Implementation method for intelligent investment research assistant Agent of financial large model

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a financial large model intelligent investment research assistant Agent implementation method, which comprises a market environment module, a long-memory investment research assistant Agent module and a layered long-term memory module, and is characterized in that the market environment module is used for monitoring key information sources of an external market, obtaining latest information and data, and sending the latest information and data to the long-term memory module; the data is pushed to a long-memory investment research assistant Agent model; and providing a query for the key index data to be called by the long memory investment research assistant Agent module. According to the method, induction and extraction of various data and information can be realized to serve as input of subsequent analysis and reasoning; required key data can be inquired from the outside and fused into analysis and reasoning; according to knowledge and historical data mastered by the model and learned new data and information, in combination with key index data inquired from the outside, an analysis and reasoning result can be generated.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a method for implementing an intelligent investment research assistant Agent for a financial large model. Background Art

[0002] In the era of information explosion, there is an urgent need for information noise reduction in investment. Investors need intelligent tools to improve means of information aggregation and management, release the value of information, and enhance the speed and rationality of their own decision-making.

[0003] With the progress of artificial intelligence technology, especially the emergence of highly intelligent large model technologies such as LLM (LLM is an abbreviation for "Language Model". It is an artificial intelligence technology mainly used to understand and generate natural language text. LLM can master the laws of language by learning a large amount of text data, so as to be able to generate fluent sentences and make appropriate responses to user inputs), the maturity of Deep Learning technology and applications, continuous large R & D investment, and the continuous breakthrough of AI computing power, many traditional financial companies and fintech companies have begun to research the application of AI in the financial field, as well as the R & D and use of related products based on AI. In particular, AI financial products based on LLM large models can greatly improve the intelligent performance of products, as well as customer investment satisfaction and activity. This reflects the following technical advantages of the LLM large model: (1) It can demonstrate human-like understanding and interaction capabilities (2) It can master knowledge in professional fields and reasonably apply it for analysis and speculation (3) It can show the memory ability of a large amount of historical data, that is, the data used during training Despite the above advantages, however, when the single LLM large model technology is applied to the financial field, there are some natural limitations or deficiencies as follows: (1) It requires a large amount of data for training, with high computing costs and long cycles (2) There is a certain lag in training data, resulting in the inability to reflect the latest information (3) Correct and effective decision-making and analysis in the financial field sometimes do not rely solely on historical experience and knowledge. Sometimes, it is necessary to rely on recent events, or even master real-time events, to obtain timely information in order to make correct decisions and analyses.

[0004] Therefore, there is an urgent need for a new technical solution to solve the above technical problems. Summary of the Invention

[0005] The purpose of the present invention is to overcome the problems of the above-mentioned existing technologies, and provides a method for implementing an intelligent investment research assistant Agent for a financial large model, so as to solve the deficiencies shown by traditional LLM large models in combining data timeliness during analysis and reasoning, thereby causing the inability to reflect the latest information, as well as the technical problems of being unable to make correct decisions and analyses.

[0006] The above purpose is achieved through the following technical solutions: A method for implementing an intelligent investment research assistant Agent for a financial large model, including a market environment module, a long-term memory investment research assistant Agent module, and a hierarchical long-term memory module. The market environment module is used to monitor key information sources in the external market, obtain the latest information and data, and push them to the long-term memory investment research assistant Agent module; and provide queries for key indicator data, and provide a unified encapsulated interface for the long-term memory investment research assistant Agent module to call; The long-term memory investment research assistant Agent module includes: A summary module, which generates the latest memorable information and data based on the latest information and data, and stores them in the hierarchical long-term memory module; An observation module, which is used to monitor and query the latest and recent key indicator data as key financial indicator data; A reflection module, which, based on the professional financial knowledge and experience it has mastered, combines the latest memorable information and data and the key financial indicator data, and instantaneously realizes investment research output according to the investment target and parameters input by the user, including an investment research briefing, reasoning details, and an investment decision.

