Information processing method and system based on large language model and AI intelligent assistant

Abstract

The invention discloses an information processing method and system based on a large language model and an AI intelligent assistant. The method comprises the steps that the large language model is trained, and an application service set is preset; inputting information into the large language model; the large language model extracts a target application service matched with the information intention from the application service set according to the information; the large language model extracts input parameters from the information according to the target application service and outputs the input parameters to the target application service; the target application service performs operation processing according to the input parameters; and the large language model extracts a result from the target application service and outputs the result. Through a mode of combining a large language model and application service, the application effect of traditional application software is improved, the application space is expanded, and the use threshold is reduced; the use depth of a traditional language model can be greatly improved, and the language model can be expanded to various business scenes, so that complex professional problems can be solved in the dialogue process.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to an information processing method, system and AI intelligent assistant based on a large language model. Background Art

[0002] With the rapid development of machine learning (ML) technology, artificial intelligence assistants based on large language models (LLM) have been widely used in many scenarios. Such intelligent assistants can solve some common and simple needs of users in daily life and work through natural language dialogue.

[0003] However, many current assistants are mainly based on question-and-answer dialogues, and their functions are usually limited to simple language-related tasks. When encountering tasks that require complex processing, relying solely on large language models is often insufficient and must be combined with traditional software applications to complete.

[0004] In contrast, traditional application software is mainly used to solve problems in specific scenarios. Its functions are more specialized and vertical, and are suitable for users with professional backgrounds. However, this type of software has a high learning cost for novices, especially when multiple applications are involved. Users need to make their own judgments and switch between different software to complete tasks, which undoubtedly increases the complexity and threshold of use. Summary of the invention

[0005] The purpose of this application is to combine traditional artificial intelligence assistants with application software, using the respective advantages of large language models and traditional application software to solve the problem that large language models cannot solve professional problems or that traditional application software has a high threshold for use.

[0006] Based on this, in order to achieve the above advantages or other advantages and purposes of the present application, the present application provides an information processing method based on a large language model, including: Train large language models and pre-set application service sets; The large language model obtains information input by the user; The large language model analyzes the information and extracts information intent, and extracts a target application service from the application service set according to the information intent; The large language model extracts input parameters required by the target application service from the information, and outputs the input parameters to the target application service; The target application service performs calculation processing based on the input parameters and outputs a result.

[0007] According to an embodiment of an application, the large language model analyzes the information and extracts the information intent, and extracts the target application service from the application service set according to the information intent, including: Step S311: The large language model analyzes the information to extract intent; Step S312: matching application services in the application service set according to the intent; Step S313: determine whether the application service is matched, if yes, proceed to step S314, if not, proceed to step S315; Step S314: extracting the matched application service as the target application service; Step S315: the large language model outputs intention guidance information according to the intention; Step S316: supplement the information according to the intention guidance information and proceed to step S311.

[0008] According to an embodiment of an application, the large language model analyzes the information and extracts the information intent, and extracts the target application service from the application service set according to the information intent, including: Step S321: presetting a scenario task set, wherein the scenario tasks in the scenario task set correspond to one or more application services; Step S322: The large language model analyzes the information to extract intent; Step S323, matching a target scenario task in the scenario task set according to the intention; Step S324: determine whether the target scene task is matched, if yes, proceed to step S325, if no, proceed to step S326; Step S325, extracting the application service corresponding to the target scenario task as the target application service; Step S326: the large language model outputs intention guidance information according to the intention; Step S337, supplement the information according to the intention guidance information, and proceed to step S322.

[0009] According to an embodiment of an application, the information processing method further includes: Preset information supplement times N1; If the number of times the information is supplemented reaches N1, the large language model processes and outputs the information.

[0010] According to an embodiment of an application, the process of extracting the input parameters required by the target application service from the information by the large language model and outputting them to the target application service includes: Step S411, the large language model extracts input parameters from the information; Step S412: determine whether the input parameters meet the target application service computing requirements, if yes, proceed to step S413, if no, proceed to step S414; Step S413: output the input parameters to the target application service; Step S414, the large language model outputs parameter guidance information; Step S415, supplement the information according to the parameter guidance information, and proceed to step S411.

[0011] According to an embodiment of an application, the information processing method further includes: The preset parameter supplement times N2; If the number of parameter supplementation times reaches N2, the large language model processes and outputs the information.

[0012] According to an embodiment of an application, the information processing method further includes: Preset application packages; When or after outputting the result, calling the target application software corresponding to the target application service; The output result or input information is deeply processed using the target application software.

