Interactive modeling system based on large model

Abstract

The invention discloses an interactive modeling system based on a large model, and the system comprises a data preprocessing module which is used for carrying out the preprocessing of modeling data based on an input first interaction instruction, and obtaining the preprocessed modeling data; the feature engineering module is used for performing feature engineering processing on the preprocessed modeling data based on an input second interaction instruction to obtain modeling data after feature engineering processing; the model construction module is used for performing model construction on the modeling data subjected to feature engineering processing based on an input third interaction instruction to obtain a model corresponding to the modeling data; the intelligent agent is used for receiving and analyzing an input modeling starting instruction, obtaining a modeling step according to input modeling data and feeding back the modeling step to a user; and respectively calling the data preprocessing module, the feature engineering module and the model building module in sequence based on the modeling step. Through the above mode, conversational interactive modeling can be realized, the modeling threshold is reduced, and the modeling efficiency and accuracy are improved through intelligent recommendation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of interactive modeling, and in particular to an interactive modeling system based on large models. BACKGROUND

[0002] With the advent of the data era, enterprises and individuals are facing an increasingly large amount of data, and data-driven decision-making has become an important means to enhance competitiveness. However, the traditional data modeling process usually requires professional statistical and programming knowledge, making it difficult for non-professionals to participate. In addition, existing data analysis tools often have complex interfaces and high usage thresholds, limiting their widespread application.

[0003] In recent years, the rapid development of deep learning and natural language processing technologies has brought new opportunities to the field of data modeling. Large-scale pre-trained models such as the GPT series and BERT have shown excellent performance in understanding and generating natural language, providing the possibility of achieving more intuitive and user-friendly interactions. At the same time, the progress of real-time data feedback and intelligent recommendation technologies has also opened up new paths to improve modeling efficiency and accuracy.

[0004] Although there have been some low-code or no-code platforms, they still fail to fully integrate the latest artificial intelligence technologies, lacking flexible and intelligent user interaction methods. Therefore, it is necessary to develop an interactive modeling platform that integrates advanced large models, supports conversational modeling, and provides real-time feedback to lower the modeling threshold and enable more users to conveniently perform data analysis and decision-making. This platform will not only meet the basic modeling needs of users, but also continuously optimize its functions through intelligent recommendations and user analysis to further enhance user experience. SUMMARY

[0005] The present application mainly provides an interactive modeling system based on large models to solve the problems of high user usage threshold and lack of real-time feedback in existing interactive modeling platforms.

[0006] To solve the above technical problems, one technical solution adopted by the present application is to provide an interactive modeling system based on large models, which includes: a data preprocessing module for preprocessing modeling data based on input first interaction instructions to obtain preprocessed modeling data; a feature engineering module for performing feature engineering processing on the preprocessed modeling data based on input second interaction instructions to obtain feature engineering processed modeling data; a model construction module for constructing a model based on input third interaction instructions on the feature engineering processed modeling data to obtain a model corresponding to the modeling data; The intelligent agent is configured to receive and analyze input modeling start instruction, and feed back modeling step to a user according to input modeling data; and sequentially call the data preprocessing module, the feature engineering module and the model construction module based on the modeling step.

[0007] In an optional embodiment of the application, the data preprocessing module comprises a data preprocessing user agent, a data preprocessing assistant and a data preprocessing manager which are communicatively connected with each other, and a data preprocessing intelligent agent which is communicatively connected with the intelligent agent and the data preprocessing manager. The data preprocessing user agent is configured to receive the first interaction instruction and forward it to the data preprocessing assistant after entering the data preprocessing module. The data preprocessing assistant is configured to analyze the first interaction instruction to obtain a data preprocessing plan, preprocess the modeling data based on the data preprocessing plan, and obtain the preprocessed modeling data. The data preprocessing manager is configured to collect all execution data in the data preprocessing module, and feed back to the user and the data preprocessing intelligent agent. The data preprocessing intelligent agent is configured to send all execution data in the data preprocessing module and a data preprocessing completion identifier to the intelligent agent.

[0008] In an optional embodiment of the application, the intelligent agent is further configured to send data preprocessing instruction and the modeling data to the data preprocessing intelligent agent based on the modeling step. The data preprocessing intelligent agent is further configured to receive the data preprocessing instruction and the modeling data sent by the intelligent agent and forward them to the data preprocessing assistant.

[0009] In an optional embodiment of the application, the feature engineering module comprises a feature engineering user agent, a feature engineering assistant and a feature engineering manager which are communicatively connected with each other, and a feature engineering intelligent agent which is communicatively connected with the intelligent agent and the feature engineering manager. The feature engineering user agent is configured to receive the second interaction instruction and forward it to the feature engineering assistant after entering the feature engineering module. The feature engineering assistant is configured to analyze the second interaction instruction to obtain a feature engineering plan, perform feature engineering processing on the preprocessed modeling data based on the feature engineering plan, and obtain the modeling data after feature engineering processing. The feature engineering manager is configured to collect all execution data in the feature engineering module, and feed back to the user and the feature engineering intelligent agent. The feature engineering agent is configured to send all execution data in the feature engineering module and a feature engineering processing completion identifier to the agent.

