Information processing method and computing device

Abstract

The embodiment of the invention provides an information processing method and computing equipment. The method comprises the steps of obtaining user input content; determining a target knowledge base which can be accessed by the user in combination with the user feature information and the knowledge base feature information; searching target knowledge content matched with the input content from the target knowledge base; and generating feedback information by using a large language model according to the target knowledge content and the input content. According to the technical scheme provided by the embodiment of the invention, the accuracy of the large language model is improved, and the data security is ensured.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular, to an information processing method and a computing device. Background Art

[0002] With the development of technologies such as deep learning, big data, and cloud computing, various large-scale pre-trained large language models have made great progress, and question-and-answer systems based on large language models have been widely used in fields such as customer service, healthcare, and education and training.

[0003] When training large language models, large-scale data is usually used for pre-training, and various common instructions are used for adjustment to make them easy to communicate with humans. Therefore, they can handle various types of tasks and have obvious advantages in tasks in general fields that require rich knowledge. However, for tasks in some specific professional fields, due to the lack of vertical and professional domain knowledge, general large language models still have certain limitations. Summary of the Invention

[0004] The embodiments of the present application provide an information processing method and a computing device to solve the problem of low accuracy of large language models and ensure data security.

[0005] In a first aspect, an information processing method is provided in the embodiments of the present application, including:

[0006] Obtain user input content;

[0007] Combine user feature information and knowledge base feature information to determine the target knowledge base that the user can access;

[0008] Search for target knowledge content matching the input content from the target knowledge base;

[0009] Generate feedback information using a large language model based on the target knowledge content and the input content.

[0010] In a second aspect, a computing device is provided in the embodiments of the present application, including a processing component and a storage component;

[0011] The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the information processing method as described in the first aspect above.

[0012] In a third aspect, a computer-readable storage medium is provided in the embodiments of the present application, storing a computer program, and when the computer program is executed by a computer, it implements the information processing method as described in the first aspect above.

[0013] In the embodiments of the present application, for the user input content, first, in combination with the user feature information and the knowledge base feature information, the target knowledge base that the user can access is determined; the target knowledge content matching the input content is searched from the target knowledge base; and according to the target knowledge content and the input content, a large language model is used to generate feedback information. Due to the characteristics of the knowledge base such as extensiveness, professionalism, and timeliness, the technical solution of the embodiments of the present application enhances the user input content, can add more prior knowledge to the user input content, makes the obtained feedback information more professional and accurate, thereby improving the adaptability and accuracy of the large language model, and only determines the target knowledge content in the target knowledge base that the user can access, improving data security.

[0014] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0016] Figure 1 The flowchart of an embodiment of an information processing method provided by the present application is shown;

[0017] Figure 2 The flowchart of another embodiment of an information processing method provided by the present application is shown;

[0018] Figure 3 The schematic diagram of the scenario interaction in an actual application of the technical solution of the embodiments of the present application is shown;

[0019] Figure 4 The schematic structural diagram of an embodiment of an information processing device provided by the present application is shown;

[0020] Figure 5 The schematic structural diagram of an embodiment of a computing device provided by the present application is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application.

[0022] In some of the processes described in the specification, claims, and above-mentioned drawings of this application, multiple operations appear in a specific order. However, it should be clearly understood that these operations can be performed not in the order in which they appear herein or in parallel. The operation numbers such as 101, 102, etc. are only used to distinguish different operations, and the numbers themselves do not represent any execution order. Additionally, these processes can include more or fewer operations, and these operations can be performed sequentially or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., do not represent a sequence, and do not limit that "first" and "second" are of different types.

[0023] The technical solution of the embodiment of this application can be applied to the application scenarios of large language models.

[0024] A large language model (LLM for short) refers to a natural language processing model with an extremely large number of parameters and computing power, which can be used for various natural language processing tasks such as text generation, machine translation, question answering systems, etc., and is an AI (Artificial Intelligence) model.

[0025] Combined with the description of the background technology, it can be seen that large language models perform well in general domain tasks, but there are still some deficiencies in terms of credibility, accuracy, and professionalism in some specific professional fields.

[0026] The inventors found in the process of implementing this application that considering that large language models cannot handle natural language tasks in professional fields well, specific professional field knowledge can be integrated to fine-tune large language models to increase their adaptability to that professional field. However, due to the need to collect high-quality annotated specific corpus data in the professional field, there are difficulties and high costs in data collection, as well as high professionalism requirements for the annotated data. Therefore, the fine-tuning training is difficult and costly.

[0027] In order to improve the adaptability and accuracy of large language models and enhance their application, the inventors proposed the technical solution of the embodiments of this application through a series of studies. In the embodiments of this application, it is not necessary to fine-tune the large language model using domain-specific corpora. Instead, the knowledge base is used to determine the target knowledge content that matches the user input content, and the target knowledge content is used as the context information of the user input content for assistance. Since the knowledge base has characteristics such as extensiveness, professionalism, and timeliness, the technical solution of the embodiments of this application enhances the user input content, adds more prior knowledge to the user input content, makes the obtained feedback information more professional and accurate, and thus improves the adaptability and accuracy of the large language model. In addition, the inventors further found that in order to ensure the quality and scale of the knowledge base, in actual applications, the large language model will integrate knowledge bases from different knowledge sources. Facing multi-source heterogeneous knowledge, these knowledge bases face security issues. Therefore, in order to improve data security while enhancing the adaptability and accuracy of the large language model, in the embodiments of this application, the target knowledge base that the user can access can be determined according to the user feature information and the knowledge base feature information, so as to determine the target knowledge content only in the target knowledge base, meet the security requirements of the knowledge base, improve data security, and also distinguish the knowledge usage rights according to the user identity to ensure the secure sharing of the knowledge base.