[0013] Another embodiment of the present invention discloses an information processing system based on a large language model, comprising: An information input window, wherein the information input window is used for information input; An application service module, which is used to store a preset application service set, perform calculations based on input parameters and output results; A large language model module, wherein the large language model module stores a trained large language model and is used to obtain information inputted in the information input window, analyze the information and extract information intent, extract a target application service in the application service module according to the information intent, extract input parameters required for the target application service from the information, and output the input parameters to the application service module; A result display window is used to display the result output by the application service module.

[0014] According to an embodiment of an application, the large language model module is also used to process the information and output a result.

[0015] According to an embodiment of an application, the information processing system based on a large language model also includes: an application software module, which stores a preset application software package, and is used to display an operation interface after the large language model module is called and perform in-depth processing on the output results or input information.

[0016] Another embodiment of the present invention discloses an AI intelligent assistant, which includes a computer-readable storage medium, in which at least one instruction is stored, and the instruction is read by a processor to enable the AI ​​intelligent assistant to execute any of the methods described above.

[0017] Beneficial effects: The information processing method, system and AI intelligent assistant based on the large language model disclosed in the present application can, by combining the large language model with application services, improve the application effect of traditional application software, expand the application space, and be more intelligent, thereby lowering the threshold for use; on the other hand, it can greatly enhance the depth of use of the traditional language model and expand it to a variety of business scenarios, enabling it to solve complex professional problems during the conversation process, greatly enhancing the application surface and user experience of the traditional language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A flow chart of an information processing method based on a large language model provided for one embodiment of the present application; Figure 2 A flowchart of step S3 of an information processing method based on a large language model provided for one embodiment of the present application; Figure 3 A flowchart of step S3 of an information processing method based on a large language model provided in another embodiment of the present application; Figure 4 A flowchart of step S4 of an information processing method based on a large language model is provided for one embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0021] Considering the existing technologies, some AI question-and-answer assistants based on large language models are mainly in the form of question-and-answer dialogues, which can meet some simple demands related to dialogue or language. The output results of user questions and answers are mainly text content. Taking scientific research scenarios as an example, AI question-and-answer assistants can help scientific researchers find information, translate texts, and polish writing, thereby improving their work efficiency. However, once it involves more complex professional task processing, large language models cannot meet the requirements of richer scientific research work.

[0022] For example, in scientific research scenarios, the scientific research process includes searching for papers, paper management, paper reading, data extraction, data analysis, data visualization, paper writing, etc., not just simple language-related tasks. Currently, AI scientific research assistants on the market are more limited to some auxiliary tasks, and are not very practical, and cannot meet the needs of scientific researchers for professional application software in their daily work.

[0023] For scientific researchers, the learning cost of using professional application software is high, especially when multiple applications are involved. Users need to make their own judgments and switch between different software to complete tasks, which undoubtedly increases the complexity and threshold of use.

[0024] In view of this, the inventor selected the strengths of both and tried to integrate them, and innovatively proposed an information processing method, system and AI intelligent assistant based on a large language model to solve the above problems.

[0025] The present invention discloses some embodiments to explain in detail the information processing method, system and AI intelligent assistant based on the large language model. The following embodiments provide a user interaction interface through a PC online product program, so that the user can interact with the system through the user interaction interface. The interaction interface includes commonly used interaction interfaces such as an input dialogue window, a result output window, and a quick operation virtual key, and also includes a virtual key matching the corresponding function disclosed in the embodiment of the present invention.

[0026] An embodiment of the present invention discloses an information processing method based on a large language model. Figure 1 As shown, the information processing method includes: Step S1, train a large language model and preset an application service set. Among them, the large language model is used to interact with users in a human-computer dialogue scenario, and its training process can be understood and implemented based on prior art knowledge, so this embodiment will not be described in detail. The application service set is encapsulated by the basic functions of multiple application software, which is used to provide basic capabilities for the operation of the input parameters. It can also be understood that the basic functions of some application software are integrated into the application service set. The application service set can be used as part of the large language model, or the application service set can be set separately and connected to the large language model for easy retrieval.

[0027] Step S2: The large language model obtains information input by the user, including text information, picture information, voice information and video information.

[0028] Take document management and reading in scientific research scenarios as an example. For example, users can input information by text input, uploading PDF documents, etc. After the user enters the information, the system can obtain the user input information. For example, in one example, the user uploaded two PDF papers and entered text "conduct an in-depth comparison and analysis of the two studies from multiple dimensions such as research purpose, theoretical framework, main findings, application and significance". These text information and document information are the user input information described in this manual.

[0029] In one implementation, the user input information may include only one piece of information input by the user most recently, or may include several pieces of information input by the user most recently, or may include all information input by the user in the past, and the present invention is not limited to this.