[0010] In an optional implementation of the embodiments of the application, the agent is further configured to send a feature engineering processing instruction and the preprocessed modeling data to the feature engineering agent based on the modeling step, all execution data in the data preprocessing module, and the data preprocessing completion identifier. The feature engineering agent is further configured to receive the feature engineering processing instruction and the preprocessed modeling data sent by the agent and forward them to the feature engineering assistant.

[0011] In an optional implementation of the embodiments of the application, the model construction module comprises a model construction user agent, a model construction assistant, and a model construction manager that are communicatively connected to each other, and a model construction agent that is communicatively connected to the agent and the model construction manager. The model construction user agent is configured to receive the third interaction instruction and forward it to the model construction assistant after entering the model construction module. The model construction assistant is configured to analyze the third interaction instruction to obtain a model construction plan, perform model construction on the feature engineering processed modeling data based on the model construction plan, and obtain a model corresponding to the modeling data. The model construction manager is configured to collect all execution data in the model construction module and feed back to a user and the model construction agent. The model construction agent is configured to receive the model construction instruction and the feature engineering processed modeling data sent by the agent and forward them to the model construction assistant, and send all execution data in the model construction module and a model construction completion identifier to the agent.

[0012] In an optional implementation of the embodiments of the application, the agent is further configured to send a model construction instruction and the feature engineering processed modeling data to the model construction agent based on the modeling step, all execution data in the feature engineering module, and the feature engineering processing completion identifier. The model construction agent is further configured to receive the model construction instruction and the feature engineering processed modeling data sent by the agent and forward them to the model construction assistant.

[0013] In an optional implementation of the embodiment of the application, the intelligent agent is further configured to collect and analyze historical input interaction instructions and historical execution data of the data preprocessing module, the feature engineering module and the model construction module to obtain user behavior preferences; and dynamically optimize the intelligent agent, the data preprocessing module, the feature engineering module and the model construction module according to the user behavior preferences.

[0014] In an optional implementation of the embodiment of the application, the intelligent agent is further configured to receive and analyze input fourth interaction instructions, and update the modeling step based on the fourth interaction instructions.

[0015] In an optional implementation of the embodiment of the application, the first interaction instruction, the second interaction instruction, the third interaction instruction and the fourth interaction instruction are one of an empty input instruction, a natural language description instruction and a termination instruction.

[0016] The beneficial effects of the application are as follows: Different from the prior art, the application discloses an interactive modeling system based on a large model. The system includes a data preprocessing module, a feature engineering module, a model construction module and an intelligent agent. The intelligent agent interacts with a user, receives a modeling start instruction and feeds back a modeling step to the user, and sequentially calls the data preprocessing module, the feature engineering module and the model construction module based on the modeling step to perform corresponding tasks. The data preprocessing module, the feature engineering module and the model construction module also interact with the user in real time, receive and analyze input interaction instructions to perform corresponding processing, thereby realizing modeling. The system models through dialog interaction. The user can describe modeling requirements through natural language, without the need for professional programming or statistical knowledge, so that non-professionals can easily model, reducing the modeling threshold. The intelligent agent can provide execution results and modeling suggestions of each module in real time during interaction with the user, helping the user to timely adjust the model and improving modeling efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor. Figure 1 is a structural schematic diagram of the interactive modeling system based on a large model provided by the application; Figure 2 is a structural schematic diagram of the data preprocessing module of the interactive modeling system based on a large model provided by the application; Figure 3is a feature engineering module structure schematic diagram of the large model-based interactive modeling system provided in the application; Figure 4 is a model construction module structure schematic diagram of the large model-based interactive modeling system provided in the application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0019] The terms “first”, “second”, “third” in the embodiments of the present application are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by “first”, “second”, “third” can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of “a plurality of” is at least two, for example, two, three, etc., unless otherwise explicitly and specifically limited. In addition, the terms “include” and “have” and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0020] In this document, referring to “embodiments” means that the specific features, structures or properties described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] The present application provides a large model-based interactive modeling system, referring to Figure 1 , Figure 1 is a structure schematic diagram of the large model-based interactive modeling system provided in the present application, which comprises: The data preprocessing module 10 is configured to preprocess the modeling data based on the input first interaction instruction to obtain preprocessed modeling data. The feature engineering module 20 is configured to perform feature engineering processing on the preprocessed modeling data based on the input second interaction instruction to obtain feature engineering processed modeling data. a model construction module 30, configured to perform model construction on the modeling data after feature engineering based on the input third interaction instruction, to obtain a model corresponding to the modeling data; an agent 40, configured to receive and analyze the input modeling start instruction, and feed back the modeling steps to the user according to the input modeling data; the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 are called in sequence based on the modeling steps.

[0022] In this application, the agent 40 is an agent based on a large language model (LLM), which can analyze the input natural language instruction using natural language processing technology (NLP) in the form of dialogue, obtain the task intention of the user and give specific solutions. Among them, the agent (Agent) is a core concept widely existing in the fields of artificial intelligence, computer science and philosophy, and usually refers to an entity that can perceive the environment, make autonomous decisions and take actions to achieve goals. Agents have been widely used in many fields, such as controlling virtual characters in games to make intelligent decisions and actions; in industrial production, they can be used for automated process control and fault diagnosis; in the medical field, they can assist doctors in disease diagnosis and treatment plan recommendation, etc.