[0028] Next, the technical solutions in the embodiments of this application will be clearly and completely described with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of this application.

[0029] It should be noted that the use of user data may be involved in the embodiments of this application. In actual applications, user-specific personal data can be used in the solutions described herein within the scope permitted by applicable laws and regulations of the country of residence (for example, with the user's explicit consent, giving the user a practical notice, etc.).

[0030] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. The collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entrances are provided for the user to choose to authorize or refuse.

[0031] It should be noted that the technical solution of the embodiment of the present application is applicable to a network virtual environment. Generally, the described user refers to a "virtual user". A real user can register a user account on the server through registration to obtain a user identity in the network environment.

[0032] Figure 1 FIG. 4 is a flowchart of an embodiment of an information processing method provided by an embodiment of the present application. In an actual application, the technical solution of the embodiment of the present application can be applied to an information processing system composed of a user terminal and a server. The user terminal can sense the user input content, and the server is used to call a large language model for processing.

[0033] Among them, a connection is established between the user terminal and the server through a network. The network provides a medium for the communication link between the user terminal and the server. The network can include various connection types, such as wired, wireless communication links, or fiber optic cables, etc. The user terminal can interact with the server through the network to receive or send information, etc.

[0034] In an actual application, the user terminal can be a browser, an APP (Application), or a web application such as an H5 (HyperText Markup Language 5) application, or a light application (also known as a small program, a lightweight application program), or a cloud application, etc. The user terminal can be deployed in an electronic device and needs to rely on the device or certain apps in the device to run, etc. The electronic device can, for example, have a display screen and support information browsing, etc., such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, a desktop computer, a smart speaker, a smart watch, etc. Various other types of applications can usually be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc. The electronic device can refer to a device used by a user and having functions such as computing, Internet access, and communication required by the user, such as a mobile phone, a tablet computer, a personal computer, a wearable device, etc. The electronic device usually includes at least one processing component and at least one storage component. The electronic device may also include basic configurations such as a network card chip, an IO bus, and audio-visual components, which are not limited in the present application. Optionally, according to the implementation form of the electronic device, some peripheral devices may also be included, such as a keyboard, a mouse, an input pen, a printer, etc., which are not limited in the present application.

[0035] The server side may include servers that provide various services. It should be noted that the server side can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), as well as big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0036] The technical solution of this embodiment can be specifically executed by the server side. Of course, in other implementation manners, it can also be executed by the client side. The large language model trained by the server side can be deployed on the client side to provide more timely processing by the client side. This application does not limit this. Among them, the information processing method may include the following steps:

[0037] 101: Obtain the user input content.

[0038] 102: Combine the user feature information and the knowledge base feature information to determine the target knowledge base that the user can access.

[0039] Among them, the knowledge base is a special database for knowledge management, facilitating the collection, collation, and extraction of relevant domain knowledge. It is a data storage and organization method for storing, managing, and retrieving structured and unstructured data. The knowledge base usually includes contents such as documents, pictures, audio, video, web pages, etc., and can be retrieved, filtered, sorted, and classified by users.

[0040] In the technical solution of this application embodiment, the system may include multiple knowledge bases. These multiple knowledge bases can be provided by one or more knowledge base providers. The multiple knowledge bases correspond to different knowledge sources. Therefore, it is possible to combine the user feature information and the knowledge base feature information to determine the target knowledge base that the user can access from multiple knowledge bases with different knowledge sources.

[0041] That is, the embodiments of the present application can integrate knowledge bases from different knowledge sources. For example, aggregate professional data from multiple departments in an organization; or integrate external knowledge from different enterprises, institutions, or individuals, etc.; or integrate domain knowledge from different fields such as healthcare and education and training. This will face the situation where the knowledge sources are dispersed and have different permission attributes. If these knowledge bases are synthesized into a single integrated knowledge base, there is a risk of data leakage. The inventors have found that, for example, in the same organization, the data of Department A can only be shared within the organization, while the data of Department B can be opened to the outside. In addition, different employees in the same organization should be given different access permissions according to their levels. Therefore, in order to improve data security, in the embodiments of the present application, the user characteristic information and the knowledge base characteristic information can be combined first to determine the target knowledge bases that the user can access, and determine the target knowledge bases with access permissions. Among them, there can be at least one target knowledge base.

[0042] Among them, there are various implementation manners for combining the user characteristic information and the knowledge base characteristic information to determine the target knowledge bases that the user can access, which will be introduced in detail in the following embodiments.

[0043] 103: Search for target knowledge content matching the input content from the target knowledge bases.

[0044] 104: Generate feedback information using a large language model based on the target knowledge content and the input content.

[0045] In the embodiments of the present application, searching for target knowledge content matching the input content from the target knowledge bases, this target knowledge content can be used as the context information of the input content, so as to provide auxiliary information for the input content, making the feedback information generated by the large language model more accurate.

[0046] Optionally, a first prompt information can be generated according to the target knowledge content and the input content; the first prompt information is input into the large language model to obtain feedback information.

[0047] Among them, the prompt information is a form of input used to prompt or guide the large language model to give an output that meets expectations. It is used to indicate what actions the large language model should take or what output it should generate when performing a specific task. The prompt information is a natural language input, similar to a command or instruction, to let the large language model know what it needs to do.