[0030] Step S3: The large language model analyzes the information and extracts information intent, and extracts target application services from the application service set according to the information intent.

[0031] In one embodiment, Figure 2 As shown, step S3 includes: Step S311: The large language model analyzes the information to extract intent; Step S312: matching application services in the application service set according to the intent; Step S313: determine whether the application service is matched, if yes, proceed to step S314, if not, proceed to step S315; Step S314: extracting the matched application service as the target application service; Step S315: the large language model outputs intention guidance information according to the intention; Step S316: supplement the information according to the intention guidance information and proceed to step S311.

[0032] In one embodiment, the large language model can efficiently perform intent recognition through its powerful natural language processing capabilities. For example, in a specific embodiment, after pre-training on large-scale text data, the large language model can understand the deep semantics of user input and infer the implicit intent. In addition, the large language model also has excellent context processing capabilities and can track changes in user intent based on conversation history. In multiple rounds of conversations, the large language model can gradually decompose complex requirements and ensure accurate understanding of intent by issuing intent guidance information. The intent guidance information can be a question or a clarification. In this embodiment, the content and form of the intent guidance information are not specifically limited. Through fine-tuning, the large language model can also be adapted to tasks in specific fields to further improve the accuracy and applicability of its intent recognition. Regarding the intent recognition algorithm, those skilled in the art can understand and implement it based on prior art knowledge, so the present invention will not be described in detail.

[0033] In another embodiment, the information processing method based on the large language model provides an application service set, but the current conversation may contain multiple intents, or two or more application services are required to implement an intent. The semantic understanding of the large language model can be used to decompose the user intent into two or more sub-intents, and call the corresponding multiple application services according to the sub-intents.

[0034] For example, in one example, during the current conversation, the user input information is "Please extract the data of Table 1 from the document and draw a map scatter plot" and a PDF document. Based on the user input information, the large model can recognize that the current conversation corresponds to two intentions, namely "data extraction" and "drawing a map". It is necessary to first call the document parsing service based on the "data extraction" task. After completing the data extraction from the document, the data visualization service can be called based on the "drawing a map" task to draw a map scatter plot.

[0035] In another embodiment, Figure 3 As shown, step S3 includes: Step S321: presetting a scenario task set, wherein the scenario tasks in the scenario task set correspond to one or more application services; Step S322: The large language model analyzes the information to extract intent; Step S323, matching a target scenario task in the scenario task set according to the intention; Step S324: determine whether the target scene task is matched, if yes, proceed to step S325, if no, proceed to step S326; Step S325, extracting the application service corresponding to the target scenario task as the target application service; Step S326: the large language model outputs intention guidance information according to the intention; Step S337, supplement the information according to the intention guidance information, and proceed to step S322.

[0036] Specifically, this embodiment packages one or more application services into a scenario task to facilitate the invocation of multiple application services. The application service corresponding to the target scenario task is the target application service.

[0037] For example, the user input information is "conduct an in-depth comparison and analysis of the two studies from multiple dimensions such as research purpose, theoretical framework, main findings, application and significance" and two PDF documents, indicating that the user expects to summarize the contents of the two papers and compare them. Therefore, the scenario task corresponding to the user's intention should be "document overview".

[0038] For another example, the user input information is "Please help me draw a map scatter plot based on the data below" and an EXCEL file of scattered data, indicating that the user expects to read the data and visualize the data, so the scenario task corresponding to the user's intention should be "data visualization".

[0039] For another example, the user input information is "Please convert the following image into a table" and a table image, indicating that the user expects to extract the table data in the image, so the scenario task corresponding to the user's intention should be "convert image into table".

[0040] From the above, we can see that different user intentions correspond to different scenario tasks, and the basic capabilities required to perform tasks in different scenario tasks are also different. For example, taking "Document Overview" as an example, in order to complete the overview and comparison of the paper, it is necessary to call the "Document Parsing" application service to answer the user's questions based on the processing results of the document parsing service. In "Data Visualization", in order to draw a map scatter plot, it is necessary to call the "Data Visualization" application service to draw a map scatter plot based on the data uploaded by the user. From this, it can be understood that in order to achieve different user intentions, different application services need to be used to achieve different scenario tasks corresponding to user intentions.

[0041] Therefore, in this implementation, some basic capabilities of the integrated multiple applications can be pre-packaged into application services called by the large language model, and a correspondence between different scenario tasks and application services can be established.

[0042] In particular, the integrated multiple applications are mainly used to solve professional problems in specific scenarios, have more specialized and vertical functions, and provide complex operation interfaces. Users can use the information processing method based on the large language model or the system based on this method to express their intentions in a more accustomed and convenient way through dialogue, and can complete tasks in the interactive interface of the conversation. We encapsulate the common basic functions of the integrated applications into application services that can be called in the conversation to help users complete tasks in specific scenarios in the conversation.

