Data processing method and device in man-machine interaction and electronic equipment

Abstract

The invention discloses a data processing method and device in human-computer interaction, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining human description information, a role knowledge base and model selection parameters configured for a to-be-generated agent; determining a first model from a pre-deployed model set according to the model selection parameters; and adding the human description information and the role knowledge base to a to-be-generated agent, and taking the first model as a model used by the to-be-generated agent to generate the agent. Therefore, according to the data processing method in human-computer interaction provided by the embodiment of the invention, the intelligent agent which meets the personalized requirements of the user and has certain settings can be created, so that the human-computer interaction experience is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, device, electronic device and computer-readable storage medium in human-computer interaction. Background Art

[0002] With the rapid development of artificial intelligence technology, human-computer interaction has been widely used in many fields such as game development and virtual assistants. At present, users can build intelligent agents with basic interaction capabilities and certain situational understanding capabilities. However, the responses of currently created intelligent agents to different users are relatively simple, and it is difficult to show the unique personality of real humans.

[0003] Therefore, how to create intelligent agents that meet more complex and diverse needs has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] This application provides a data processing method, device, electronic device and computer-readable storage medium in human-computer interaction, which can create an intelligent agent with a certain personality that meets the personalized needs of users, thereby improving the human-computer interaction experience. The specific scheme is as follows:

[0005] In a first aspect, an embodiment of the present application provides a data processing method in human-computer interaction, the method comprising:

[0006] Obtaining the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated;

[0007] Determining a first model from a set of pre-deployed models according to the model selection parameters;

[0008] The character description information and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0009] In a second aspect, an embodiment of the present application provides a data processing device in human-computer interaction, the device comprising:

[0010] An acquisition unit, used to acquire the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated;

[0011] a determining unit, configured to determine a first model from a set of pre-deployed models according to the model selection parameters;

[0012] A generating unit is used to add the character description information and the role knowledge base to the intelligent agent to be generated, and use the first model as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0013] In a third aspect, the present application further provides an electronic device, including:

[0014] Processor; and

[0015] The memory is used to store a data processing program. After the electronic device is powered on and the program is run by the processor, the method described in the first aspect is executed.

[0016] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium storing a data processing program, which is executed by a processor to perform the method described in the first aspect.

[0017] Compared with the prior art, this application has the following advantages:

[0018] The data processing method in human-computer interaction provided by the embodiment of the present application includes the following steps: obtaining the human description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated; determining the first model from the pre-deployed model set according to the model selection parameters; adding the human description information and the role knowledge base to the intelligent agent to be generated, and using the first model as the model used by the intelligent agent to be generated to generate the intelligent agent. It can be seen that in the present application, first, the parameters configured by the user for the intelligent agent to be generated are obtained, and the configured parameters include the human description information that meets the personalized needs of the user, the role knowledge base that stores the knowledge that the intelligent agent to be generated needs to master, and the model selection parameters that indicate which model the intelligent agent to be generated uses. In this way, based on the model selection parameters, the first model used by the intelligent agent to be generated can be determined from the pre-deployed model set. By adding the human description information and the role knowledge base to the intelligent agent to be generated, the obtained intelligent agent has a certain human setting and a certain knowledge reserve, and by using the first model as the model used by the intelligent agent to be generated, the obtained intelligent agent can reply to the interactive information input by the user based on the algorithm of the first model. In summary, in the process of human-computer interaction, the generated intelligent agent can be based on the knowledge in the corresponding first person setting and the corresponding role knowledge base, and call the algorithm of the first model to generate personalized replies. Therefore, the data processing method in human-computer interaction provided by the embodiment of the present application can create an intelligent agent with a certain person setting that meets the personalized needs of users, thereby improving the human-computer interaction experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a human-computer interaction system diagram for implementing a data processing method in human-computer interaction provided by an embodiment of the present application;

[0020] Figure 2 is a flow chart of a data processing method in human-computer interaction provided by an embodiment of the present application;

[0021] Figure 3 It is a flowchart of model scheduling in the data processing method in human-computer interaction provided in an embodiment of the present application;

[0022] Figure 4 It is a structural block diagram of an example of a data processing device in human-computer interaction provided in an embodiment of the present application;

[0023] Figure 5 It is a structural block diagram of an example of an electronic device for data processing in human-computer interaction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0025] It should be noted that the terms "first", "second", "third", etc. in the claims, description and drawings of the present application are used to distinguish similar objects and are not used to describe a specific order or sequence. The data used in this way are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including", "having" and their variants are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It should be understood that in the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" is merely a way to describe the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "Including A, B and / or C" means including any one, any two, or any three of A, B, and C.

[0027] It should be understood that in the embodiments of the present application, "B corresponding to A", "B corresponding to A", "A corresponds to B", or "B corresponds to A" means that B is associated with A, and B can be determined according to A. Determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0028] It should be noted that the instructions or operations that appear in the subsequent detailed introduction of the data processing method in the human-computer interaction provided by the embodiment of the present application can be regarded as instructions or operations triggered by the player through fingers or controlling the mouse, keyboard, game controller or stylus or voice. The specific triggering operation can be determined according to the type of electronic device. For example, when the electronic device is a touch screen device such as a mobile phone, a tablet computer, a game console, etc., the player can trigger instructions or operations on the touch screen through any suitable object or accessory such as a finger or stylus, or through voice. When the terminal device is a non-touch screen terminal device such as a desktop computer or a laptop computer, the player can trigger instructions or operations through external devices such as a mouse, keyboard, and game controller.

[0029] Based on background technical reasons, in order to create an intelligent entity that meets the user's personalized needs and has a certain personality, thereby improving the human-computer interaction experience, the first embodiment of the present application provides a data processing method in human-computer interaction. The method is applied to an electronic device, which can be a desktop computer, a laptop computer, a mobile phone, a tablet computer, an electronic watch, etc., or other electronic devices that can perform data processing in human-computer interaction, which is not specifically limited in the embodiments of the present application.

[0030] For example, in combination with the above introduction, Figure 1 A human-computer interaction system 100 for implementing a data processing method in human-computer interaction provided by an embodiment of the present application is shown, and the human-computer interaction system 100 may include a client 101 and a server 102. The client 101 may be connected to the server 102 via a network. The client may run on any device with computing hardware that can support and execute software application tools corresponding to human-computer interaction.