[0048] Optionally, the target knowledge content and the input content can be filled into a prompt template to generate a first prompt message. For example, the target prompt template can be "Please answer the user's question 'XX' based on the following background knowledge 'AA'", where the 'AA' part is used to fill in the target knowledge content, and the 'XX' part is used to fill in the input content, thus generating the first prompt message. The target knowledge content can serve as the background knowledge of the input content, providing the required knowledge for the large language model, reducing factual errors in the reasoning of the large language model, giving full play to the powerful understanding and reasoning abilities of the large language model, improving the accuracy of the large language model, and thereby enhancing the user experience.

[0049] The large language model in the embodiments of the present application can be, for example, GPT-3 (Generative Pre-trained Transformer-3, the third generation of generative pre-trained model), GPT-4 (Generative General Pre-trained Transformer-4, the fourth generation of generative pre-trained model), BERT (Bidirectional Encoder Representation from Transformers, a bidirectional encoder model based on Transformers), Turing NLG (Turing Natural language Generation), and so on. The large language model can be widely applied in fields such as customer service, healthcare, and education and training. For example, the intelligent customer service robot built based on the large language model can automatically handle various service consultation questions of users; the intelligent Q&A system in the healthcare field built based on the large language model can assist doctors in answering patients' questions; the intelligent education software built based on the large language model can answer students' learning doubts based on the knowledge base. These applications utilize the powerful natural language processing ability of the pre-trained large language model, greatly improving the Q&A experience and efficiency. In actual applications, for example, the patient knowledge involved in the healthcare field often has a certain degree of privacy. Therefore, through the technical solution of the embodiments of the present application, different user access permissions can be granted, so that the knowledge scope that only the user can obtain only includes the knowledge base to which the user has access permissions. The large language model can summarize and provide answers based on the knowledge content that the user can access, which can not only ensure the accuracy of the large language model but also ensure that the user cannot access the knowledge without access permissions, thereby avoiding data leakage and ensuring data security.

[0050] As an optional implementation manner, the user feature information may include a user identifier, and the knowledge base feature information may include a knowledge base identifier; the determination of the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information may include:

[0051] According to the user identifier and the knowledge base identifier, search for the first permission mapping relationship to determine the target knowledge base to which the user has access permission; the first permission mapping relationship is used to indicate whether the user has access permission to the knowledge base.

[0052] Among them, the first permission mapping relationship can be preset to indicate whether there is access permission to the knowledge base.

[0053] In some embodiments, it may be determined according to the first determination result of whether the user meets the first access requirements respectively configured for different knowledge bases, the first permission mapping relationship between the user and different knowledge bases is determined, and the first permission mapping relationship is saved;

[0054] That is, the knowledge base provider can provide the knowledge base and can perform configuration operations, and the first access requirements can be configured by the knowledge base provider for the knowledge base. Among them, if the user meets the first access requirements of any knowledge base, it can be considered that the user has the right to access the knowledge base, otherwise it can be considered that the user has no right to access the knowledge base. The first permission mapping relationship can be saved according to the user identifier and the knowledge base identifier.

[0055] Among them, in addition to the user identifier, the user characteristic information may further include at least one attribute characteristic, such as user name, affiliated organization, rank, position, registration time, age, and gender, etc.

[0056] Therefore, it may be combined with the user identifier and / or at least one attribute characteristic to determine whether the user meets the first access requirements respectively configured for different knowledge bases.

[0057] In some embodiments, it may also be defined according to the user registration information the first permission mapping relationship between the user and different knowledge base providers, and the first permission mapping relationship is saved.

[0058] Among them, the user registration information can be generated when the user submits a registration request. In one implementation, the user characteristic information may include the user registration information, and the user registration information may include the user identifier and one or more attribute characteristics in the above user characteristic information, etc.

[0059] Among them, the first permission mapping relationship between the user and different knowledge bases can be defined by searching for the corresponding relationship between different user registration information and different knowledge bases pre-configured in the system, and the first permission mapping relationship is saved.

[0060] In addition, the user registration information may further include the expected knowledge base identifier, so that the knowledge base indicated by the expected knowledge base identifier can be determined as the knowledge base to which the user has access permission in combination with the expected knowledge base identifier, etc.

[0061] Among them, the first permission mapping relationship can be saved through a relational database. For example, it can be represented by a permission table, which may include fields such as user identification, knowledge base identification, and whether there is permission, so as to save the first permission item mapping relationship between different users and different knowledge bases into the permission table.

[0062] Of course, other information can also be saved through other data tables in the relational database. For example, the user table saves user-related information, which may include fields such as user identification, user name, user information addition time, user information update time, etc. Another example is that the knowledge base table is used to save knowledge base-related information, such as fields that may include knowledge base identification, knowledge base name, knowledge base information addition time, knowledge base information update time, first access requirement, etc.

[0063] In addition, the first permission mapping relationship can also be marked through other types of databases such as key-value databases, and this application does not limit this.

[0064] As another alternative, the user characteristic information may include user identification, and the knowledge base characteristic information may include knowledge base identification; combining the above user characteristic information and knowledge base characteristic information, determining the target knowledge base that the user can access may include:

[0065] According to the user identification and the knowledge base provider identification, search for the second permission mapping relationship to determine the knowledge base provider for which the user has access permission; regard all at least one knowledge base provided by the target knowledge base provider as the target knowledge base.

[0066] Among them, the second permission mapping relationship is used to represent whether the user has access permission to the knowledge base provider.

[0067] In practical applications, a knowledge base provider can provide one or more knowledge bases. In the embodiments of this application, the permission mapping relationship between the user and the knowledge base provider can be preset. If the user has permission to access a certain knowledge base provider, then the user can access all the knowledge bases provided by that knowledge base provider.

[0068] In some embodiments, it may be determined according to the second determination result of whether the user meets the second access requirement corresponding to any knowledge base provider, and the second permission mapping relationship between the user and the knowledge base provider is determined and saved.