[0023] In this application, when a user builds a model for a group of modeling data through an interactive modeling system based on a large model, for example, models a group of marketing data, the user usually needs to sequentially perform data preprocessing, feature engineering processing and model construction processing on the group of marketing data, respectively, to obtain a model corresponding to the group of marketing data, so as to predict and score the group of marketing data and conduct targeted marketing behavior.

[0024] Before starting modeling, the user needs to upload the modeling data to the interactive modeling system. The way to upload the modeling data can be to upload it by interacting with the interactive modeling system, that is, in the form of dialogue. The uploaded modeling data can be in various forms, such as directly uploading numbers, tables and other modeling data that can be directly read, or an address that can be accessed by the interactive modeling system to read the modeling data.

[0025] In this application, the interactive modeling system can be encapsulated into an interactive modeling platform, which processes interface requests and business logic through chainlit, and is used to build an intuitive user interface, so that the user can access the interactive modeling system through a graphical interface on the interactive modeling platform. At this time, the user can input natural language instructions in the form of dialogue, such as inputting “Please help me model, my data is ‘**.csv,label=”*”’”.

[0026] Unlike the operation interface of many modeling platforms in the prior art, which is not intuitive enough and has a complicated use process, it is difficult for users to quickly start operation when performing data modeling, affecting the use experience and the modeling efficiency. The interactive modeling system based on a large model provided in the present application provides a simple and easy-to-use graphical interface, enabling users to quickly get started, reducing learning costs and improving modeling efficiency.

[0027] In the present application, when entering the interactive modeling system for a dialogue, the instructions input by the user are received by the agent 40 before entering any execution module (data preprocessing module 10, feature engineering module 20 or model construction module 30). After the user inputs an instruction containing a modeling start intent, the agent 40 receives the modeling start instruction and analyzes the modeling start instruction through natural language processing technology, obtains the modeling intent and modeling data of the user, and feeds back the corresponding modeling steps to the user, such as "the first step is data preprocessing, including data cleaning, outlier processing and missing value filling; the second step is feature engineering processing, including feature selection and feature derivation …". At this time, the user can selectively make suggestions to the modeling steps given by the agent 40, such as inputting the instruction "skip the outlier processing in the first step of data preprocessing". After receiving the instruction, the agent 40 will give the modeling steps again, "the first step is data preprocessing, including data cleaning and missing value filling; the second step …". If the user does not input any instruction after the agent 40 gives the modeling steps, the agent 40 will call the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 in sequence based on the order of the modeling steps to process the modeling data.

[0028] Unlike the data modeling tools in the prior art, which often require users to have a deep foundation in statistics and programming, non-professionals are difficult to participate in the data modeling process, thereby limiting the popularization of data-driven decision-making. The interactive modeling system based on a large model provided in the present application enables users to describe their requirements in natural language through conversational interactive modeling, without the need for professional programming or statistical knowledge, so that non-professionals can also easily perform data modeling, reducing the operation threshold, improving the accessibility and efficiency of modeling, and helping enterprises and individuals to better utilize data for decision-making, promoting the popularization and development of data-driven culture.

[0029] In the present application, with reference to Figure 2 , Figure 2 is a data preprocessing module structure diagram of the interactive modeling system based on a large model provided in the present application, the data preprocessing module 10 includes a data preprocessing user agent 11, a data preprocessing assistant 12 and a data preprocessing manager 13 which are communicatively connected, and a data preprocessing agent 14 which is communicatively connected with the agent 40 and the data preprocessing manager 13: The data preprocessing user agent 11 is configured to receive the first interaction instruction and forward the first interaction instruction to the data preprocessing assistant 12 after entering the data preprocessing module 10. The data preprocessing assistant 12 is configured to analyze the first interaction instruction to obtain a data preprocessing plan, and preprocess the modeling data based on the data preprocessing plan to obtain preprocessed modeling data. The data preprocessing manager 13 is configured to collect all execution data in the data preprocessing module 10 and feed back to the user and the data preprocessing agent 14. The data preprocessing agent 14 is configured to send all execution data in the data preprocessing module 10 and a data preprocessing completion identifier to the agent 40.

[0030] In the present application, the agent 40 is further configured to send a data preprocessing instruction and modeling data to the data preprocessing agent 14 based on the modeling step. The data preprocessing agent 14 is further configured to receive the data preprocessing instruction and the modeling data sent by the agent 40 and forward the data preprocessing instruction and the modeling data to the data preprocessing assistant 12.

[0031] In the present application, the agent 40 initiates a corresponding instruction to an execution module required to be called by the first modeling step based on the modeling step. If the first modeling step is data preprocessing, the agent 40 sends a data preprocessing instruction and modeling data to the data preprocessing agent 14 in the data preprocessing module 10.