[0031] In the above-mentioned human-computer interaction system 100, the client 101 can be a human-computer interaction application installed on an electronic device through an installation package, or a human-computer interaction application accessed through a browser, a small program, or other client. The client includes modules 201, 202, 203, and 204. The user can register and log in to the human-computer interaction application through module 201, search for "official roles" opened by the development team through module 202, and chat with existing roles through module 203, or search for historical conversation records with roles that have been interacted with, and customize the role through module 204, or modify the configuration information of the customized role, and view the information of the personal center and the current data statistics.

[0032] In the above-mentioned human-computer interaction system 100, the server 102 may include a system background 205 and a model scheduling 206, wherein the system background 205 may include an account management module and a system log module, the account management module is used to manage the accounts created in the human-computer interaction application, and the system log module is used to record various logs generated during the operation of the human-computer interaction application. Multiple models are connected to the model scheduling 206, including model 1, model 2, ... model n, and the model scheduling is used to provide users with matching models to achieve corresponding functions. The multiple models may include but are not limited to a model for realizing a face pinching function, a model for extracting timbre, and a model for realizing at least one of the functions of text-generated text, picture-generated text, text-generated picture, and picture-generated picture.

[0033] Optionally, an apikey (interface key) can be created in advance in the personal center of the human-computer interaction application, so that users who log in to the human-computer interaction application can quickly access the system and quickly call the pre-deployed model to achieve rapid customization of the intelligent body.

[0034] It should be noted that Figure 1 The human-computer interaction system schematic diagram shown is merely an example. The human-computer interaction system 100 described in the embodiment of the present application is intended to more clearly illustrate the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application.

[0035] The technical solution of the present application is described in detail below through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0036] The following, combined Figure 2 and Figure 3 The present invention introduces a data processing method in human-computer interaction provided by an embodiment of the present application.

[0037] like Figure 2 As shown, the data processing method in human-computer interaction provided by the present application includes the following steps S101 to S103.

[0038] Step S101: Obtain the personality description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated.

[0039] In the agent creation interface, users can configure parameters for the agent to be generated. Specifically, the configured parameters include personality description information, role knowledge base and model selection parameters.

[0040] The above-mentioned character description information is used to indicate the character attributes of the intelligent body to be generated. The character description information may include but is not limited to at least one of the following: image resources representing the image, audio resources representing the timbre, personality information, and worldview information.

[0041] The image resource representing the image can be an image uploaded by the user or an image selected by the user from a preset gallery. The image resource can be one of a real person image, an animal image, an animation image, a film and television character image, etc. By using the image resource representing the image, an intelligent agent presented in a specific image can be customized.

[0042] The audio resource representing the timbre may be audio uploaded by the user, or audio selected by the user from a preset audio library. The audio resource may be audio of a real person, audio of an animation character, or audio of a character in a film or TV series.

[0043] The personality information may include at least one of cheerfulness, optimism, honesty, courage, humor, calmness, etc.

[0044] Worldview information refers to the cognitive and understanding framework set for the intelligent agent to be generated in the virtual world of human-computer interaction, including basic assumptions, beliefs, values ​​and behavioral norms of the virtual world. Worldview information is used to define the way the intelligent agent to be generated "views" and "understands" the virtual world in which it lives. Worldview information can generally include but is not limited to at least one of the following: background information, value information, and emotional information. Among them, background information represents the background of the intelligent agent to be generated. For example, the intelligent agent to be generated can be set to live in a science fiction world of the future, or it can also be set to live in ancient times. Value information is used to influence the views and opinions of the intelligent agent to be generated on things. Emotional information is used to give the intelligent agent to be generated the ability to express emotions.

[0045] Exemplarily, the personality description information configured by the user for the intelligent agent to be generated is: name - Xiao Ming, age - 6 years old, gender - male, hobbies - likes singing and dancing, favorite thing - watching cartoons, favorite person - mother.

[0046] The above-mentioned role knowledge base refers to the knowledge that is established and required to be mastered by the intelligent agent to be generated, and is used to store the knowledge mastered by the intelligent agent to be generated, so as to generate response information based on the role knowledge base during the human-computer interaction process.

[0047] The above model selection parameters characterize the model used by the agent to be generated.

[0048] After the user configures the parameters for the intelligent agent to be generated, the character description information, role knowledge base and model selection parameters can be obtained accordingly.

[0049] Step S102: Determine a first model from a set of pre-deployed models according to the model selection parameters.

[0050] After obtaining the above-mentioned character description information, role knowledge base and model selection parameters, a first model can be determined from a pre-deployed model set according to the model selection parameters. The first model is the first model used by the to-be-generated intelligent agent in the human-computer interaction process.

[0051] The above model set may include general models and multiple specialized models. General models are used to solve a wide range of diverse problems and have high generalization capabilities. General models can usually cover knowledge in multiple fields. Specialized models are used to solve a certain type of specific problem or complete a certain task. Usually, specialized models perform better than general models in specific fields.

[0052] Combined with the types of models included in the model collection, when setting the model selection parameters in the creation interface, the user can select a suitable model based on actual needs. For example, if the user wants to create an intelligent agent that performs well in a specific field, he or she can choose a dedicated model that excels in that specific field. For example, if the user wants to create an intelligent agent that is proficient in various fields, he or she can choose a general model.

[0053] For example, in response to the question "What are the reasons for the formation of the Himalayas and their impact on the climate of the surrounding areas", the general model's answer is relatively broad, such as "The Himalayas are located in southern Asia and are one of the longest and highest mountain ranges in the world. It was formed due to the collision of the Indian Plate and the Eurasian Plate. This geological process began about 50 million years ago. The Himalayas have a significant impact on the climate of the surrounding areas, blocking the cold air from the north and forming the South Asian monsoon system"; the dedicated model's answer is more detailed and accurate, such as "The formation of the Himalayas is due to the continuous collision of the Indian Plate and the Eurasian Plate. This process began about 5000 years ago. Ten thousand years ago. As the Indian plate moved northward, it gradually rose to form the Himalayas. This geological process not only created the highest peaks in the world, but also triggered frequent seismic activity. The Himalayas have a profound impact on the climate of the surrounding areas. It blocks the cold air from the north, allowing South Asia to form a unique monsoon climate. In summer, the Himalayas block the moist air from the Indian Ocean, resulting in strong monsoon rainfall in South Asia; in winter, the Himalayas block the cold air from the north, making South Asia relatively warm. In addition, the Himalayas also affect the surrounding ecosystems, forming a diverse biome."