[0069] The knowledge base provider can provide the knowledge base and perform configuration operations. The second access requirement can be configured by the knowledge base provider for itself. Among them, if a user meets the second access requirement of any knowledge base provider, it can be considered that the user has the permission to access all the knowledge bases provided by the knowledge base provider. Otherwise, it can be considered that the user has no permission to access any of the knowledge bases provided by the knowledge base provider. The second permission mapping relationship can be saved corresponding to the user identifier and the knowledge base provider identifier, and this second permission mapping relationship is used to indicate whether the user has the permission to access the knowledge base provider, etc.

[0070] Among them, in addition to the user identifier, the user characteristic information can also include at least one attribute characteristic, such as user name, affiliated organization, rank, position, registration time, age, and gender, etc.

[0071] Therefore, it can be combined with the user identifier and / or at least one attribute characteristic to determine whether the user meets the second access requirements respectively configured by different knowledge base providers.

[0072] In addition, in some embodiments, it can also be based on the user registration information to define the second permission mapping relationship between the user and different knowledge base providers, and save the second permission mapping relationship.

[0073] Among them, the user registration information can be created when the user submits a registration request. In one implementation, the user characteristic information can include at least part of the information in the user registration information. For example, the user registration information can include the user identifier and one or more attribute characteristics in the above user characteristic information, etc.

[0074] Among them, the second permission mapping relationship between the user and different knowledge base providers can be defined and saved by looking up the corresponding relationship between different user registration information and different knowledge base providers pre-configured in the system.

[0075] Among them, the second permission mapping relationship can be saved through a relational database. For example, it can be represented by a knowledge base provider table. The knowledge base provider table can include fields such as user identifier, knowledge provider identifier, and whether there is permission, etc., so as to save the second permission mapping relationship between different users and different knowledge base providers into the knowledge base provider table.

[0076] In addition, the second permission mapping relationship can also be saved through other types of databases such as key-value databases, etc. This application does not limit this.

[0077] As another alternative, the user feature information may include a user identifier and / or at least one attribute feature; the knowledge base feature information may include a knowledge base identifier, a knowledge base provider identifier, a first access requirement corresponding to the knowledge base, and a second access requirement corresponding to the knowledge base provider to which the knowledge base belongs, etc.

[0078] The at least one attribute feature may be, for example, one or more of a user name, an affiliated organization, a user level, a position, an age, and a gender.

[0079] The above determination of the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information may include:

[0080] Combining the user feature information to determine whether the user meets at least one determination condition in the first access requirement corresponding to any knowledge base and / or the second access requirement corresponding to any knowledge base provider;

[0081] Regarding at least one knowledge base corresponding to the user feature information meeting the first access requirement, and / or at least one knowledge base provided by the knowledge base provider corresponding to the user feature information meeting the second access requirement, respectively as the target knowledge base.

[0082] That is, in the embodiments of the present application, the first permission mapping relationship or the second permission mapping relationship may be determined in advance based on the first access requirement or the second access requirement, etc. By looking up the first permission mapping relationship or the second permission mapping relationship based on the user identifier and the knowledge base identifier, it can be determined whether the user has the access permission to a certain knowledge base or a certain knowledge base provider.

[0083] Of course, it may also be to determine the target knowledge base that the user can access in real time based on the first access requirement or the second access requirement.

[0084] From the above several alternative methods, it can be seen that the first access requirement or the second access requirement can be obtained by configuring the knowledge base provider. Therefore, in some embodiments, the device may further include:

[0085] Sending a first registration prompt message to the knowledge base provider; generating knowledge base registration information according to the first registration request sent by the knowledge base provider.

[0086] Among them, the knowledge base registration information may include a knowledge base identifier, a knowledge base name, a knowledge base provider identifier to which the knowledge base belongs, and the above first access requirement and / or second access requirement, etc. The knowledge base feature information may include the knowledge base registration information.

[0087] In some embodiments, the method may further include:

[0088] Sending a second registration prompt message to the user terminal; creating user registration information according to the registration request sent by the user terminal.

[0089] The user registration information may include user identification, user identification, etc. The user characteristic information may include user registration information.

[0090] In some embodiments, the first access requirement or the second access requirement may include one or more of the following access conditions:

[0091] having a configured predetermined user identification;

[0092] having one or more configured predetermined attribute characteristics;

[0093] as well as,

[0094] Within the validity period of the permission set by one or more of the above access conditions.

[0095] That is, the knowledge base provider can configure the knowledge base development scope through user identification or attribute characteristics. For example, if a user has user identification X set by the knowledge base provider, the user can access all knowledge bases under the knowledge base provider; for another example, if the user has attribute characteristics 1 and attribute characteristics 2 set by the knowledge base provider for knowledge base A, such as the user's position is general manager and belongs to the product department, the user can access the knowledge base A; for another example, if the user has attribute characteristics 1 and attribute characteristics 2 set by the knowledge base provider for knowledge base A, and the access conditions are within the validity period of the authority, then it can be considered that the user can access the knowledge base A.

[0096] In addition, as another optional manner, the above-mentioned determination of the target knowledge base that the user can access by combining the user characteristic information and the knowledge base characteristic information may include:

[0097] The knowledge base whose matching degree between knowledge base feature information and user feature information meets the matching requirements is regarded as the target knowledge base that the user can access.

[0098] The knowledge base feature information and the user feature information may be vectorized and converted into vector features, and then the degree of matching is determined by calculating the vector distance. For example, the degree of matching may be greater than a predetermined value.