[0032] In the data preprocessing module 10, after receiving the data preprocessing instruction and the modeling data sent by the agent 40, the data preprocessing agent 14 first forwards the data preprocessing instruction and the modeling data to the data preprocessing manager 13, and then forwards the data preprocessing instruction and the modeling data to the data preprocessing assistant 12 by the data preprocessing manager 13. After receiving the data preprocessing instruction and the modeling data from the agent 40, the data preprocessing assistant 12 analyzes the data preprocessing instruction by natural language processing technology and starts to plan a data preprocessing plan.

[0033] The data preprocessing user agent 11 (data_preprocessing_user_proxy) is responsible for receiving the interaction instruction input by the user and interacting with the user after entering the data preprocessing module 10. At this time, the user can still input the first interaction instruction to the data preprocessing user agent 11 at any time, and then the data preprocessing user agent 11 forwards the first interaction instruction to the data preprocessing assistant 12.

[0034] The data preprocessing assistant 12 is responsible for analyzing the first interaction instruction received to identify the key tasks contained therein, such as "data cleaning", etc., through natural language processing technology, and generating a corresponding data preprocessing plan in combination with the data preprocessing instructions and modeling data sent by the agent 40. In this link, the user can choose to input the first interaction instruction or not, i.e., the first interaction instruction is an empty input instruction, and the user can freely choose according to actual needs. The data preprocessing assistant 12 needs to analyze the data preprocessing instructions and the first interaction instruction when planning the data preprocessing plan.

[0035] After generating the data preprocessing plan, the data preprocessing assistant 12 preprocesses the modeling data based on the data preprocessing plan, and executes the Python code through the code interpreter to achieve preprocessing, such as using the Pandas library for data cleaning and filling, using the imbalanced-learn library to process unbalanced data, etc., to finally obtain the preprocessed modeling data. Among them, data preprocessing can include processing missing values, filling abnormal values, and deduplication, etc., compatible with multiple data types, supporting different data types of data processing methods; the code interpreter is a program or tool that can read, parse and execute code line by line, it does not compile the entire code file at once, but interprets and runs the code in real time.

[0036] The data preprocessing manager 13, as the manager of the data preprocessing module 10, is used to collect and summarize all execution data inside the data preprocessing module 10, including user input, execution status, execution flow and execution result, and can view the interaction record of each round in the data preprocessing module 10. The data preprocessing manager 13 saves all execution data in the data preprocessing module 10 collected in the background, and at the same time feeds back to the user through the display on the current dialogue page, and forwards to the data preprocessing agent 14. After the user completes each feedback, the interactive modeling system will ask whether further help is needed to ensure that the user's needs are met and the subsequent interaction is optimized.

[0037] The data preprocessing agent 14 is responsible for further analyzing and summarizing all execution data in the data preprocessing module 10 received from the data preprocessing manager 13, and delivering all execution data in the data preprocessing module 10 with a data preprocessing completion identifier (such as "data preprocessing module has been completed") to the agent 40 for interaction, so that the agent 40 can master more data information to call appropriate execution modules to continue processing the preprocessed modeling data.

[0038] Unlike the modeling tools in the prior art, which usually cannot provide instant feedback during model construction, users cannot quickly obtain results and suggestions during modeling and analysis, reducing modeling efficiency and effectiveness. The interactive modeling system based on large models provided in the present application schedules the data preprocessing module 10 to perform data preprocessing operations through the agent 40, and supports users to input first interaction instructions at any time for dynamic adjustment during the entire operation process. Users can intervene in model construction in the form of dialogue, and provide modeling results and suggestions in real time during interaction with users, support rapid iteration, help users adjust the model in time, and improve modeling efficiency and accuracy.

[0039] In the present application, the agent 40 is also used to receive and analyze the input fourth interaction instruction, and update the modeling step based on the fourth interaction instruction.

[0040] In the present application, after the agent 40 receives the feedback of the data preprocessing agent 14, it can dynamically adjust the next modeling step according to all execution data in the data preprocessing module 10. At this time, the user can continue to input the fourth interaction instruction to directly interact with the agent 40, and the agent 40 updates the modeling step based on the fourth interaction instruction. For example, the agent 40 gives the next modeling step as “data preprocessing has been completed, and the next step is feature engineering processing, including feature construction, feature transformation, feature selection, and feature encoding…” according to all execution data in the data preprocessing module 10. The user inputs the fourth interaction instruction “skip feature selection” according to actual needs, and the agent 40 updates the modeling step to “the next step is feature engineering processing, including feature construction, feature transformation, and feature encoding…”.

[0041] Unlike the prior art, the interactive modeling system based on large models provided in the present application provides real-time interaction and feedback with the user during modeling, timely understands the user's intention and dynamically adjusts the modeling step, to provide the most suitable modeling step for the user's needs, and improves the accuracy of modeling.

[0042] In the present application, the first interaction instruction, the second interaction instruction, the third interaction instruction, and the fourth interaction instruction are one of an empty input instruction, a natural language description instruction, and a termination instruction.

[0043] Among them, the empty input instruction means that the user input is empty, at this time the interactive modeling system will automatically check the previous dialogue record to identify whether there is a task to be executed, if there is, continue to process the task to be executed, if there is not, continue to dialogue with the user; the natural language description instruction is analyzed and understood by the corresponding module through natural language processing technology to extract the key task according to the user's demand; the termination instruction (“TERMINATE” instruction) means to terminate the operation of the current module, and give feedback to the user.