[0054] In step S102, if the model selection parameter is a parameter indicating a general model, the general model may be determined as the first model; if the model selection parameter is a parameter indicating a first dedicated model among multiple dedicated models, the first dedicated model may be determined as the first model.

[0055] Step S103: Add the character description information and the role knowledge base to the intelligent agent to be generated, and use the first model as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0056] After obtaining the above-mentioned character description information, role knowledge base, and determining the first model to be used, an intelligent entity can be successfully created that has a first character that matches the character description information, can master the knowledge in the role knowledge base, and can understand and respond to the interaction information input by the user in human-computer interaction through the algorithm of the first model.

[0057] In an optional implementation, when the character description information includes image resources, step S103 may be implemented by the following steps:

[0058] Performing face pinching processing on the image resource to obtain a target image;

[0059] The target image and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated, so as to obtain an intelligent agent presented with the target image.

[0060] It should be noted that the multiple models included in the model scheduling may include a model for face pinching. "Face pinching" refers to the process of creating or customizing the appearance of a character by adjusting the facial features of the character in digital content creation, especially in electronic games, virtual reality (VR) and 3D modeling software. When the user uploads an image resource representing the image in the creation interface of the intelligent agent, the interface of the model for face pinching in the model scheduling can be called to adjust the image resource representing the image uploaded by the user to obtain the target image.

[0061] It should be noted that by performing face-pinching processing on image resources, a game-level, realistic target image can be obtained.

[0062] Optionally, the target image can be a 2D image or a 3D image. A 2D image refers to a character or object represented on a two-dimensional plane, usually with a flat and concise visual effect; a 3D image refers to a character or object represented in a three-dimensional space, usually with a three-dimensional and realistic visual effect.

[0063] In this embodiment, the target image can be assigned to the intelligent agent to be generated, and a role knowledge base can be added to store the intelligent agent to be generated, and the first model can be used as the model used by the intelligent agent to be generated, thereby creating an intelligent agent that is presented with the target image, masters the knowledge in the role knowledge base, and can understand and respond to the interactive information input by the user in human-computer interaction through the algorithm of the first model.

[0064] The data processing method in human-computer interaction provided by the embodiment of the present application includes the following steps: obtaining the human description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated; determining the first model from the pre-deployed model set according to the model selection parameters; adding the human description information and the role knowledge base to the intelligent agent to be generated, and using the first model as the model used by the intelligent agent to be generated to generate the intelligent agent. It can be seen that in the present application, first, the parameters configured by the user for the intelligent agent to be generated are obtained, and the configured parameters include the human description information that meets the personalized needs of the user, the role knowledge base that stores the knowledge that the intelligent agent to be generated needs to master, and the model selection parameters that indicate which model the intelligent agent to be generated uses. In this way, based on the model selection parameters, the first model used by the intelligent agent to be generated can be determined from the pre-deployed model set. By adding the human description information and the role knowledge base to the intelligent agent to be generated, the obtained intelligent agent has a certain human setting and a certain knowledge reserve, and by using the first model as the model used by the intelligent agent to be generated, the obtained intelligent agent can reply to the interactive information input by the user based on the algorithm of the first model. In summary, in the process of human-computer interaction, the generated intelligent agent can be based on the knowledge in the corresponding first person setting and the corresponding role knowledge base, and call the algorithm of the first model to generate personalized replies. Therefore, the data processing method in human-computer interaction provided by the embodiment of the present application can create an intelligent agent with a certain person setting that meets the personalized needs of users, thereby improving the human-computer interaction experience.

[0065] In an optional implementation manner, the data processing method in human-computer interaction provided in the embodiment of the present application may further include the following steps:

[0066] Obtaining open configuration parameters for the agent;

[0067] The agent is configured to an open state indicated by the open configuration parameter.

[0068] In this implementation, the user can also set the open state of the obtained intelligent agent. Specifically, the open state can be set when configuring parameters in the creation interface of the intelligent agent before the intelligent agent is created, or the open state can be set for the created intelligent agent after the intelligent agent is created. This application does not limit the timing of setting the open state.

[0069] The set open configuration parameter may be a first parameter for indicating that the agent is configured to be visible to all users who log in to the human-computer interaction application, a second parameter for indicating that the agent is configured to be visible to some users who log in to the human-computer interaction application, or a third parameter for indicating that the agent is configured to be visible to only the user who created the agent. The aforementioned some users may be, for example, users in a certain region or users of a certain age group.

[0070] In this way, the agent can be configured to the open state indicated by the open configuration parameter. Specifically, when the open configuration parameter is the first parameter, the agent will be configured to the first state visible to all users who log in to the human-computer interaction application, that is, all users who log in to the human-computer interaction application can interact with the agent; when the open configuration parameter is the second parameter, the agent will be configured to the second state visible to some users who log in to the human-computer interaction application, that is, these some users can interact with the agent; when the open configuration parameter is the third parameter, the agent will be configured to the third state visible only to the user who created the agent, that is, only the user who created the agent can interact with the agent.

[0071] By configuring the open configuration parameters, users can freely and flexibly set the open state of the intelligent agent, which improves the flexibility in human-computer interaction.

[0072] The above is an introduction to the process of creating an intelligent agent in the data processing method in human-computer interaction provided in the embodiment of the present application.

[0073] The following introduces the use process of the intelligent agent in the human-computer interaction process in the data processing method in the human-computer interaction provided in the embodiment of the present application.

[0074] Optionally, the data processing method in human-computer interaction provided in the embodiment of the present application may further include the following steps:

[0075] In response to a user inputting first interaction information on an interaction interface of the agent, determining first reply information according to the role knowledge base and a model knowledge base corresponding to the first model;

[0076] Extracting personality feature data corresponding to the personality description information;

[0077] The character feature data and the first reply information are integrated to obtain second reply information that conforms to the first character.

[0078] After the intelligent agent is created, the user can interact with the intelligent agent. The user is a user among the interactive users corresponding to the intelligent agent. It can be the user who created the intelligent agent or other users who did not create the intelligent agent and logged in to the human-computer interaction application.