[0099] Of course, the matching degree between the knowledge base feature information and the user feature information can also be calculated through a matching model. The matching model can be trained in advance based on sample knowledge base feature information and sample user feature information that matches it.

[0100] In the case where there is a knowledge base that the user can access, the input content can be directly input into the large language model to obtain feedback information.

[0101] In addition, in the case where there is no knowledge base accessible to the user, in order to further improve the accuracy of the large language model, in some embodiments, the method may further include:

[0102] When the determination result of the target knowledge base is empty, based on the input content, call at least one search engine to perform a web search; extract the target search content from the search results; according to the target search content and the input content, use the large language model to obtain feedback information of the input content.

[0103] That is, it is possible to call at least one search engine for searching based on the input content through web search. The search results may include multiple web links, and the target search content can be extracted from the web content corresponding to at least one of the web links.

[0104] Among them, it may be to generate a second prompt message according to the target search content and the input content; use the large language model to process the second prompt message to obtain the feedback information of the input content.

[0105] Among them, it may be to input the target search content and the input content into a prompt template to generate a second prompt message.

[0106] Embodiments of the present application may also adopt the method of searching the entire network data, so as to combine the target search content as auxiliary information for the input content, thereby improving the processing accuracy of the large language model.

[0107] In addition, in the case where there is no knowledge base accessible to the user, in order to further improve the accuracy of the large language model, in some embodiments, the method may further include:

[0108] When the determination result of the target knowledge base is empty, use an intent recognition model to identify the target intent corresponding to the input content; according to the target intent and the input content, use the large language model to obtain the feedback information of the input content.

[0109] That is, embodiments of the present application can identify the target intent corresponding to the input content, and use the target intent as the context information of the input content to improve the processing accuracy of the large language model.

[0110] Among them, it may be to generate a third prompt message according to the target intent and the input content; use the large language model to process the third prompt message to obtain the feedback information of the input content.

[0111] The target intent and the input content can be filled into a preset prompt template to generate a third prompt message.

[0112] In addition, in some embodiments, the method may further include:

[0113] In the case where the determination result of the target knowledge base is empty, use an intent recognition model to recognize the target intent corresponding to the input content;

[0114] In the case where the determination result of the target knowledge base is empty, based on the input content, call at least one search engine for web search; extract the target search content from the search results;

[0115] According to the target search content, target intent, and input content, use a large language model to generate feedback information.

[0116] Among them, it can be to generate a fourth prompt message with the target search content, target intent, and input content, and use the large language model to process the fourth prompt message to obtain the feedback information of the input content.

[0117] The target search content and target intent can be used as the context information of the input content at the same time, adding more prior knowledge to the input content, making the obtained feedback information more accurate, thereby improving the accuracy of the large language model for assistance.

[0118] In addition, in some embodiments, after finding the target knowledge content matching the input content from the target knowledge base, the target search content and target intent can be combined to enhance the input content. Therefore, according to the target knowledge content, target search content, target intent, and input content, use a large language model to generate feedback information.

[0119] In some embodiments, the above-mentioned finding of the target knowledge content matching the input content from the target knowledge base may include:

[0120] Convert the input content into a feature vector; perform a similarity search on the vector data corresponding to the knowledge base content of the target knowledge base with the feature vector to determine the target knowledge content corresponding to the target vector data that meets the similarity requirement with the feature vector.

[0121] That is, the target knowledge content can be determined from the target knowledge base through vector retrieval technology.

[0122] Among them, the knowledge content in the knowledge base can be pre-converted into vector data and saved in a vector database, so that the target vector data matching the feature vector can be found from the vector database, and then the target knowledge content corresponding to the vector data can be determined from the knowledge base.

[0123] Among them, the similarity between the feature vector and different vector data can be determined by calculating the vector distance, etc. The similarity requirement can be, for example, that the similarity is greater than a specified threshold, etc.

[0124] Among them, the vector data can be obtained by first converting the knowledge content into text format, switching it into multiple text blocks through tokenization technology, and then vectorizing each text block to obtain the vector form of each text block. The text block, the corresponding digital vector, and text block-related information such as the text to which the text block belongs and the vector database can be correspondingly stored in the vector database. Thus, the corresponding target knowledge content can be determined according to the text to which the text block corresponding to the target vector data that meets the similarity requirement with the feature vector belongs.

[0125] In a practical application, the technical solution of the embodiment of the present application can be applied to an intelligent question-and-answer scenario. The content input by the user is the content of the user's question; the feedback information is the response content output by the large language model. Figure 2 The processing schematic diagram of the technical solution of the embodiment of the present application in a practical application is shown. Based on the knowledge base provided by the knowledge base provider, the permission mapping relationship (the first permission mapping relationship and / or the second permission mapping relationship) 201 with the user can be first determined.

[0126] After that, the knowledge content in the knowledge base can be converted into text format such as TXT (Text) format text 202; the text is segmented to form file blocks 203. Each text block can include, for example, 50 - 1000 words or the like; then each text block can be vectorized, and the content of each file block is converted into vector data in vector form 204; the text block, the corresponding vector data, and text block-related information are stored in the vector database 205.

[0127] Before the user asks a question, registration can be first performed to generate user registration information, so that the above permission mapping relationship can be determined in combination with the user registration information and the knowledge base feature information, etc.

[0128] After that, the user's question content is obtained; the user's question content can be vectorized and converted into the corresponding feature vector 206.

[0129] After that, based on the permission mapping relationship, the target knowledge base to which the user has access permission can be first determined, and the vector data corresponding to the target knowledge base is obtained from the vector database. Then, the similarity search 207 can be performed between the feature vector and the vector data corresponding to the target knowledge base to find the target vector data with high semantic similarity, and then the target knowledge content corresponding to the target vector data is determined 208.