[0044] In the present application, with reference to Figure 3 , Figure 3 is a feature engineering module structure schematic diagram of the large model-based interactive modeling system provided in the present application, the feature engineering module 20 includes a feature engineering user agent 21, a feature engineering assistant 22 and a feature engineering manager 23 which are in communication connection with each other, and a feature engineering agent 24 which is in communication connection with the intelligent agent 40 and the feature engineering manager 23: The feature engineering user agent 21 is configured to receive the second interaction instruction and forward it to the feature engineering assistant 22 after entering the feature engineering module 20; The feature engineering assistant 22 is configured to analyze the second interaction instruction to obtain a feature engineering plan, and perform feature engineering processing on the preprocessed modeling data based on the feature engineering plan to obtain feature engineering processed modeling data; The feature engineering manager 23 is configured to collect all execution data in the feature engineering module 20 and feed back to the user and the feature engineering agent 24; The feature engineering agent 24 is configured to send all execution data in the feature engineering module 20 and a feature engineering processing completion identifier to the intelligent agent 40.

[0045] In the present application, the intelligent agent 40 is further configured to send a feature engineering processing instruction and preprocessed modeling data to the feature engineering agent 24 based on the modeling step, all execution data in the data preprocessing module 10 and the data preprocessing completion identifier; The feature engineering agent 24 is further configured to receive the feature engineering processing instruction and preprocessed modeling data sent by the intelligent agent 40 and forward them to the feature engineering assistant 22.

[0046] In the present application, the intelligent agent 40 initiates corresponding instructions to the execution module required to be called to execute the second step modeling step based on the modeling step, and if the second step modeling step is feature engineering processing, the intelligent agent 40 sends a feature engineering processing instruction and preprocessed modeling data to the feature engineering agent 24 in the feature engineering module 20.

[0047] In the feature engineering module 20, after receiving the feature engineering processing instruction and preprocessed modeling data sent by the intelligent agent 40, the feature engineering agent 24 first forwards them to the feature engineering manager 23, and then forwards them to the feature engineering assistant 22 by the feature engineering manager 23. After receiving the feature engineering processing instruction and preprocessed modeling data from the intelligent agent 40, the feature engineering assistant 22 analyzes the feature engineering processing instruction through natural language processing technology and starts to plan a feature engineering processing plan.

[0048] The feature engineering user proxy 21 is responsible for receiving the user inputted interaction instructions and interacting with the user after entering the feature engineering module 20. At this time, the user can still input a second interaction instruction to the feature engineering user proxy 21, which will forward the second interaction instruction to the feature engineering assistant 22.

[0049] The feature engineering assistant 22 is responsible for analyzing the second interaction instruction to identify the key tasks contained therein, such as "feature selection", etc., through natural language processing technology after receiving the second interaction instruction, and generating a corresponding feature engineering processing plan in combination with the feature engineering processing instructions sent by the agent 40 and the preprocessed modeling data, such as calling relevant algorithms (such as Pearson correlation coefficient) for feature selection. In this link, the user can choose to input the second interaction instruction or not, i.e. the second interaction instruction is an empty input instruction, which is freely selected according to the actual needs of the user. The feature engineering assistant 22 needs to analyze the feature engineering processing instructions and the second interaction instruction when planning the feature engineering processing plan.

[0050] After the feature engineering assistant 22 generates the feature engineering processing plan, it performs feature engineering processing on the preprocessed modeling data based on the feature engineering processing plan, and implements feature engineering processing by calling a code interpreter to execute Python code, such as scikit-learn library and feature-engine library, etc. Finally, the modeling data after feature engineering processing is obtained. Among them, the feature engineering processing can include feature construction (Feature Creation), feature transformation (Feature Transformation), feature selection (Feature Selection), feature encoding (Feature Encoding), dimensionality reduction (Dimensionality Reduction) and processing time series features, etc.

[0051] The feature engineering manager 23, as the manager of the feature engineering module 20, is used to collect and summarize all execution data inside the feature engineering module 20, including user input, execution status, execution flow and execution result, and can view the interaction record of each round in the feature engineering module 20. The feature engineering manager 23 saves all the execution data collected in the feature engineering module 20 in the background, at the same time feedbacks to the user through the way of displaying on the current dialogue page, and forwards to the feature engineering agent 24.

[0052] The feature engineering agent 24 is responsible for further analyzing and summarizing all execution data in the feature engineering module 20 received from the feature engineering manager 23, and delivering all execution data in the feature engineering module 20 with a feature engineering processing completion identifier (such as "feature engineering module has been completed") to the agent 40 for interaction, so that the agent 40 can master more data information to call appropriate execution modules to continue processing the modeling data after feature engineering processing.

[0053] In the present application, after receiving the feedback of the feature engineering agent 24, the agent 40 can dynamically adjust the subsequent modeling steps according to all execution data in the feature engineering module 20. At this time, the user can continue to input the fourth interaction instruction to directly interact with the agent 40, and the agent 40 updates the modeling steps based on the fourth interaction instruction.