[0079] The user can input the first interactive information in the interactive interface of the agent, and the presentation form of the first interactive information can be at least one of text and image, and can specifically be query information, interactive information, etc. For example, the first interactive information is in the form of text such as "Who is your good friend?" or "Please recommend me a good movie," or the first interactive information is in the form of a picture and text + image such as "Please analyze the picture." In this way, a multimodal dialogue of the agent can be realized.

[0080] When receiving the first interaction information input by the user, firstly, the first reply information can be determined according to the knowledge in the role knowledge base added by the agent and the knowledge in the model knowledge base corresponding to the first model itself used by the agent.

[0081] It should be noted that in the embodiment of the present application, the first reply information can be determined based on the historical conversation messages between the user and the agent as a reference. By referring to the historical conversation messages, the agent can better understand the context of the current conversation, thereby generating a more coherent and natural response. Moreover, based on the historical conversation messages, the agent can quickly acquire the user's preferences and habits and provide more personalized services.

[0082] Afterwards, the personality feature data corresponding to the personality description information added by the agent can be extracted. The personality feature data can be a vector that represents the personality. Specifically, when the personality description information includes an audio resource that represents the timbre, the timbre feature data can be extracted from the audio resource through a speech model in a pre-deployed model set. For example, if the audio resource is an audio clip of a little boy, the timbre feature data of the little boy can be extracted through a speech model; when the personality description information includes character information, the character feature data can be extracted from the character information; when the personality description information includes worldview information, the worldview feature data can be extracted from the worldview information. Afterwards, the timbre feature data, the character feature data, and the worldview feature data can be superimposed to obtain the personality feature data.

[0083] Afterwards, based on the first reply information, the character feature data can be integrated to obtain the second reply information that matches the first character. In other words, the second reply information is the information after the character feature data is added to the first reply information.

[0084] For example, the first interactive information is "What is photosynthesis?", and the first reply information can be "Photosynthesis is the process by which plants, algae, and some bacteria use light energy to convert carbon dioxide and water into organic matter and oxygen. This process mainly occurs in the chloroplasts of plants, where light energy drives chemical reactions to produce energy storage and oxygen release. Photosynthesis is an important part of the carbon cycle in ecosystems and is essential for maintaining oxygen levels in the Earth's atmosphere." If the character of the intelligent agent is a curious little boy, then the second reply information obtained by integrating the character feature data and the first reply information can be "Wow, photosynthesis is like the super magic of plants! Did you know that plants are actually little scientists who experiment with sunlight, water, and carbon dioxide in the air, and put these things in a small factory in the leaves, a place called chloroplasts, and then use the sun's energy to turn them into sugar and oxygen. Sugar is the food of the plant, and oxygen is released for us to breathe. Don't you think this is cool?"

[0085] It should be noted that, without setting the character description information, the generated agent's responses are usually more general, without obvious personality traits or specific worldviews, and are mainly used to provide information and functional help without involving too much emotion or personal color. When the character description information is set, the generated agent's responses will reflect specific character traits, making them more personalized.

[0086] In an optional implementation, the role knowledge base corresponding to the agent can be stored in the form of a vector database. Therefore, the data processing method in human-computer interaction provided in the embodiment of the present application can also include the following steps:

[0087] Each piece of knowledge in the role knowledge base is vector-embedded to obtain a vector database corresponding to the role knowledge base.

[0088] In this implementation, a vector database can be used to store and retrieve knowledge in the role knowledge base. A vector database is a database specifically used to store, search, and manage vectors. The vector database can process unstructured data such as text, voice, image, and video, and convert these data into vector form through vectorization technology for efficient management and query.

[0089] Specifically, the knowledge in the role knowledge base corresponding to the intelligent agent can be text fragments, question-answer pairs, and keywords. When the role knowledge base corresponding to the intelligent agent is obtained, the knowledge contained therein can be vector embedded, thereby converting it into vector form and storing it in the vector database.

[0090] This setting method, by storing various knowledge in the character knowledge base in a vector database, facilitates the management and query of information and can improve the response efficiency of the intelligent agent in the process of human-computer interaction.

[0091] Optionally, when each piece of knowledge in the role knowledge base is stored in the form of a vector, the above step of "determining the first reply information according to the role knowledge base and the model knowledge base corresponding to the first model" can be implemented by the following steps:

[0092] Converting the first interaction information into corresponding first vector data;

[0093] Determining, from the vector database, a preset number of vector entries having a correlation with the first vector data greater than a preset threshold;

[0094] generating initial reply information according to the vector entry and a preset prompt template;

[0095] The initial reply information is completed through the model knowledge base to obtain the first reply information.

[0096] In an embodiment of the present application, in response to a user inputting first interaction information in the interactive interface of the intelligent agent, the first interaction information can be converted into corresponding first vector data, and the search characteristics of the vector database can be used to find a preset number of vector entries with a high degree of relevance, that is, a preset number of vector entries with a relevance to the first vector data greater than a preset threshold. Afterwards, the preset number of vector entries can be integrated into a preset prompt template, and the preset prompt template is used to guide the intelligent agent to generate customized reply information, thereby generating initial reply information. For example, the top 10 most relevant vector entries can be integrated into the preset prompt template to obtain the corresponding initial reply information. Afterwards, the model interface corresponding to the first model can be called, and the initial reply information can be completed through the model knowledge base corresponding to the first model, thereby obtaining the first reply information.

[0097] It should be noted that since the role knowledge base is the knowledge reserve information set by the user for the agent, the initial reply information generated based on the knowledge in the role knowledge base meets the customized requirements of the user who created the agent. The initial reply information is further supplemented by the model knowledge base to obtain a more humanized and more complete first reply information.

[0098] Exemplarily, when a user asks the agent: "In which district is Beijing's Fragrant Hills Park located?", the agent will convert the question into vector form, query the vector database corresponding to the role knowledge base, find the knowledge items related to "Beijing Fragrant Hills Park", and fill them in based on the preset prompt template. The initial response information obtained may be "Fragrant Hills Park is located in Haidian District, Beijing". Afterwards, the initial response information is completed by the first model used by the agent, and the first response information obtained may be "Fragrant Hills Park is located in Haidian District, Beijing. In autumn, the mountains are covered with red leaves, which is very beautiful."

[0099] In an optional implementation, the above step of "determining the first reply information according to the role knowledge base and the model knowledge base corresponding to the first model" can be implemented by the following steps:

[0100] Acquire model constraint parameters configured for the first model; the model constraint parameters include at least one of the following: a response diversity parameter, a response divergence parameter, and a memory round parameter;

[0101] Determining a response method of the agent according to the model constraint parameters;

[0102] First reply information is determined according to the role knowledge base, the model knowledge base corresponding to the first model, and the reply method.

