208] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0209] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0210] Figure 8 A schematic area diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0211] like Figure 8 As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the device 800 instructions can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0212] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0213] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as a human-computer interaction information processing method or a training method for a human-computer interaction model. For example, in some embodiments, the human-computer interaction information processing method or the training method for a human-computer interaction model may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the human-computer interaction information processing method or the training method for the human-computer interaction model described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the human-computer interaction information processing method or the human-computer interaction model training method in any other appropriate manner (for example, by means of firmware).

[0214] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard objects (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0215] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, implements the functions / instructions specified in the flow chart and / or the area diagram. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0216] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0217] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0218] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0219] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services. The server may also be a server of a distributed system, or a server combined with a blockchain.

[0220] Artificial intelligence is a discipline that studies how to use computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.). It includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, machine learning / deep learning technology, big data processing technology, knowledge graph technology, and other major directions.

[0221] Cloud computing refers to a technology system that uses network access to elastically scalable shared physical or virtual resource pools. Resources can include servers, instruction systems, networks, software, applications, and storage devices, and can be deployed and managed on demand and in a self-service manner. Cloud computing technology can provide efficient and powerful data processing capabilities for the application of technologies such as artificial intelligence and blockchain, as well as model training.

[0222] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution provided by this disclosure can be achieved, and this document does not limit this.

[0223] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for processing human-computer interaction information, comprising: Get user questions sent by interactive users; Acquire a first identifier of the interactive user; Inputting the user question and the first identifier into a human-computer interaction model for processing, and outputting interaction content; The interactive content is sent to the interactive user.

2. The method according to claim 1, wherein: The obtaining the first identifier of the interactive user includes: The first identification of the interactive user is searched in the pre-stored identifications; different interactive users correspond to different first identifications.

3. The method according to claim 1, wherein: The step of inputting the user question and the first identifier into a human-computer interaction model for processing and outputting interaction content includes: Combining the user question, the first identifier and a preset prompt template to obtain input data; The input data is input into a human-computer interaction model for processing, and the interaction content is output.

4. The method according to any one of claims 1 to 3, wherein: The sending the interactive content to the interactive user includes: Extracting a second identifier from the interactive content; different interactive users correspond to different second identifiers; When the second identifier corresponds to the interactive user, determining feedback content according to the reply content in the interactive content; The feedback content is sent to the interactive user.

5. The method according to claim 4, wherein: The first identifier is a natural language, and the first identifier and the second identifier are determined according to information associated with the interactive user.

6. The method according to claim 4, wherein: The step of determining feedback content according to the reply content in the interactive content includes: determining the reply content in the interactive content as feedback content; or According to the complexity level corresponding to the second identifier, the reply content in the interactive content is processed to obtain feedback content.

7. The method according to claim 1, wherein: The step of inputting the user question and the first identifier into a human-computer interaction model for processing and outputting interaction content includes: When the interactive user is a first user, generating first content according to the user question and a first identifier of the first user; When the interactive user is a second user, second content is generated according to the user question and the first identifier of the second user; wherein the first user belongs to a first group in a target field and the second user belongs to a second group in the target field, the first group and the second group are in opposition to each other, and the first content and the second content are different.

8. A training method for a human-computer interaction model, comprising: Obtaining a training sample of a sample user, wherein the training sample includes a sample question and a sample answer; Obtaining a first identifier of the sample user; The human-computer interaction model is trained according to the first identifier and the training sample.

9. The method according to claim 8, wherein: The training of the human-computer interaction model according to the first identifier and the training sample includes: Acquire at least one second identifier of the sample user; Determine at least one standard output data according to each of the second identifiers and the sample answers; The human-computer interaction model is trained according to the first identifier, the sample question and each of the standard output data.

10. The method according to claim 9, wherein: The step of determining at least one standard output data according to each of the second identifiers and the sample answers comprises: When the number of the second identifier is one, determining the second identifier and the sample answer as standard output data; When there are multiple second identifiers, respectively combine each of the second identifiers with the sample answer to obtain standard output data corresponding to each of the second identifiers; or When there are multiple second identifiers, obtaining, for each of the second identifiers, a complexity level corresponding to the second identifier; Processing the sample answer according to the complexity level corresponding to the second identifier to obtain content of the corresponding complexity level; The second identifier and the corresponding complexity content are determined as standard output data corresponding to the second identifier.