[0054] Unlike the prior art, the interactive modeling system based on large models provided in the present application schedules the feature engineering module 20 to perform feature engineering processing operations through the agent 40, and supports the user to input the second interaction instruction at any time for dynamic adjustment during the entire operation process. The feature engineering assistant 22 will intelligently recommend feature engineering processing steps in combination with the second interaction instruction. The user can intervene in model construction in the form of dialogue, and provide modeling results and suggestions in real time during the interaction with the user, support rapid iteration, help the user to adjust the model in time, and improve the modeling efficiency and accuracy.

[0055] In the present application, with reference to Figure 4 , Figure 4 is a model construction module structure schematic diagram of the interactive modeling system based on large models provided in the present application. The model construction module 30 includes a model construction user agent 31, a model construction assistant 32 and a model construction manager 33 which are in communication connection with each other, and a model construction agent 34 which is in communication connection with the agent 40 and the model construction manager 33: The model construction user agent 31 is used to receive the third interaction instruction and forward it to the model construction assistant 32 after entering the model construction module 30; The model construction assistant 32 is used to analyze the third interaction instruction to obtain a model construction plan, and perform model construction on the modeling data after feature engineering processing based on the model construction plan to obtain a model corresponding to the modeling data; The model construction manager 33 is used to collect all execution data in the model construction module 30 and feed back to the user and the model construction agent 34; The model building agent 34 is configured to receive the model building instruction and the modeling data processed by the feature engineering from the agent 40 and forward them to the model building assistant 32, and send all the execution data in the model building module 30 and the model building completion identifier to the agent 40.

[0056] In the present application, the agent 40 is further configured to send the model building instruction and the modeling data processed by the feature engineering to the model building agent 34 based on the modeling step, all the execution data in the feature engineering module 20 and the feature engineering processing completion identifier. The model building agent 34 is further configured to receive the model building instruction and the modeling data processed by the feature engineering from the agent 40 and forward them to the model building assistant 32.

[0057] In the present application, the agent 40 initiates corresponding instructions to the execution module required to be called by the third modeling step, and if the third modeling step is model building, the agent 40 sends the model building instruction and the modeling data processed by the feature engineering to the model building agent 34 in the model building module 30.

[0058] In the model building module 30, after receiving the model building instruction and the modeling data processed by the feature engineering from the agent 40, the model building agent 34 first forwards them to the model building manager 33, and then forwards them to the model building assistant 32 by the model building manager 33. After receiving the model building instruction and the modeling data processed by the feature engineering from the agent 40, the model building assistant 32 analyzes the model building instruction by natural language processing technology and starts to plan the model building plan.

[0059] The model building user proxy 31 is responsible for receiving the interactive instruction input by the user and interacting with the user after entering the model building module 30. At this time, the user can still input the third interactive instruction to the model building user proxy 31, and then the model building user proxy 31 forwards the third interactive instruction to the model building assistant 32.

[0060] The model building assistant 32 is responsible for analyzing the third interaction instruction to identify the key tasks contained therein, such as "XGBoost model building", etc., through natural language processing technology after receiving the third interaction instruction, and generating a corresponding model building plan in combination with the model building instructions sent by the agent 40 and the modeling data processed by feature engineering, such as using which algorithm for model building, and for example, whether optimization is needed according to the model index. In this link, the user can choose to input the third interaction instruction, or can not input, i.e. the third interaction instruction is empty input instruction, and the user can freely choose according to the actual needs. The model building assistant 32 needs to analyze the model building instructions and the third interaction instruction when planning the model building plan.

[0061] After generating the model building plan, the model building assistant 32 performs module construction on the preprocessed modeling data based on the model building plan, and executes Python code to realize feature engineering processing through the calling of code interpreter, such as scikit-learn library, XGBoost library and LightGBM library, etc., and finally obtains the model corresponding to the modeling data. Among them, multiple machine learning algorithms are integrated in the model building, mainly according to the characteristics of the data to build a suitable model, for example, for a set of house price data modeling for house price prediction, the characteristics of the data are numerical features and continuous target, so linear regression model and XGBoost regression model are recommended; for example, for a set of sales data modeling for sales trend prediction, the characteristics of the data are time series data, so ARIMA model and LSTM model are recommended.

[0062] The model building manager 33, as the manager of the model building module 30, is used to collect and summarize all execution data in the model building module 30, including user input, execution state, execution flow and execution result, and can view the interaction record of each round in the model building module 30. All execution data in the model building module 30 collected by the model building manager 33 is saved in the background, and is fed back to the user through the display on the current dialogue page, and is forwarded to the model building agent 34.

[0063] The model building agent 34 is responsible for further analyzing and summarizing all execution data in the model building module 30 received from the model building manager 33, and delivering all execution data in the model building module 30 with a model building completion identifier (such as "model building module has been completed") to the agent 40 for interaction, so that the agent 40 can master more data information to judge whether to continue processing the built model.