[0103] In an embodiment of the present application, the parameter configuration performed by the user in the creation interface of the agent may also include the configuration of model constraint parameters, which are used to constrain the agent when replying to the user's interactive information. Specifically, after the user has configured the model selection parameters and selected the first model, the model constraint parameters may be configured for the first model. The model constraint parameters may include, but are not limited to, a reply diversity parameter, a reply divergence parameter, a memory round parameter, and a maximum number of reply characters.

[0104] Among them, the response diversity parameter represents that the agent responds in a more diverse way when generating response information. This response means that when answering questions, it can provide information from different dimensions, use different methods, or cite different materials, so that the responses generated by the agent can include different viewpoints and opinions, and be more comprehensive and objective. For example, when discussing a social phenomenon, it can be analyzed from multiple perspectives such as economy, culture, and history.

[0105] The response divergence parameter characterizes the agent's response in a more divergent way when generating response information. This response method refers to a way of thinking that starts from a problem and explores in multiple directions. It enables the agent to try to think about the problem from different angles and put forward novel ideas or solutions.

[0106] The memory turn parameter represents the number of historical conversation messages (such as the most recent 20 messages, or the messages in the last week) that the agent refers to when generating a reply message.

[0107] It can be seen that based on the configured model constraint parameters, the response method of the agent can be determined, so that the first response information matching the response method can be determined according to the role knowledge base and the model knowledge base corresponding to the first model and the response method. For example, if the model constraint parameter is a response diversity parameter, the first response information including multiple different dimensions can be determined; if the model constraint parameter is a response divergence parameter, the first response information including multiple different angles can be determined; if the model constraint parameter is a memory round parameter, the first response information can be determined based on the number of historical conversation messages indicated by the memory round parameter.

[0108] This setting method, by configuring the model constraint parameters of the first model, can constrain the response method of the intelligent agent, thereby generating matching response information.

[0109] As can be seen from the above introduction, the model set includes a general model and multiple special models. In an optional implementation, when receiving the first interaction information input by the user in the interaction interface of the agent, it can be determined whether to perform model scheduling based on whether the agent is configured to use a special model when it is created.

[0110] like Figure 3 , which is a flowchart of model scheduling in a data processing method in human-computer interaction provided in an embodiment of the present application, including steps S201 to S206.

[0111] First, if Figure 3 As shown in step S201 and step S202 in the embodiment, when receiving the first interaction information input by the user in the interaction interface of the agent, it can be determined whether the first model used by the agent is a dedicated model. If the first model is a general model, the execution Figure 3 Steps S203 to S205 in the above example are executed when the first model is a dedicated model. Figure 3 Step S206 in .

[0112] That is, the above step of "in response to the user inputting the first interaction information in the interactive interface of the agent, determining the first reply information according to the role knowledge base and the model knowledge base corresponding to the first model" specifically includes:

[0113] Step S206: If the first model is the dedicated model, in response to the user inputting first interaction information in the interactive interface of the agent, first reply information is determined according to the role knowledge base and the model knowledge base corresponding to the first model.

[0114] In this embodiment, when a user creates an agent and selects a dedicated model as the model used by the agent in the human-computer interaction process, the algorithm of the dedicated model will always be used to generate reply information during the human-computer interaction of the agent.

[0115] When the first model is a general model, the data processing method in human-computer interaction provided in the embodiment of the present application may also include: Figure 3 The following steps S203 to S205 are as follows:

[0116] Step S203: if the first model is the general model, in response to the user inputting the first interaction information in the interaction interface of the agent, determining a second model from each of the dedicated models;

[0117] Step S204: switching the first model used by the agent to the second model;

[0118] Step S205: Determine third reply information according to the role knowledge base and the model knowledge base corresponding to the second model.

[0119] In this embodiment, when a user creates an agent and selects a general model as the model used by the agent in the human-computer interaction process, each time the agent receives first interaction information input by the user, it will determine a second model from multiple dedicated models in a pre-deployed model set to achieve real-time scheduling of the model, and switch the first model used by the agent to the second model. In this way, based on the first interaction information input by the user, combined with the vector database corresponding to the role knowledge base added by the agent and the model knowledge base corresponding to the second model, the third reply information can be determined.

[0120] Similar to the above-mentioned fusion of personality characteristic data and the first reply information to obtain the second reply information, after obtaining the third reply information here, the third reply information and the personality characteristic data can also be fused to obtain the fourth reply information. The fourth reply information is the reply of the intelligent agent using the second model to the user's first interaction information.

[0121] Optionally, the above step of "determining the second model from each of the dedicated models" may include: obtaining the information type of the first interactive information, and determining the second model matching the information type from each of the dedicated models. The information type may include, but is not limited to, one of geography, history, mathematics, physics, chemistry, medicine, computer science, music, and sports.

[0122] For example, when the user selects a general model to create an agent, if the user asks the agent a question 1 about music, then model A that is good at music-related knowledge can be determined from the dedicated models, and question 1 can be answered through model A; if the user asks the agent a question 2 about geography, then model B that is good at geography-related knowledge can be determined from the dedicated models, and question 2 can be answered through model B, and so on.

[0123] In this setting, when the user selects a general model to generate an agent, during the human-computer interaction of the agent, a matching second model is selected from the dedicated model in real time and returned to the agent, so as to give a response in combination with the vector database corresponding to the persona and role knowledge base and the model knowledge base corresponding to the second model. In this way, through the scheduling of the model during the use of the agent, the agent can always provide users with high-quality and professional answers.

[0124] Optionally, the above step of “determining the second model from each of the dedicated models” may include the following steps:

[0125] Acquire model evaluation data for each of the dedicated models, wherein the model evaluation data represents the performance of each of the dedicated models in a target dimension, wherein the target dimension includes at least one of user portrait information, interaction information, and persona information;

[0126] Based on the first dimension data corresponding to the target dimension and the model evaluation data, a matching second model is determined from the dedicated model, wherein the first dimension data includes at least one of the user portrait information of the user, the first interaction information and the personality description information.