11. The method according to claim 9, wherein: The obtaining of at least one second identifier of the sample user includes: At least one second identifier is determined according to the information associated with the sample user, where the information associated with the sample user includes at least one of the following: the first identifier, the question type of the sample question, and user authority information.

12. A human-computer interaction information processing device, comprising: User question acquisition module, used to obtain user questions sent by users; A first identification acquisition module, used to acquire the first identification of the user; An interaction content generation module, used for inputting the user question and the first identifier into a human-computer interaction model for processing, and outputting interaction content; The interactive content feedback module is used to send the interactive content to the user.

13. The device according to claim 12, wherein: The first identification acquisition module includes: The first identification query unit is used to query the first identification corresponding to the user in the pre-stored identifications; different users correspond to different first identifications.

14. The device according to claim 12, wherein: The interactive content generation module includes: An input data acquisition unit, used for combining the user question, the first identifier and a preset prompt template to obtain input data; The model processing unit is used to input the input data into the human-computer interaction model for processing and output the interaction content.

15. The device according to any one of claims 12 to 14, wherein: The interactive content feedback module shown includes: A second identification acquisition unit, used to extract a second identification from the interactive content; different users correspond to different second identifications; a feedback content determination unit, configured to determine feedback content according to reply content in the interactive content when the second identifier corresponds to the user; A content feedback unit is used to send the feedback content to the user.

16. The device according to claim 15, wherein: The first identifier is a natural language, and the first identifier and the second identifier are determined according to information associated with the user.

17. The device according to claim 15, wherein: The feedback content determination unit includes: an interactive content direct feedback subunit, configured to determine the reply content in the interactive content as feedback content; or The interactive content processing feedback subunit is used to process the reply content in the interactive content according to the complexity corresponding to the second identifier to obtain feedback content.

18. The device according to claim 12, wherein: The interactive content generation module includes: A first content generating unit, configured to generate first content according to the user question and a first identifier of the first user when the interactive user is a first user; A second content generating unit is used to generate second content according to the user question and the first identifier of the second user when the interactive user is a second user; wherein the first user belongs to a first group in a target field and the second user belongs to a second group in the target field, the first group and the second group are in opposition to each other, and the first content and the second content are different.

19. A training device for a human-computer interaction model, comprising: A training sample acquisition module, used to acquire training samples of sample users, wherein the training samples include sample questions and sample answers; A first identification acquisition module, used to acquire the first identification of the sample user; A model training module is used to train the human-computer interaction model according to the first identifier and the training sample.

20. The device according to claim 19, wherein The model training module includes: A second identification obtaining unit, configured to obtain at least one second identification of the sample user; a standard output determination unit, configured to determine at least one standard output data according to each of the second identifiers and the sample answer; A model training unit is used to train the human-computer interaction model according to the first identifier, the sample question and each standard output data.

21. The device according to claim 20, wherein: The standard output determination unit comprises: a single identification determination subunit, configured to determine the second identification and the sample answer as standard output data when the number of the second identification is one; A first multiple identification determination subunit is used for, when there are multiple second identifications, respectively combining each of the second identifications with the sample answer to obtain standard output data corresponding to each of the second identifications; a complexity determination subunit, configured to obtain, for each of the second identifiers, a complexity corresponding to the second identifier when there are multiple second identifiers; A complexity content generation subunit, configured to process the sample answer according to the complexity corresponding to the second identifier to obtain content of the corresponding complexity; The second multi-identification determination subunit is used to determine the second identification and the content of the corresponding complexity as the standard output data corresponding to the second identification.

22. The device according to claim 20, wherein: The second identification obtaining unit includes: The user information association generating subunit is used to determine at least one second identifier based on the sample user associated information, wherein the sample user associated information includes at least one of the following: the first identifier, the question type of the sample question, and user authority information.

23. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the human-computer interaction information processing method described in any one of claims 1-7, or the human-computer interaction model training method described in any one of claims 8-11.

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

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

Images