[0064] In the present application, after receiving the feedback of the model construction agent 34, the agent 40 can determine whether to continue processing the model based on all the execution data in the model construction module 30, for example, it can enter the model construction module 30 again to optimize the constructed model, etc. At this time, the user can continue to input the fourth interaction instruction to directly interact with the agent 40, and the agent 40 selectively enters the corresponding execution module again based on the fourth interaction instruction.

[0065] Unlike the prior art, the interactive modeling system based on large models provided in the present application schedules the model construction module 30 to perform model construction operations through the agent 40, and supports the user to input the third interaction instruction at any time for dynamic adjustment during the entire operation process. The model construction assistant 32 will intelligently recommend model construction algorithms and model optimization steps in combination with the third interaction instruction. The user can intervene in the model construction in the form of dialogue, and provide modeling results and suggestions in real time during the interaction with the user, support rapid iteration, help the user to adjust the model in time, and improve the modeling efficiency and accuracy.

[0066] In the entire interactive modeling system, the agents and managers in each execution module interact through a clear information transmission and coordination mechanism. The user submits the corresponding interaction instruction through the user agent of each execution module. These interaction instructions are analyzed and planned into executable small tasks by the assistant and processed. In addition, the user can also submit interaction instructions through the agent 40. These instructions are summarized by the agent 40 and finally disassembled into executable small tasks, which are assigned to the agents of the corresponding execution modules for processing. This structure ensures the efficiency and flexibility of the interactive modeling system, which can quickly respond to user needs and automatically perform complex data preprocessing and feature engineering tasks.

[0067] In the present application, the agent 40 is also used to collect and analyze the historical input interaction instructions and all the historical execution data in the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 to obtain the user behavior preference; and dynamically optimize the agent 40, the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 according to the user behavior preference.

[0068] In this application, all interactive instructions and operations of user historical input and execution data in each module will be collected and recorded by the agent 40, and these historical information will be analyzed and summarized by GPT-4o to obtain user behavior preferences, so as to make intelligent recommendations to users and optimize the agent 40, the data preprocessing module 10, the feature engineering module 20 and the model building module 30. For example, a user's behavior preference is to enter the data preprocessing module 10 for data cleaning multiple times, so when this user needs to model a set of data, the agent 40 will add the step of multiple data cleaning feedback to the user when generating the modeling step, reducing the user's subsequent operation and improving the smoothness of the modeling process.

[0069] Unlike the modeling tools in the prior art, which have weak response ability to user needs, lack of intelligent recommendation mechanism based on user input and historical behavior, leading to difficulties for users in selecting models and parameters, and lack of in-depth analysis of user interaction behavior, failing to effectively identify user needs and usage patterns, making it difficult to continuously optimize system functions and improve user satisfaction. The interactive modeling system based on large models provided in this application obtains user behavior preferences through historical input interactive instructions and all historical execution data in the data preprocessing module 10, the feature engineering module 20 and the model building module 30, and the platform can intelligently recommend suitable models and parameter configurations combined with user behavior preferences to meet individual needs, enhance user experience and improve modeling accuracy.

[0070] It should be noted that the above modeling steps are not fixed as "first data preprocessing, second feature engineering processing, and third model building", and in this application, this common model building step is used for ease of description. In the actual modeling process, the specific modeling steps are intelligently recommended by the agent 40 according to the modeling data and user needs combined with user behavior preferences, which are not limited here.

[0071] Different from the prior art, the application discloses an interactive modeling system based on a large model. The system comprises a data preprocessing module 10, a feature engineering module 20, a model construction module 30 and an intelligent agent 40, which interacts with the user through the intelligent agent 40, receives the modeling starting instruction and feeds back the modeling step to the user, and sequentially calls the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 based on the modeling step to perform corresponding tasks, and the data preprocessing module 10, the feature engineering module 20 and the model construction module 30 also interact with the user in real time, receive and analyze the input interaction instruction for corresponding processing to realize modeling. The system models through dialog interaction, the user can describe the modeling requirement through natural language, without the need of professional programming or statistical knowledge, so that non-professionals can also easily model, reducing the modeling threshold; the intelligent agent can provide the execution result of each module and modeling suggestion in the process of interaction with the user, helping the user to timely adjust the model, improving the modeling efficiency and accuracy.

[0072] In an embodiment, the interactive modeling system based on a large model provided by the application can also be packaged as an interactive modeling product based on a large model. The structure of the interactive modeling product can comprise a user input module, an input analysis unit, an execution module and a feedback and interaction module, wherein: The user input module is used to receive the interaction instruction input by the user, comprising a text box recognition interface; The input analysis unit comprises a natural language processing engine and a state detector, wherein the natural language processing engine is used to analyze and understand the natural language description instruction input by the user; the state detector is used to monitor the state of the last dialogue to determine whether there is a task to be executed; The execution module comprises a task scheduler, a data processing engine, a feature engineering engine and a model construction engine, wherein the task scheduler is used to manage the execution order of the to-be-processed tasks; the data processing engine is used to execute specific data preprocessing operations, such as data cleaning, outlier processing and missing value filling, etc.; the feature engineering engine is used to execute specific feature engineering operations, such as feature screening and feature derivation, etc.; the model construction engine is used to execute specific model construction operations, such as LightGBM, xgboost model construction and model optimization, etc.; The feedback and interaction module is used to feed back the execution result to the user and perform subsequent interaction.