[0127] In an embodiment of the present application, each dedicated model in the model set can be evaluated in advance from a target dimension, and the target dimension can include at least one of a user portrait dimension, an interactive information type dimension, and a personality dimension, so as to obtain model evaluation data that characterizes the performance of each dedicated model in the target dimension.

[0128] It should be noted that the user portrait dimension may include attributes such as gender, age, personality, hobbies, etc. The interactive information type dimension may include attributes such as geography, history, mathematics, physics, chemistry, medicine, computer science, music, sports, etc. The persona dimension may include attributes such as martial arts characters, doctors, teachers, and students.

[0129] By evaluating each special model through the user portrait dimension, we can get the user portrait type that each special model is suitable for, for example, model A is suitable for women aged 20 to 30, and model B is suitable for men aged 30 to 40; by evaluating each special model through the interactive information type dimension, we can get the field that each special model is good at, for example, model A is good at music knowledge, and model B is good at geography knowledge; by evaluating each special model through the personality dimension, we can get the personality that each special model is adapted to, for example, model A performs well in martial arts roles, model B performs well in teacher roles, etc.

[0130] In this way, in response to the user inputting the first interaction information in the interactive interface of the intelligent agent, the first dimension data corresponding to the above-mentioned target dimension can be obtained. For example, when the target dimension includes the user portrait dimension, the interaction information type dimension and the personality dimension, the first dimension data may include the user portrait information of the user, the target session information and the personality description information corresponding to the intelligent agent. It should be noted that in the embodiment of the present application, the target dimension is introduced as an example including the user portrait dimension, the interaction information type dimension and the personality dimension, and the present application does not make specific restrictions on this.

[0131] In this way, based on the first dimension data and the model evaluation data, a matching second model can be determined from the dedicated model, so that more accurate, professional and high-quality answers can be provided through the second model.

[0132] Optionally, the data processing method in human-computer interaction provided in the embodiment of the present application may further include the following steps:

[0133] Obtaining comment data of users on each actual reply message of the agent;

[0134] According to the comment data and each of the actual reply information, adjusting the model parameters of the first model to obtain a trained first model;

[0135] The first model used by the agent is replaced with the trained first model.

[0136] In the embodiment of the present application, the user can comment on the actual reply information generated by the agent during the human-computer interaction process, such as likes, dislikes, etc. The system can collect the user's comment data on each actual reply information of the agent, and the comment data corresponding to an actual reply information represents the quality of the actual reply information.

[0137] The types of comment data include positive type and negative type. It can be understood that positive type of comment data indicates that the quality of the corresponding actual reply information is good, while negative type of comment data indicates that the quality of the corresponding actual reply information is poor.

[0138] After obtaining the user's comment data on each actual reply message of the intelligent agent, the comment data and the corresponding actual reply messages can be used as sample data to reversely train the first model. Thereby, when the human-computer interaction application is already online, the first model can be trained through the feedback of online users, and the model parameters of the first model can be adjusted to obtain the trained first model. In this way, the first model in the pre-deployed model set can be updated to a new version, thereby replacing the first model used by the intelligent agent with the trained first model.

[0139] Optionally, for an intelligent agent, corresponding comment data may be collected according to a preset period (such as 2 days, a week, etc.), so as to train the first model used by the intelligent agent based on the comment data.

[0140] With this setting, after the human-computer interaction application is launched, by collecting users' comment data on the actual reply information of the intelligent agent, the first model used by the intelligent agent is reversely trained, thereby retraining the model deployed in the human-computer interaction application, improving the robustness and prediction accuracy of the model, and making the model perform more stably and accurately on new data that has not been seen.

[0141] In an optional specific implementation, the above step of "adjusting the model parameters of the first model according to the comment data and each of the actual reply information to obtain the trained first model" can be implemented by the following steps:

[0142] For each of the actual reply information, the actual reply information, the interaction information input by the user corresponding to the actual reply information, and the comment data corresponding to the actual reply information are used as a training sample pair to obtain a training sample set;

[0143] Inputting the training sample set into the first model to obtain predicted reply information for each of the interaction information input by the user corresponding to each of the actual reply information;

[0144] Adjusting the model parameters of the first model so that the difference between the predicted reply information and the actual reply information corresponding to the positive sample pair is less than or equal to a first threshold, and the difference between the predicted reply information and the actual reply information corresponding to the negative sample pair is greater than a second threshold, to obtain a trained first model;

[0145] The positive sample pairs are training sample pairs corresponding to the positive type of the comment data in the training sample set, and the negative sample pairs are training sample pairs corresponding to the negative type of the comment data in the training sample set.

[0146] First, a training sample set is constructed. For each actual reply information commented by a user, the actual reply information, the interaction information input by the user corresponding to the actual reply information, and the comment data of the user on the actual reply information are taken as a training sample pair, thereby obtaining multiple training sample pairs to form a training sample set.

[0147] Afterwards, the training sample set is input into the first model, so that the first model generates predicted reply information for each interaction information input by the user corresponding to each actual reply information;

[0148] Afterwards, based on the type (positive or negative) corresponding to each comment data in the training sample set, the model parameters of the first model are adjusted so that the difference between the predicted reply information corresponding to the positive sample pair and the actual reply information is less than or equal to the first threshold, and the difference between the predicted reply information corresponding to the negative sample pair and the actual reply information is greater than the second threshold.

[0149] That is to say, the first model is trained with the training objectives of minimizing the difference between the predicted response information and the actual response information corresponding to the positive sample pairs and maximizing the difference between the predicted response information and the actual response information corresponding to the negative sample pairs, thereby obtaining the trained first model.

[0150] The above is a specific implementation of a model training example. In the specific implementation, the model training can also be performed in other feasible ways, and the embodiments of the present application are not limited thereto.

[0151] Corresponding to the data processing method in human-computer interaction provided in the first embodiment of the present application, the second embodiment of the present application further provides a data processing device in human-computer interaction, such as Figure 4 As shown, the data processing device 400 in the human-computer interaction includes:

[0152] An acquisition unit 401 is used to acquire the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated;

[0153] A determination unit 402 is used to determine a first model from a set of pre-deployed models according to the model selection parameters;

[0154] The generating unit 403 is used to add the character description information and the role knowledge base to the intelligent agent to be generated, and use the first model as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0155] Optionally, the data processing device in the human-computer interaction further includes a reply unit, and the reply unit is used to:

[0156] In response to a user inputting first interaction information on an interaction interface of the agent, determining first reply information according to the role knowledge base and a model knowledge base corresponding to the first model;

[0157] Extracting personality feature data corresponding to the personality description information;

[0158] The character feature data and the first reply information are integrated to obtain second reply information that conforms to the first character.