[0043] For example, in one embodiment, the integrated applications include document management applications, document reading applications, map-to-table applications, chart drawing applications, geological map recognition applications, data extraction applications, etc., and the corresponding application services that can be abstracted include document management services, document parsing services, map-to-table services, data visualization services, geological map recognition services, data extraction services, etc.

[0044] The document management application is used by users to upload and manage personal documents, and provides operations such as document upload and document deletion, etc. The document management service is used to select personal documents in a session and enter them into the session, or to save documents in the session into personal documents.

[0045] The document reading application is used to open the user's documents and perform detailed reading on the documents, providing operations such as selecting translation, selecting dialogue, selecting recognition table, etc. The document parsing service is used to understand the documents and questions entered by the user in the session and output the results.

[0046] The Image-to-Table application is used by users to upload images and convert them into tables, providing operations such as editing, inserting, deleting, and exporting. The Image-to-Table service is used to convert images sent in a session into tables and output the results.

[0047] The geological map recognition application is used to identify the coordinates and legends of the geological maps uploaded by users, and provides operations such as coordinate extraction and legend extraction. The geological map recognition service is used to read the coordinates and legend information of the geological map sent in the session and output the results.

[0048] The data extraction application is used for documents uploaded by users. You can define extraction tasks, including metadata, table data, or data in other specified fields. You can modify the extraction results and save the extraction results into the database.

[0049] The document extraction service application is used to extract information of specified fields from the documents sent in the session and output the results.

[0050] Of course, those skilled in the art will appreciate that the pre-built application services are not limited to the above examples, and may also include other types of application service modules, which cannot be exhaustively listed in the present disclosure and are not limited thereto.

[0051] In one implementation, after encapsulating application services with different functions based on the integrated applications, all application services can be integrated to obtain an application service set, that is, the application service set includes multiple application services corresponding to different functions, and each application service can implement some corresponding functions by calling the basic capabilities of an application software.

[0052] As can be seen from the above, when creating tasks in different scenarios, the application services that need to be called may be different. Therefore, in the embodiments of the present disclosure, corresponding application services can be set in advance for different scenario tasks, so as to obtain the correspondence between scenario tasks and application services. For example, in an example, the correspondence between scenario tasks and application services can be shown in the following Table 1:

[0053] It is worth noting that for some common and frequently used tasks and scenarios, the corresponding relationship in the example of Table 1 can be pre-built based on the application services integrated by this method. For some problems that require multiple application services or multiple intentions, the scenario task can be mapped to two or more application services, and then the application service to be called can be quickly determined based on the corresponding relationship.

[0054] In another embodiment, the information processing method based on the large language model provides an application service set, but the current conversation may contain multiple scenario tasks, which do not maintain a corresponding relationship with the application services. In this case, the semantic understanding of the large language model can be used to identify the multiple scenario tasks contained in the current conversation based on the user's intention, and the corresponding multiple application services can be combined and called.

[0055] However, due to the diversity of user needs for text input, the recognition ability of a large language model will be limited when used in different scenarios or after different training. Based on this, the present invention discloses another embodiment, in which the information processing method further includes: Preset information supplement times N1; If the number of times the information is supplemented reaches N1, the large language model processes and outputs the information.

[0056] That is, when the large language model cannot recognize the intent or scenario task, and the problem cannot be solved by calling the application service, the large language model's own conversational capabilities are used to respond to user questions.

[0057] Step S4: The large language model extracts input parameters required by the target application service from the information, and outputs the input parameters to the target application service.

[0058] Specifically, Figure 4 As shown, step S4 includes the following steps: Step S411, the large language model extracts input parameters from the information; Step S412: determine whether the input parameters meet the target application service computing requirements, if yes, proceed to step S413, if no, proceed to step S414; Step S413: output the input parameters to the target application service; Step S414, the large language model outputs parameter guidance information; Step S415, supplement the information according to the parameter guidance information, and proceed to step S411.

[0059] In the embodiments of this specification, when calling the application service, it is necessary to extract the parameters required by the application from the information input by the user based on the large language model. Through a deep understanding of natural language, the trained large language model can automatically extract key parameters related to the target application service from the user's unstructured input. For example, when a user wants to draw a map scatter plot, the user enters the longitude, latitude, eigenvalue, etc. of the sampling point in the conversation. The large language model can automatically identify this information according to the context and convert it into data in the parameter format required by the target application service.