[0007] Further, the reflection module includes an immediate reflection module and an extended reflection module. The immediate reflection module, based on the professional financial knowledge and experience it has mastered, combines the latest memorable information and data and the key financial indicator data, and analyzes and generates the investment research briefing according to the investment target and parameters input by the user; the extended reflection module analyzes and generates the reasoning details and the investment decision according to the historical data of the investment target and the key financial indicator data, combined with news that is likely to cause emotional positive and negative impacts, according to the investment target and parameters input by the user.

[0008] Further, the investment target and parameters include a stock code and a trading date.

[0009] Further, the latest information and data include hot discussions in WeChat groups, hot public opinions on Weibo, information news of listed companies, information disclosure of listed companies, and annual reports of listed companies.

[0010] Furthermore, the key financial indicator data includes daily stock prices, important fund indices, important currency exchange rates, and gold prices.

[0011] Furthermore, according to the timeliness of the received latest memorable information and data, the hierarchical long-term memory module stores the latest memorable information and data in: A shallow processing unit for storing the latest memorable information and data released within one month; A middle processing unit for storing the latest memorable information and data released more than one month and within one year; A deep processing unit for storing the latest memorable information and data released more than one year ago.

[0012] Furthermore, the shallow processing unit, the middle processing unit, and the deep processing unit perform self-update on the latest information and data stored by them according to time, including: If the storage time of the latest memorable information and data in the shallow processing unit exceeds one month, it is transferred to the middle processing unit; If the storage time of the latest memorable information and data in the middle processing unit exceeds one year, it is transferred to the deep processing unit.

[0013] Furthermore, in the deep processing unit, according to the time and quantity of the stored latest memorable information and data, a threshold for old data and information processing is set, and the latest memorable information and data exceeding the threshold for old data and information processing are regarded as old data and information, and fine-tuning training of the basic large model is started. After the training is completed, the old data and information are then eliminated.

[0014] The implementation method of a financial large model intelligent investment research assistant Agent provided by the present invention can effectively solve the deficiencies shown by traditional LLM large models in combining data timeliness during analysis and reasoning, thereby causing the inability to reflect the latest information, as well as the technical problems of being unable to make correct decisions and analyses. It can achieve the induction and extraction of various data and information as the input for subsequent analysis and reasoning; it can query the required key data from the outside and integrate it into the analysis and reasoning; it can generate the results of analysis and reasoning based on the knowledge and historical data mastered by the model, as well as the newly learned data and information, and in combination with the key indicator data queried from the outside. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a functional framework diagram of the long-term memory investment research assistant Agent module in the implementation method of a financial large model intelligent investment research assistant Agent of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0017] As Figure 1 shown, in order to facilitate the use of AI technology to extract valuable data from the learned professional financial knowledge, historical information and data, and a large amount of various real-time information mastered, and perform reasoning and analysis, so as to provide investment reference suggestions and viewpoints for ordinary investment users or professional investment personnel. This solution provides a method for implementing a financial large model intelligent investment research assistant Agent, including a market environment module, a long-term memory investment research assistant Agent module, and a hierarchical long-term memory module, where: The market environment module is used to monitor key information sources in the external market, obtain the latest information and data, and push them to the long-term memory investment research assistant Agent module; and provide queries for key indicator data, and provide a unified encapsulated interface for the long-term memory investment research assistant Agent module to call; in this embodiment, the latest information and data include hot discussions in WeChat groups, hot public opinions on Weibo, information news of listed companies, information disclosure of listed companies, and annual reports of listed companies, etc.; the key financial indicator data includes daily stock prices, important fund indexes, important currency exchange rates, and gold prices, etc.

[0018] The long-term memory investment research assistant Agent module trains a basic large model in the financial field based on a large amount of financial knowledge, data and information, monitors changes in external important information sources, and processes the change information according to different real-time standards according to the importance and urgency, such as some information needs to be processed immediately, some can be processed at specific idle time points, and some are processed once every hour, etc. Its specific capabilities are as follows: The ability to summarize and extract various data and information, as the input for subsequent analysis and reasoning, that is, the function of the summary module of the basic large model; The ability to query the required key data from the outside and integrate it into the analysis and reasoning, that is, the function of the observation module of the basic large model; It has the ability of analysis and reasoning, and can generate the results of analysis and reasoning according to the knowledge and historical data mastered by the model, as well as the newly learned data and information, and combined with the key indicator data queried from the outside, that is, the function of the reflection module of the basic large model.