[0159] Optionally, the reply unit is specifically used for:

[0160] Acquire model constraint parameters configured for the first model; the model constraint parameters include at least one of the following: a response diversity parameter, a response divergence parameter, and a memory round parameter;

[0161] Determining a response method of the agent according to the model constraint parameters;

[0162] First reply information is determined according to the role knowledge base, the model knowledge base corresponding to the first model, and the reply method.

[0163] Optionally, the model set includes a general model and multiple special models; and the reply unit is specifically used to:

[0164] If the first model is the dedicated model, in response to the user inputting first interaction information in the interactive interface of the agent, first reply information is determined according to the role knowledge base and the model knowledge base corresponding to the first model.

[0165] Optionally, the reply unit is further used for:

[0166] If the first model is the general model, in response to the user inputting the first interaction information in the interaction interface of the agent, determining a second model from each of the dedicated models;

[0167] Switching the first model used by the agent to the second model;

[0168] The third reply information is determined according to the role knowledge base and the model knowledge base corresponding to the second model.

[0169] Optionally, the reply unit is further specifically configured to:

[0170] Acquire model evaluation data for each of the dedicated models, wherein the model evaluation data characterizes the performance of each of the dedicated models in a target dimension, wherein the target dimension includes at least one of a user portrait dimension, an interactive information type dimension, and a person setting dimension;

[0171] Based on the first dimension data corresponding to the target dimension and the model evaluation data, a matching second model is determined from the dedicated model, the first dimension data including user portrait information of the user, target session information, and at least one of the personality description information, and the target session information including the first interaction information and at least part of the historical session information before it.

[0172] Optionally, the generating unit 503 is further configured to:

[0173] Each piece of knowledge in the role knowledge base is vector-embedded to obtain a vector database corresponding to the role knowledge base.

[0174] Optionally, the reply unit is further specifically configured to:

[0175] Converting the first interaction information into corresponding first vector data;

[0176] Determining, from the vector database, a preset number of vector entries having a correlation with the first vector data greater than a preset threshold;

[0177] generating initial reply information according to the vector entry and a preset prompt template;

[0178] The initial reply information is completed through the model knowledge base to obtain the first reply information.

[0179] Optionally, the data processing device in the human-computer interaction further includes a training unit, and the training unit is used to:

[0180] Obtaining comment data of users on each actual reply message of the agent;

[0181] According to the comment data and each of the actual reply information, adjusting the model parameters of the first model to obtain a trained first model;

[0182] The first model used by the agent is replaced with the trained first model.

[0183] Optionally, the training unit is specifically used for:

[0184] For each of the actual reply information, the actual reply information, the interaction information input by the user corresponding to the actual reply information, and the comment data corresponding to the actual reply information are used as a training sample pair to obtain a training sample set;

[0185] Inputting the training sample set into the first model to obtain predicted reply information for each of the interaction information input by the user corresponding to each of the actual reply information;

[0186] Adjusting the model parameters of the first model so that the difference between the predicted reply information and the actual reply information corresponding to the positive sample pair is less than or equal to a first threshold, and the difference between the predicted reply information and the actual reply information corresponding to the negative sample pair is greater than a second threshold, to obtain a trained first model;

[0187] The positive sample pairs are training sample pairs corresponding to the positive type of the comment data in the training sample set, and the negative sample pairs are training sample pairs corresponding to the negative type of the comment data in the training sample set.

[0188] Optionally, the character description information acquired by the acquisition unit 401 includes at least one of the following: image resources representing image, audio resources representing timbre, personality information, and worldview information.

[0189] Optionally, when the character description information includes the image resource, the generating unit is specifically configured to:

[0190] Performing face pinching processing on the image resource to obtain a target image;

[0191] The target image and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated, so as to obtain an intelligent agent presented with the target image.

[0192] Optionally, the acquiring unit 401 is further configured to:

[0193] Obtaining open configuration parameters for the agent;

[0194] The agent is configured to an open state indicated by the open configuration parameter.

[0195] Corresponding to the data processing method in human-computer interaction provided in the first embodiment of the present application, the third embodiment of the present application also provides an electronic device for data processing in human-computer interaction.

[0196] like Figure 5 , which is a structural block diagram of an example of an electronic device for data processing in human-computer interaction provided in an embodiment of the present application.

[0197] In this embodiment, an optional hardware structure of the electronic device 500 can be as follows: Figure 5 As shown, it includes: at least one processor 501, at least one memory 502 and at least one communication bus 505; the memory 502 contains a program 503 and data 504.

[0198] The bus 505 may be a communication device for transmitting data between components inside the electronic device 500, such as an internal bus (e.g., a CPU-memory bus, where the processor is a central processing unit, CPU for short), an external bus (e.g., a universal serial bus port, a peripheral component interconnect express port), etc.

[0199] In addition, the electronic device also includes: at least one network interface 506, at least one peripheral interface 507. The network interface 506 provides wired or wireless communication related to an external network 508 (for example, the Internet, an intranet, a local area network, a mobile communication network, etc.); in some embodiments, the network interface 506 may include any number of network interface controllers (NICs), radio frequency (RF) modules, repeaters, transceivers, modems, routers, gateways, any combination of wired network adapters, wireless network adapters, Bluetooth adapters, infrared adapters, near field communication (NFC) adapters, cellular network chips, etc.

[0200] The peripheral interface 507 is used to connect to the peripheral device, and the peripheral device can be the peripheral device 1 ( Figure 5 509), peripheral 2 ( Figure 5 510) and peripheral 3 ( Figure 5 511 in the figure). Peripherals are peripheral devices, which may include but are not limited to cursor control devices (such as mice, touch pads or touch screens), keyboards, displays (such as cathode ray tube displays, liquid crystal displays), displays or light emitting diode displays, video input devices (such as cameras or communication interfaces coupled to video files), etc.