[0060] In addition, large language models excel in identifying context and implicit information. When the user's input is not clear enough, the model can further extract the required parameters by inferring or guiding questions. For example, in a conversation, the user may only mention "Please draw the sampling point data in the table above into a map scatter plot." The model can automatically extract the sampling point data in the historical conversation and convert it into the parameter format required by the application service.

[0061] If the large language model fails to extract the parameters required by the application service from the user input information in the current conversation process, the large language model will output parameter guidance information to prompt or remind the user to further input information.

[0062] Regarding the application parameter extraction using a large language model, those skilled in the art can understand and fully implement it by combining relevant technologies, and this disclosure will not elaborate on this.

[0063] However, the user's personal knowledge reserves are limited. When they are unable to correctly enter the required parameters after multiple prompts, a large language model is needed to intervene and process the information input by the user.

[0064] Based on this, the present invention further discloses an embodiment, in which the information processing method further includes: The preset parameter supplement times N2; If the number of parameter supplementation times reaches N2, the large language model processes and outputs the information.

[0065] That is, when the user is unable to correctly input the required parameters due to personal reasons, the large language model's own conversational capabilities can be used to respond to the user's questions.

[0066] It should be noted that the preset information supplement times N1 and the preset parameter supplement times N2 can be pre-fixed and embedded in the large language model, or a control interface can be displayed to the user for the user to manually adjust the preset information supplement times N1 and the preset parameter supplement times N2.

[0067] Step S5: The target application service performs calculations based on the input parameters and outputs results.

[0068] The output results can be displayed in the result output area of ​​the user interaction interface, and the displayed information includes text information, formulas, tables, pictures, maps, links, lists, etc. In addition, the historical output results of the current conversation process can also be recorded and displayed in the user interaction interface in a waterfall flow manner.

[0069] In one embodiment, the user may also only input text information. For example, the user inputs "What is the distribution of plateaus on the earth?" The large language model can determine the intention of the user's input information. For such basic questions of professional knowledge, the application service generally does not need to be involved, or such simple application services are not preset. In this case, after matching and determination, the user's question will be answered by the ability of the large language model.

[0070] The matching judgment process in this embodiment can be to match application services according to user intentions, and when the number of information supplements reaches N1, the large language model processes and outputs the information; or to match scenario task sets according to user intentions, and set the basic questions of such professional knowledge as a scenario task set, and the scenario task set corresponds to the large language model, that is, after matching the scenario task set, the large language model directly processes and outputs the information.

[0071] The information processing method based on the large language model disclosed in the present invention combines the large language model with application services. On the one hand, it can improve the application effect of traditional application software, expand the application space, and be more intelligent, thereby lowering the threshold for use. On the other hand, it can greatly enhance the depth of use of the traditional language model and expand it to a variety of business scenarios, so that it can solve complex professional problems during the conversation process, greatly enhancing the application surface and user experience of the traditional language model.

[0072] In another embodiment, the information processing method further includes: Preset application packages; When or after outputting the result, calling the target application software corresponding to the target application service; The output result or input information is deeply processed using the target application software.

[0073] In the previous embodiment, after the dialogue task is created, the dialogue task will be executed immediately and the output result of the task execution will be output. For example, when the user enters the information "Please convert the following image into a table" and uploads a picture of a table, the scene task is "image to table" based on the intent recognition, the target application service is called and the dialogue scene "image to table" task is executed, and the data results are presented in the form of a table.

[0074] In this embodiment, in the dialogue result output area, the data results can be further analyzed or processed. That is, the information processing method also supports seamless jump to the corresponding application software for detailed operations.

[0075] In another embodiment, after creating a dialog task, processing and outputting the result, the user can perform simple processing on the result output by the current task, including editing, exporting, etc. For example, the user can modify the output result of the "graph to table" target application service in the current dialog where an error occurs in recognition, and then export the data result after modification.

[0076] When the user needs to further process or analyze the result of the conversation, the application software corresponding to the target application service is called to assist the user to jump to the corresponding application for further operation. For example, when the user is not satisfied with the output result of the "Image to Table" target application service, the user can go to the "Image to Table" application function interface through "Advanced Operation" to modify and edit the output result.

[0077] In the question-and-answer dialogue interaction interface, the functions of some application software are interactively integrated, and the function entrances of all application software are provided in the system navigation bar based on the information processing method based on the large language model disclosed by the present invention. For example, the "My Document" application is provided for document management functions. When the user wants to upload a document in the conversation input box, it supports users to upload directly from the local (From Local) or select the document to be uploaded from the "My Documents" (From MyDocuments) provided by the system. For documents uploaded from the local in the conversation, the "Save" operation is also provided in the conversation interface to save the document to "My Documents".