[0130] The target knowledge base content and the user's question content can be input into the template, so as to generate the first prompt information 209 finally input into the large language model.

[0131] The first prompt information is input into the large language model 210, and thus the response content can be obtained.

[0132] Among them, the method of vectorizing the input content or knowledge content can be implemented by, for example, the One Hot Model, the Bag of Words Model, the Term Frequency-Inverse Document Frequency (TF-IDF), the N-Gram model, the Word2vec model, the Doc2vec model, etc. This application does not make specific limitations on this.

[0133] In some embodiments, the above-mentioned obtaining of the user input content may be obtaining the user input content sent by the user terminal;

[0134] After generating the feedback information using the large language model, the method may further include: sending the feedback information to the user terminal for the user terminal to display the feedback information.

[0135] For ease of understanding, as Figure 3 shown, a schematic diagram of scenario interaction in an actual application of an embodiment of this application is shown.

[0136] The server 301 may send a first registration prompt message to the client 302. The client 302 may face the knowledge base provider. The knowledge base provider may perform registration operations, configuration operations, etc. based on the first registration prompt message and send a first registration request to the server 301. The server 301 may generate knowledge base registration information accordingly. The knowledge base feature information may include at least part of the information in the knowledge base registration information, etc.

[0137] The server 301 may send a second registration prompt message to the user terminal 303. The user may perform registration operations, configuration operations, etc. based on the second registration prompt message and send a second registration request to the server 301. The server 301 may generate user registration information accordingly. The user feature information may include at least part of the information in the user registration information.

[0138] The user terminal 303 may receive the user input content and send the user input content to the server 301.

[0139] The server 301 may combine the user feature information and the knowledge base feature information to determine the target knowledge base that the user can access, search for the target knowledge content matching the input content from the target knowledge base, and then may call the large language model to generate feedback information according to the target knowledge content and the input content. The specific execution operations of the server 301 can be seen in the corresponding embodiments described above, for example Figure 2 as described in the embodiments shown. Details will not be repeated here.

[0140] The feedback information generated by the server 301 can be sent to the client 303 for the client 303 to display the feedback information to the user and complete the intelligent question-answering operation.

[0141] In some embodiments, in order to further improve data security, etc., determining the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information may include:

[0142] Import the input content into the trusted execution environment;

[0143] In the trusted execution environment, execute operations such as determining the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information.

[0144] Searching for the target knowledge content matching the input content from the target knowledge base as described above is also to search for the target knowledge content matching the input content from the target knowledge base in the trusted execution environment

[0145] Generating the feedback information according to the target knowledge content and the input content using the large language model as described above can also be generating the feedback information according to the target knowledge content and the input content using the large language model in the trusted execution environment.

[0146] Of course, the vectorization of the knowledge content and the vectorization of the user input content described above can also be performed in the trusted execution environment.

[0147] In addition, the method may further include: exporting the feedback information from the trusted execution environment. Thus, the feedback information can be sent to the client for the user to view, etc.

[0148] Among them, the trusted execution environment can be, for example, a sandbox created in the device system; or the trusted execution environment can be created based on the security unit built into the CPU, such as Intel SGX ( Software Guard Extensions), ARM Trustzone (an embedded platform security technology), or AMD PSP (AMD Platform Security Processor), etc.; or it can be created by an external security element (SE, Secure Element) configured in the device, etc. The present application does not specifically limit this.

[0149] Among them, the trusted execution environment can be a running environment that coexists with the operating system. It can run independently in the device, can ensure the security of data and code in the trusted execution environment, and achieve the purpose of isolating from the external execution environment. The external execution environment referred to herein can be the operating system of the device.

[0150] Through the technical solution of the embodiments of the present application combined with the knowledge content of the knowledge base, the accuracy of the large language model is improved, and hallucinations inconsistent with the input content or facts are avoided. In addition, the permission mapping relationship between different knowledge bases and different users can be defined, so that the knowledge scope that can be obtained when the user asks questions only includes the knowledge bases that the user has permission to access. The large language model will summarize the question and answer according to the content of the knowledge bases accessible to the user, ensuring that the user cannot access the knowledge without permission, thereby avoiding data leakage and ensuring data security.

[0151] Figure 4 FIG. 5 is a schematic structural diagram of an embodiment of an information processing device provided by an embodiment of the present application. The device may include:

[0152] A content acquisition module 401, configured to acquire user input content;

[0153] A knowledge base determination module 402, configured to determine a target knowledge base that the user can access by combining user feature information and knowledge base feature information;

[0154] A knowledge search module 403, configured to search for target knowledge content matching the input content from the target knowledge base;

[0155] An information generation module 404, configured to generate feedback information by using a large language model according to the target knowledge content and the input content.

[0156] In some embodiments, the user feature information includes a user identifier; the knowledge base feature information includes a knowledge base identifier; the knowledge base determination module is specifically configured to search for a first permission mapping relationship according to the user identifier and the knowledge base identifier to determine the target knowledge base to which the user has access permission; the first permission mapping relationship is used to indicate whether the user has access permission to the knowledge base.

[0157] In some embodiments, the device may further include:

[0158] A first relationship determination module, configured to determine the first permission mapping relationship between the user and different knowledge bases according to the first determination result of whether the user meets the first access requirements respectively configured for different knowledge bases, and save the first permission mapping relationship; or, define the first permission mapping relationship between the user and different knowledge bases according to the user registration information, and save the first permission mapping relationship.