[0073] The above is only an embodiment of the application, and does not limit the patent scope of the application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the application.

Claims

1. An interactive modeling system based on a large model, characterized in that: include: A data preprocessing module, configured to preprocess the modeling data based on the input first interaction instruction to obtain preprocessed modeling data; A feature engineering module is used to perform feature engineering processing on the preprocessed modeling data based on the input second interactive instruction to obtain modeling data after feature engineering processing; A model building module, configured to build a model for the modeling data after the feature engineering process based on the input third interactive instruction, to obtain a model corresponding to the modeling data; An intelligent agent, configured to receive and analyze an input modeling start instruction, and obtain modeling steps based on the input modeling data and provide feedback to the user; The data preprocessing module, the feature engineering module and the model building module are called in sequence based on the modeling steps.

2. The interactive modeling system according to claim 1, characterized in that The data preprocessing module includes a data preprocessing user agent, a data preprocessing assistant, and a data preprocessing manager that are communicatively connected to each other, and a data preprocessing agent that is communicatively connected to the agent and the data preprocessing manager: The data preprocessing user agent is configured to receive the first interaction instruction and forward it to the data preprocessing assistant after entering the data preprocessing module; The data preprocessing assistant is configured to analyze the first interactive instruction to obtain a data preprocessing plan, and preprocess the modeling data based on the data preprocessing plan to obtain the preprocessed modeling data; The data preprocessing manager is used to collect all execution data in the data preprocessing module and feed it back to the user and the data preprocessing agent; The data preprocessing agent is used to send all execution data in the data preprocessing module and a data preprocessing completion mark to the agent.

3. The interactive modeling system according to claim 2, characterized in that The agent is further configured to send a data preprocessing instruction and the modeling data to the data preprocessing agent based on the modeling step; The data preprocessing agent is further configured to receive the data preprocessing instructions and the modeling data sent by the agent and forward them to the data preprocessing assistant.

4. The interactive modeling system according to claim 3, characterized in that The feature engineering module includes a feature engineering user agent, a feature engineering assistant, and a feature engineering manager that are communicatively connected to each other, and a feature engineering agent that is communicatively connected to the agent and the feature engineering manager: The feature engineering user agent is configured to receive the second interaction instruction and forward it to the feature engineering assistant after entering the feature engineering module; The feature engineering assistant is configured to analyze the second interactive instruction to obtain a feature engineering plan, and perform feature engineering processing on the preprocessed modeling data based on the feature engineering plan to obtain the modeling data after the feature engineering processing; The feature engineering manager is used to collect all execution data in the feature engineering module and feed it back to the user and the feature engineering agent; The feature engineering agent is used to send all execution data in the feature engineering module and a feature engineering processing completion mark to the agent.

5. The interactive modeling system according to claim 4, characterized in that The agent is further configured to send a feature engineering processing instruction and the preprocessed modeling data to the feature engineering agent based on the modeling steps, all execution data in the data preprocessing module, and the data preprocessing completion flag; The feature engineering agent is also used to receive the feature engineering processing instructions and the pre-processed modeling data sent by the agent and forward them to the feature engineering assistant.

6. The interactive modeling system according to claim 5, characterized in that The model building module includes a model building user agent, a model building assistant, and a model building manager that are communicatively connected to each other, and a model building agent that is communicatively connected to the agent and the model building manager: The model building user agent is configured to receive the third interaction instruction and forward it to the model building assistant after entering the model building module; The model building assistant is configured to analyze the third interactive instruction to obtain a model building plan, and to perform model building on the modeling data after the feature engineering process based on the model building plan to obtain a model corresponding to the modeling data; The model building manager is used to collect all execution data in the model building module and feed it back to the user and the model building agent; The model building agent is used to receive the model building instructions and the modeling data after feature engineering processing sent by the agent and forward them to the model building assistant; and send all execution data in the model building module and the model building completion mark to the agent.

7. The interactive modeling system according to claim 6, characterized in that The agent is further configured to send a model building instruction and the modeling data after the feature engineering process to the model building agent based on the modeling steps, all execution data in the feature engineering module, and the feature engineering process completion flag; The model building agent is also used to receive the model building instructions and the modeling data processed by the feature engineering sent by the agent and forward them to the model building assistant.

8. The interactive modeling system according to claim 7, characterized in that: The intelligent agent is also used to collect and analyze historical input interaction instructions and all historical execution data in the data preprocessing module, the feature engineering module and the model building module to derive user behavior preferences; and dynamically optimize the intelligent agent, the data preprocessing module, the feature engineering module and the model building module according to the user behavior preferences.

9. The interactive modeling system according to claim 1, characterized in that: The intelligent agent is further configured to receive and analyze a fourth interaction instruction input, and update the modeling step based on the fourth interaction instruction.

10. The interactive modeling system according to claim 9, characterized in that: The first interaction instruction, the second interaction instruction, the third interaction instruction and the fourth interaction instruction are each one of an empty input instruction, a natural language description instruction and a termination instruction.

Images