[0078] Similarly, the system provides a "Document Reader" application. If the user is not satisfied with the output of the document parsing service in the conversation, he or she can click the "Read" operation in the conversation interface to directly open the document with the reader for in-depth reading.

[0079] Similarly, the system provides a "Table Extraction" application. Users can directly jump to the "Table Extraction" application by clicking the "Advanced Operations" provided in the output results of the Table Extraction task in the dialog to further process the results. Users can also directly open the chart in the "Table Extraction" application to perform operations such as table conversion, editing, and export.

[0080] In the existing related technologies, the AI ​​assistant based on the large language model mainly interacts with the user in the form of dialogue and communication, and the output content is mainly in the form of text, which cannot meet some scientific research scene tasks, and the actual help to the user's scientific research work is not great, resulting in the user's willingness to use the AI ​​scientific research assistant is not high. The information processing method disclosed in the present invention can also be applied to the AI ​​assistant of the scientific research scene, realizes autonomous task creation and processing for various user demand scenes, and combines the intention recognition of the large language model, the extraction of application parameters and the call and control of application services, which reduces the use threshold of traditional tools, and the output results will not only present text, but also display pictures, tables, maps and even videos and other rich display forms, and provide further operation buttons, such as export, edit, advanced operations, etc., to jump to the application software to further process and analyze the output results. Based on this, the information processing method can well meet the user needs in the scientific research scene, not only realizes the professional functions of traditional tools, but also makes full use of the convenient interaction method of AI question and answer, and improves the use willingness of AI assistants.

[0081] Another embodiment of the present invention discloses an information processing system based on a large language model, comprising: An information input window, wherein the information input window is used for information input; An application service module, which is used to store a preset application service set, perform calculations based on input parameters and output results; A large language model module, wherein the large language model module stores a trained large language model and is used to obtain information inputted in the information input window, analyze the information and extract information intent, extract a target application service in the application service module according to the information intent, extract input parameters required for the target application service from the information, and output the input parameters to the application service module; A result display window is used to display the result output by the application service module.

[0082] In another embodiment, the large language model module is also used to process the information and output a result.

[0083] In another embodiment, the information processing system based on the large language model also includes: an application software module, which stores a preset application software package and is used to display an operation interface after the large language model module is called and perform in-depth processing on the output results or input information.

[0084] An information processing system based on a large language model disclosed in the present invention is a system based on the information processing method based on a large language model disclosed in the above embodiment, which is one of the specific implementations of the method.

[0085] Another embodiment of the present invention further discloses an AI intelligent assistant, which includes a computer-readable storage medium, in which at least one instruction is stored, and the instruction is read by a processor to enable the AI ​​intelligent assistant to execute the information processing method described in any of the above embodiments.

[0086] The AI ​​intelligent assistant disclosed in the present invention is executed based on the information processing method based on the large language model disclosed in the above embodiment, which is a specific form of implementing the information processing method.

[0087] It should be noted that the AI ​​intelligent assistant can also be understood as a computer program, and the computer program is flexible and adaptable, so that the execution steps can be adjusted according to different inputs and conditions, so as to adapt to the ever-changing needs and technical environment and maintain the advancement and practicality of the computer program. In addition, the architecture of the computer device allows the implementation of multiple functions, so that the program in the memory can contain multiple functional modules, and the computer program is executed by the processor to meet different application needs, thereby improving the function of the computer program and enhancing the integration of the computer device.

[0088] In another embodiment, the information processing method based on the large language model disclosed in the present invention is explained by taking the "graph to table" recognition scenario in a scientific research scenario as an example.

[0089] This embodiment discloses an AI intelligent assistant based on the information processing method based on a large language model disclosed in the present invention. The AI ​​intelligent assistant displays some application scenario recommendations in the form of cards at the top of the homepage for some frequently used scientific research scenarios, such as "find literature", "convert image to table", "literature analysis", etc. By clicking on the card, the user can view the use case associated with the scenario, which is convenient for quickly understanding the application capabilities encapsulated by the system in the session.

[0090] The user can select text, pictures, documents, etc. in the input box to input information, so that the large language model can obtain the user input information. It can be seen from the above embodiments that multiple forms can be provided for user input information, for example, user input information can include text information, picture information, and uploaded PDF documents, Excel files, and can even be input in the form of video.

[0091] For example, in one instance, the user enters "Please convert the following image into a table" and uploads a picture.

[0092] After obtaining the user input information, the large language model can identify the user input information, obtain the scenario task corresponding to the current dialogue based on the user intent of "converting the image to the table", and select the target application service "Table Extraction" that matches the current dialogue task based on the scenario task. Alternatively, the user intent of "converting the image to the table" can be obtained by identifying the user input information, and the target application service "Table Extraction" that matches the current intent can be selected from the preset application service set.