[0159] In some embodiments, the user feature information includes a user identifier; the knowledge base feature information includes a knowledge base provider identifier; the knowledge base determination module is specifically configured to find a second permission mapping relationship according to the user identifier and the knowledge base provider identifier, so as to determine a target knowledge base provider to which the user has access permission; the second permission mapping relationship is used to indicate whether the user has access permission to the knowledge base provider; and use at least one knowledge base provided by the target knowledge base provider as the target knowledge base.

[0160] In some embodiments, the apparatus may further include:

[0161] A second relationship determination module, configured to determine a second permission mapping relationship between the user and the knowledge base provider according to a second determination result of whether the user meets a second access requirement corresponding to any knowledge base provider, and save the second permission mapping relationship.

[0162] Alternatively, define a second permission mapping relationship between the user and different knowledge base providers according to the user registration information, and save the second permission mapping relationship.

[0163] In some embodiments, the user feature information includes a user identifier and / or at least one attribute feature; the knowledge base feature information includes a knowledge base identifier, a knowledge base provider identifier, a first access requirement corresponding to the knowledge base, and a second access requirement corresponding to the knowledge base provider to which the knowledge base belongs;

[0164] The knowledge base determination module is specifically configured to combine the user feature information to determine whether the user meets at least one determination condition in a first access requirement corresponding to any knowledge base and a second access requirement corresponding to any knowledge base provider; use at least one knowledge base corresponding to the first access requirement that the user feature information meets, and / or at least one knowledge base provided by the knowledge base provider corresponding to the second access requirement that the user feature information meets, as the target knowledge base respectively.

[0165] In some embodiments, the apparatus may further include:

[0166] A first registration module, configured to send a first registration prompt message to the knowledge base provider;

[0167] Generate knowledge base registration information according to a first registration request sent by the knowledge base provider; the knowledge base registration information includes a first access requirement and / or a second access requirement.

[0168] In some embodiments, the apparatus may further include:

[0169] A second registration module, configured to send a second registration prompt message to the user terminal; create user registration information according to a registration request sent by the user terminal; the user feature information includes at least part of the information in the user registration information.

[0170] In some embodiments, the first access requirement or the second access requirement includes one or more of the following access conditions:

[0171] Having a configured predetermined user identifier;

[0172] Having one or more configured predetermined attribute characteristics;

[0173] And,

[0174] Within the valid time of the permissions set by the above one or more access conditions.

[0175] In some embodiments, the above knowledge base determination module may specifically be used to use the knowledge base whose matching degree between the knowledge base feature information and the user feature information meets the matching requirement as the target knowledge base that the user can access.

[0176] In some embodiments, the above information generation module is specifically used to generate a first prompt message according to the target knowledge content and the input content; input the first prompt message into a large language model to obtain feedback information.

[0177] In some embodiments, the device may further include:

[0178] A first processing module, configured to, when the determination result of the target knowledge base is empty, perform a network search by invoking at least one search engine based on the input content; extract target search content from the search results; and obtain feedback information of the input content by using a large language model according to the target search content and the input content.

[0179] In some embodiments, the device may further include:

[0180] When the determination result of the target knowledge base is empty, identify the target intent corresponding to the input content by using an intent recognition model; and obtain feedback information of the input content by using a large language model according to the target intent and the input content.

[0181] In some embodiments, the above knowledge search module is specifically used to convert the input content into a feature vector;

[0182] Perform a similarity search on the vector data corresponding to the knowledge base content of the target knowledge base with the feature vector to determine the target knowledge content corresponding to the target vector data that meets the similarity requirement with the feature vector.

[0183] In some embodiments, the device may further include:

[0184] A security execution module, configured to import the input content into a trusted execution environment; wherein, the knowledge base is deployed in the trusted execution environment;

[0185] Specifically, the above knowledge base determination module determines the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information in the trusted execution environment.

[0186] The device may further include an information export module for exporting the feedback information out of the trusted execution environment.

[0187] Figure 4 The described information processing device can execute Figure 1 the information processing method described in the illustrated embodiment. The implementation principle and technical effects will not be elaborated further. For the information processing device in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiment related to the method, and will not be elaborated here in detail.

[0188] An embodiment of the present application also provides a computing device, as Figure 5 shown. The computing device may include a storage component 501 and a processing component 502;

[0189] The storage component 501 stores one or more computer instructions, where the one or more computer instructions are called and executed by the processing component 502 to implement the information processing method described in the illustrated embodiment. Figure 1

[0190] Of course, the computing device may necessarily also include other components, such as an input / output interface, a display component, a communication component, etc.

[0191] The input / output interface provides an interface between the processing component and the peripheral interface module, and the above peripheral interface module may be an output device, an input device, etc. The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0192] Among them, the processing component may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components for executing the above method.

[0193] ​The storage component is configured to store various types of data to support the operations of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0194] The display component can be an electroluminescent (EL) element, a liquid crystal display, or a microdisplay with a similar structure, or a retinal direct display or a similar laser scanning display.

[0195] It should be noted that in the case where the above computing device is deployed on the server side, it can be a physical device or an elastic computing host provided by a cloud computing platform, etc. It can be implemented as a distributed cluster composed of multiple servers or terminal devices, or can be implemented as a single server or a single terminal device.

[0196] In the case where the above computing device is deployed on the client side, it can be specifically implemented as an electronic device. The electronic device can refer to a device used by a user and having functions such as computing, Internet access, and communication required by the user. For example, it can be a mobile phone, a tablet computer, a personal computer, a wearable device, etc.

[0197] The embodiment of the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a computer, it can implement the above Figure 1 information processing method described in the illustrated embodiment. The computer-readable medium can be included in the electronic device described in the above embodiment; or it can exist separately without being assembled into the electronic device.