[0201] The processor 501 may be a CPU, or an application specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0202] The memory 502 may include a high-speed RAM (full name: Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0203] The processor 501 calls the program and data stored in the memory 502 to perform the following steps:

[0204] Obtaining the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated;

[0205] Determining a first model from a set of pre-deployed models according to the model selection parameters;

[0206] The character description information and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0207] Corresponding to the data processing method in human-computer interaction provided in the first embodiment of the present application, the fourth embodiment of the present application provides a computer-readable storage medium storing a program of the data processing method in human-computer interaction, the program is executed by a processor to perform the following steps:

[0208] Obtaining the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated;

[0209] Determining a first model from a set of pre-deployed models according to the model selection parameters;

[0210] The character description information and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated to generate the intelligent agent.

[0211] It should be noted that for the detailed description of the devices, electronic devices and computer-readable storage media provided in the second, third and fourth embodiments of the present application, reference can be made to the relevant description of the first embodiment of the present application, and no further details will be given here.

[0212] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

[0213] In a typical configuration, a node device in a blockchain includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0214] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0215] 1. Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), random access memory (RAM) of other attributes, read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage media or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer-readable media does not include non-transitory media such as modulated data signals and carrier waves.

[0216] 2. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0217] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A data processing method in human-computer interaction, characterized in that: The method comprises: Obtaining the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated; Determining a first model from a set of pre-deployed models according to the model selection parameters; The character description information and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated to generate the intelligent agent.

2. The method according to claim 1, characterized in that: The method further comprises: In response to a user inputting first interaction information on an interaction interface of the agent, determining first reply information according to the role knowledge base and a model knowledge base corresponding to the first model; Extracting personality feature data corresponding to the personality description information; The character feature data and the first reply information are integrated to obtain second reply information that conforms to the first character.

3. The method according to claim 2, characterized in that The determining of first reply information according to the role knowledge base and the model knowledge base corresponding to the first model includes: Acquire model constraint parameters configured for the first model; the model constraint parameters include at least one of the following: a response diversity parameter, a response divergence parameter, and a memory round parameter; Determining a response method of the agent according to the model constraint parameters; First reply information is determined according to the role knowledge base, the model knowledge base corresponding to the first model, and the reply method.

4. The method according to claim 2, characterized in that: The model set includes a general model and a plurality of special models; in response to the user inputting first interaction information on the interactive interface of the agent, determining first reply information according to the role knowledge base and the model knowledge base corresponding to the first model, including: If the first model is the dedicated model, in response to the user inputting first interaction information in the interactive interface of the agent, first reply information is determined according to the role knowledge base and the model knowledge base corresponding to the first model.

5. The method according to claim 4, characterized in that The method further comprises: If the first model is the general model, in response to the user inputting the first interaction information in the interaction interface of the agent, determining a second model from each of the dedicated models; Switching the first model used by the agent to the second model; The third reply information is determined according to the role knowledge base and the model knowledge base corresponding to the second model.

6. The method according to claim 5, characterized in that The step of determining the second model from each of the dedicated models comprises: Acquire model evaluation data for each of the dedicated models, wherein the model evaluation data characterizes the performance of each of the dedicated models in a target dimension, wherein the target dimension includes at least one of a user portrait dimension, an interactive information type dimension, and a person setting dimension; Based on the first dimension data corresponding to the target dimension and the model evaluation data, a matching second model is determined from the dedicated model, the first dimension data including user portrait information of the user, target session information, and at least one of the personality description information, and the target session information including the first interaction information and at least part of the historical session information before it.

7. The method according to claim 2, characterized in that The method further comprises: Each piece of knowledge in the role knowledge base is vector-embedded to obtain a vector database corresponding to the role knowledge base.

8. The method according to claim 7, characterized in that The determining of first reply information according to the role knowledge base and the model knowledge base corresponding to the first model includes: Converting the first interaction information into corresponding first vector data; Determining, from the vector database, a preset number of vector entries having a correlation with the first vector data greater than a preset threshold; generating initial reply information according to the vector entry and a preset prompt template; The initial reply information is completed through the model knowledge base to obtain the first reply information.

9. The method according to claim 1, characterized in that: The method further comprises: Obtaining comment data of users on each actual reply message of the agent; According to the comment data and each of the actual reply information, adjusting the model parameters of the first model to obtain a trained first model; The first model used by the agent is replaced with the trained first model.

10. The method according to claim 9, characterized in that The step of adjusting the model parameters of the first model according to the comment data and each of the actual reply information to obtain the trained first model includes: For each of the actual reply information, the actual reply information, the interaction information input by the user corresponding to the actual reply information, and the comment data corresponding to the actual reply information are used as a training sample pair to obtain a training sample set; Inputting the training sample set into the first model to obtain predicted reply information for each of the interaction information input by the user corresponding to each of the actual reply information; Adjusting the model parameters of the first model so that the difference between the predicted reply information and the actual reply information corresponding to the positive sample pair is less than or equal to a first threshold, and the difference between the predicted reply information and the actual reply information corresponding to the negative sample pair is greater than a second threshold, to obtain a trained first model; The positive sample pairs are training sample pairs corresponding to the positive type of the comment data in the training sample set, and the negative sample pairs are training sample pairs corresponding to the negative type of the comment data in the training sample set.

11. The method according to claim 1, characterized in that: The character description information includes at least one of the following: image resources representing the image, audio resources representing the timbre, personality information, and worldview information.

12. The method according to claim 11, characterized in that In a case where the character description information includes the image resource, adding the character description information and the role knowledge base to the intelligent agent to be generated, and using the first model as a model used by the intelligent agent to be generated, to generate the intelligent agent, comprises: Performing face pinching processing on the image resource to obtain a target image; The target image and the role knowledge base are added to the intelligent agent to be generated, and the first model is used as the model used by the intelligent agent to be generated, so as to obtain an intelligent agent presented with the target image.

13. A data processing device in human-computer interaction, characterized in that: The device comprises: An acquisition unit, used to acquire the character description information, role knowledge base and model selection parameters configured for the intelligent agent to be generated; a determining unit, configured to determine a first model from a set of pre-deployed models according to the model selection parameters; A generating unit is used to add the character description information and the role knowledge base to the intelligent agent to be generated, and use the first model as the model used by the intelligent agent to be generated to generate the intelligent agent.

14. An electronic device, characterized in that: include: processor; as well as The memory is used to store a data processing program. After the electronic device is powered on and the program is run by the processor, the method according to any one of claims 1 to 12 is executed.

15. A computer-readable storage medium, characterized in that: A data processing program is stored, and the program is run by a processor to execute the method according to any one of claims 1 to 12.

Images