[0093] Extract the uploaded image from the user input information, call the target application service and create a dialogue task, execute the dialogue task to generate and output the result. If the dialogue output result meets expectations, you can click "Export" on the table to export the table data; if you are not satisfied with the dialogue output result, you can click "Edit" on the table to edit the data in the table for further processing; if the output result does not handle the merged rows and columns well, then you need to perform more complex merged rows or columns on the dialogue result, then you can click "Advance" on the table to seamlessly jump to the "Table Extraction" in the system for operation.

[0094] The above takes the scenario task of "converting a graph to a table" as an example to illustrate the AI ​​smart assistant based on the information processing method based on the large language model disclosed in the present invention. In actual use, the current conversation between the user and the AI ​​smart assistant may cover multiple tasks, which may be multiple tasks in the same scenario or multiple tasks in different scenario tasks.

[0095] In another embodiment, the "image to table" scenario is still taken as an example. When the user inputs the information "Please draw a map scatter plot based on the data in the table in the picture" and uploads a table picture. The current conversation includes two intentions, which respectively convert the picture into a table and can draw a map scatter plot based on the data in the table. Then the AI ​​smart assistant needs to successively implement the creation and execution of these two tasks based on the information processing method based on the large language model disclosed in the present invention, and respond with the results of displaying the data. The specific process will not be repeated here.

[0096] In another embodiment, take the "data visualization" scenario as an example. A PDF paper was uploaded in the user's historical conversation, and the data of Table 1 in the document was extracted. The conversation results were displayed in the first round of conversation. The current user inputs the information again as "Please draw a map scatter plot of the above data". The current conversation includes an intention to draw a map scatter plot, but the specific application parameters need to be obtained from the historical conversation. Then the AI ​​intelligent assistant needs to understand the content of the context and identify valid application parameters from it, and implement the creation and execution of tasks based on the information processing method based on the large language model disclosed in the present invention, and display the results of the data accordingly. The specific process is not repeated here. If the output result of the conversation meets expectations, you can click "Export" on the table to export the visualized data; if you are not satisfied with the output result of the conversation, you can click "Edit" on the table to switch the type of chart to be drawn, and edit the data of the chart to be drawn, regenerate the visualization task and execute it; if you need to perform more complex operations on the output results, you can click "Advance" on the table to seamlessly jump to "Draw Charts" in the system for operation.

[0097] It can be seen from the above embodiments that the information processing method based on the large language model disclosed in the present invention pre-encapsulates the basic capabilities of the application software in the application service set. In the process of dialogue recognition, by identifying the user's intention, the required application service is called to complete the corresponding task construction to meet the user's different scientific research scene requirements. By providing editing, exporting and jumping operations for the results of the conversation output, the user can further process and handle the output results, meet the user's professional requirements for the application, and realize the combination of the intelligence of the AI ​​smart assistant and the capabilities of professional application software.

[0098] It can be seen from the above embodiments that the information processing method based on the large language model disclosed in the present invention has obvious advantages over the prior art. First, it can provide a one-stop service experience, that is, the system can integrate multiple applications based on different business scenarios, provide an integrated service platform, and reduce the trouble of users switching between different applications. Second, it can automate cross-application task processing, that is, through a simple question-and-answer dialogue interaction, the system can automatically call different application services, process multiple types of tasks and feedback the results, without the need for users to manually operate complex processes. Third, it can improve the flexible depth processing capability, that is, when the user wants to further analyze or process the dialogue results, the system supports seamless jump to the corresponding system application for detailed operations, so that the user can deeply process data or tasks. Fourth, it has the improvement of scalability and intelligence, that is, on the surface, the overall application scenario of this technical solution is similar to the prior art, but the characteristics of this method determine that it will be more expanded and more intelligent in specific application effects, which can greatly improve the depth of use of intelligent assistants. The scalability of the system is reflected in the fact that it can be expanded to a variety of business scenarios and support applications of different task types. Intelligence is reflected in the integration of the conversational capabilities of a large language model and the deep functions of various professional applications, enabling the system to solve complex professional problems during the conversation process, greatly improving the application scope and user experience of the intelligent assistant.

[0099] It should be emphasized that the embodiments of the present invention are all displayed in a dialogue window on the PC side, but in addition to the PC side, the program, module or system based on the method can also be integrated on the mobile side, or the computing part can be placed on the cloud and the other parts can be placed on the PC side or the mobile side.