[0198] The embodiment of the present application also provides a computer program product, which includes a computer program carried on a computer-readable storage medium, and when the computer program is executed by a computer, it can implement the information processing method described in the illustrated embodiment as above Figure 1 In such an embodiment, the computer program can be downloaded and installed from the network and / or installed from a removable medium. When the computer program is executed by a processor, it executes various functions defined in the system of the present application.

[0199] It should be noted that the use of user data may be involved in the embodiment of the present application. In actual applications, it can be used in the solutions described herein within the scope permitted by applicable laws and regulations in compliance with the requirements of applicable laws and regulations of the country where it is located (for example, with the user's explicit consent, giving the user a practical notice, etc.).

[0200] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0201] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0202] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or equivalently replace some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An information processing method, characterized in that, Including: Obtain user input content; Combine user characteristic information and knowledge base characteristic information to determine the target knowledge base that the user can access; Search for target knowledge content that matches the input content from the target knowledge base; Generate feedback information using a large language model based on the target knowledge content and the input content.

2. The method according to claim 1, wherein The user characteristic information includes a user identifier; the knowledge base characteristic information includes a knowledge base identifier; the combining of the user characteristic information and the knowledge base characteristic information to determine the target knowledge base that the user can access includes: Search for a first permission mapping relationship based on the user identifier and the knowledge base identifier to determine the target knowledge base to which the user has access permission; the first permission mapping relationship is used to indicate whether the user has access permission to the knowledge base.

3. The method according to claim 2, wherein Also included is: Determine the first permission mapping relationship between the user and different knowledge bases according to the first determination result of whether the user meets the first access requirements respectively configured for different knowledge bases, and save the first permission mapping relationship; Or, define the first permission mapping relationship between the user and different knowledge bases according to the user registration information, and save the first permission mapping relationship.

4. The method according to claim 1, wherein The user characteristic information includes a user identifier; the knowledge base characteristic information includes a knowledge base provider identifier; the combining of the user characteristic information and the knowledge base characteristic information to determine the target knowledge base that the user can access includes: Search for a second permission mapping relationship based on the user identifier and the knowledge base provider identifier to determine the target knowledge base provider to which the user has access permission; the second permission mapping relationship is used to indicate whether the user has access permission to the knowledge base provider; Use at least one knowledge base provided by the target knowledge base provider as the target knowledge base.

5. The method according to claim 4, characterized in that, Also included is: Determine the second permission mapping relationship between the user and the knowledge base provider according to the second determination result of whether the user meets the second access requirements corresponding to any knowledge base provider, and save the second permission mapping relationship; Or, define the second permission mapping relationship between the user and different knowledge base providers according to the user registration information, and save the second permission mapping relationship.

6. The method according to claim 1, characterized in that The user characteristic information includes a user identifier and / or at least one attribute characteristic; the knowledge base characteristic information includes a knowledge base identifier, a knowledge base provider identifier, a first access requirement corresponding to the knowledge base, and a second access requirement corresponding to the knowledge base provider to which the knowledge base belongs; the combining of the user characteristic information and the knowledge base characteristic information to determine the target knowledge base that the user can access includes: Combine the user characteristic information to determine whether the user meets at least one determination condition among the first access requirements corresponding to any knowledge base and the second access requirements corresponding to any knowledge base provider; Use at least one knowledge base corresponding to the first access requirement that the user characteristic information meets, and / or at least one knowledge base provided by the knowledge base provider corresponding to the second access requirement that the user characteristic information meets, as the target knowledge base respectively.

7. The method according to claim 3, 5 or 6, characterized in that, Also included is: Send a first registration prompt message to the knowledge base provider; Generate knowledge base registration information according to the first registration request sent by the knowledge base provider; The knowledge base registration information includes a first access requirement and / or a second access requirement.

8. The method according to claim 3, 5 or 6, characterized in that It further includes: Send a second registration prompt message to the user terminal; Create user registration information according to the registration request sent by the user terminal; the user feature information includes at least part of the information in the user registration information.

9. The method according to claim 6, characterized in that, The first access requirement or the second access requirement includes one or more of the following access conditions: Having a configured predetermined user identifier; Having one or more configured predetermined attribute features; And, Within the valid time of the permissions set by the above one or more access conditions.

10. The method according to claim 1, characterized in that, It further includes: In the case where the determination result of the target knowledge base is empty, based on the input content, call at least one search engine for network search; Extract target search content from the search results; According to the target search content and the input content, use the large language model to obtain feedback information of the input content.

11. The method according to claim 1, characterized in that, It further includes: In the case where the determination result of the target knowledge base is empty, use an intent recognition model to identify the target intent corresponding to the input content; According to the target intent and the input content, use the large language model to obtain feedback information of the input content.

12. The method according to claim 1, wherein The finding of the target knowledge content matching the input content from the target knowledge base includes: Convert the input content into a feature vector; Perform a similarity search on the feature vector in the vector data corresponding to the knowledge base content of the target knowledge base to determine the target knowledge content corresponding to the target vector data that meets the similarity requirement with the feature vector.

13. The method according to claim 1, wherein It further includes: Import the input content into a trusted execution environment; The determination of the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information includes: In the trusted execution environment, execute to determine the target knowledge base that the user can access by combining the user feature information and the knowledge base feature information; After generating the feedback information using the large language model, the method further includes: Export the feedback information from the trusted execution environment.

14. A computing device, characterized in that, It includes a processing component and a storage component; The storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the information processing method according to any one of claims 1 to 13.

15. A computer-readable storage medium, characterized in that, A computer program is stored, and when the computer program is executed by the computer, it implements the information processing method according to any one of claims 1 to 13.

Images