[0100] It should be noted that when a component is referred to as being "fixed to" or "disposed on" another component, it may be directly on the other component or there may also be a central component. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may be a central component at the same time. The terms "vertical", "horizontal", "upper", "lower", "left", "right" and similar expressions used in the specification of this application are for illustrative purposes only and do not represent the only implementation method.

[0101] In addition, the terms "first" and "second" are used only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or equipment.

[0102] In the present application, unless otherwise clearly specified and limited, a first feature being “above” or “below” a second feature may mean that the first feature is directly in contact with the second feature, or the first feature and the second feature are indirectly in contact through an intermediate medium. Moreover, a first feature being “above”, “above” or “above” a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being “below”, “below” or “below” a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0103] Unless otherwise defined, all technical and scientific terms used in the specification of this application have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in the specification of this application includes any and all combinations of one or more related listed items.

[0104] The technical features of the above embodiments can be combined without changing the basic principles of the present application. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the scope of patent protection of the present application shall be subject to the attached claims.

Claims

1. An information processing method based on a large language model, characterized in that: include: Train large language models and pre-set application service sets; The large language model obtains information input by the user; The large language model analyzes the information and extracts information intent, and extracts a target application service from the application service set according to the information intent; The large language model extracts input parameters required by the target application service from the information, and outputs the input parameters to the target application service; The target application service performs calculation processing based on the input parameters and outputs a result.

2. The information processing method according to claim 1, characterized in that: The process of the large language model analyzing the information and extracting the information intent, and extracting the target application service from the application service set according to the information intent, includes: Step S311: The large language model analyzes the information to extract intent; Step S312: matching application services in the application service set according to the intent; Step S313: determine whether the application service is matched, if yes, proceed to step S314, if not, proceed to step S315; Step S314: extracting the matched application service as the target application service; Step S315: the large language model outputs intention guidance information according to the intention; Step S316: supplement the information according to the intention guidance information and proceed to step S311.

3. The information processing method according to claim 1, characterized in that: The process of the large language model analyzing the information and extracting the information intent, and extracting the target application service from the application service set according to the information intent, includes: Step S321: presetting a scenario task set, wherein the scenario tasks in the scenario task set correspond to one or more application services; Step S322: The large language model analyzes the information to extract intent; Step S323, matching a target scenario task in the scenario task set according to the intention; Step S324: determine whether the target scene task is matched, if yes, proceed to step S325, if no, proceed to step S326; Step S325, extracting the application service corresponding to the target scenario task as the target application service; Step S326: the large language model outputs intention guidance information according to the intention; Step S337, supplement the information according to the intention guidance information, and proceed to step S322.

4. The information processing method according to claim 2 or 3, characterized in that: Also includes: Preset information supplement times N1; If the number of times the information is supplemented reaches N1, the large language model processes and outputs the information.

5. The information processing method according to claim 1, characterized in that: The process in which the large language model extracts input parameters required by the target application service from the information and outputs the input parameters to the target application service includes: Step S411, the large language model extracts input parameters from the information; Step S412: determine whether the input parameters meet the target application service computing requirements, if yes, proceed to step S413, if no, proceed to step S414; Step S413: output the input parameters to the target application service; Step S414, the large language model outputs parameter guidance information; Step S415, supplement the information according to the parameter guidance information, and proceed to step S411.

6. The information processing method according to claim 5, characterized in that: Also includes: The preset parameter supplement times N2; If the number of parameter supplementation times reaches N2, the large language model processes and outputs the information.

7. The information processing method according to claim 1, characterized in that: Also includes: Preset application packages; When or after outputting the result, calling the target application software corresponding to the target application service; The output result or input information is deeply processed using the target application software.

8. An information processing system based on a large language model, characterized in that: include: An information input window, wherein the information input window is used for information input; An application service module, which is used to store a preset application service set, perform calculations based on input parameters and output results; A large language model module, wherein the large language model module stores a trained large language model and is used to obtain information inputted in the information input window, analyze the information and extract information intent, extract a target application service in the application service module according to the information intent, extract input parameters required for the target application service from the information, and output them to the application service module; A result display window is used to display the result output by the application service module.

9. The information processing system based on a large language model according to claim 8, characterized in that: The large language model module is also used to process the information and output the result.

10. The information processing system based on a large language model according to claim 8, characterized in that: Also includes: An application software module stores a preset application software package, which is used to display an operation interface after the large language model module is called and perform in-depth processing on the output results or input information.

11. An AI intelligent assistant, characterized in that: It includes a computer-readable storage medium, in which at least one instruction is stored, and the instruction is read by a processor to enable the AI ​​intelligent assistant to execute the method as described in any one of claims 1 to claim 7.

Images