[0019] As a specific description of the above summary module, observation module and reflection module, as follows: Summary module, which generates the latest memorable information and data based on the latest information and data, and stores them in the hierarchical long-term memory module for subsequent precise analysis and reasoning; by summarizing the external changing data and information, it generates data and information with higher real-time performance than large models; Observation module, which is used to monitor and query the latest and recent key indicator data as key financial indicator data; Reflection module, which, based on the professional financial knowledge and experience it has mastered, combines the latest memorable information and data and the key financial indicator data, and instantaneously realizes investment research output, including investment research briefings, reasoning details, and investment decisions according to the investment targets and parameters input by the user. Specifically, according to the financial knowledge, historical information, and data it has mastered, it obtains various updated and time-sensitive data learned from the hierarchical long-term memory module. Each type of data has different influence factors on the final result, and usually the latest data has a greater influence; and it combines the functions of the observation module of the long-term memory investment research assistant Agent module to query key indicator data externally for reasoning and analysis; finally, it generates an investment research report, gives reasoning details, and output results such as investment decisions.

[0020] As an optimization of the reflection module in this embodiment, the reflection module includes: Immediate reflection module, which analyzes and generates the investment research briefing according to the professional financial knowledge and experience it has mastered, combines the latest memorable information and data and the key financial indicator data, and according to the investment targets and parameters input by the user; Extended reflection module, which analyzes and generates the reasoning details and the investment decision according to the historical data of the investment target and the key financial indicator data, combined with the news that is likely to cause emotional positive and negative impacts, and according to the investment targets and parameters input by the user.

[0021] The investment targets and parameters in this embodiment include stock codes and trading dates, but are not limited to, for example, the latest unpublished information mastered by the user; the user inputs the investment targets and parameters to be analyzed, such as stock codes and trading dates, and calls the functions of the summary module in the long-term memory investment research assistant Agent module.

[0022] In this embodiment, the hierarchical long-term memory module stores the latest memorable information and data in the shallow processing unit, middle processing unit, and deep processing unit according to the timeliness of the received latest memorable information and data. Specifically: The shallow processing unit is used to store the latest memorable information and data released within one month, that is, the latest information and data, such as daily market news from mainstream media, hot topics in social media, information disclosure by listed companies, etc. The middle-level processing unit is used to store the latest memorable information and data released for more than one month and within one year, such as monthly reports, quarterly summaries, semi-annual reports, etc. of listed companies; The deep processing unit is used to store the latest memorable information and data released more than one year ago, such as company annual reports, annual plans and goals, etc.

[0023] As an optimization of this embodiment, the shallow processing unit, the middle processing unit and the deep processing unit self-update the latest information and data stored in each according to time, that is, data and information will enter the middle processing unit from the shallow processing unit and enter the deep processing unit from the middle processing unit according to time, specifically: If the storage time of the latest memorizable information and data in the shallow processing unit exceeds one month, it is transferred to the middle processing unit; If the storage time of the latest memorizable information and data in the middle-level processing unit exceeds one year, it will be transferred to the deep-level processing unit.

[0024] As a further optimization of this embodiment, when the data and information stored in the deep processing unit reaches a certain time and quantity, in order to avoid excessive increase in the processing overhead of analytical reasoning, resulting in a decrease in processing speed, this solution customizes the long-term data and information processing threshold in the deep processing unit according to the time and quantity of the latest memorable information and data stored, and treats the latest memorable information and data exceeding the long-term data and information processing threshold as long-term data and information, and starts fine-tuning training of the basic large model. After the training is completed, the long-term data and information are eliminated, so as not to lose the impact of these data and information on analytical reasoning.

[0025] The above description is only for illustrating the implementation mode of the present invention and is not intended to limit the present invention. For those skilled in the art, any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for implementing a financial large model intelligent investment research assistant Agent, characterized in that: It includes a market environment module, a long-memory investment research assistant Agent module and a hierarchical long-term memory module. The market environment module is used to monitor key information sources of the external market, obtain the latest information and data, and push them to the long-memory investment research assistant Agent module; And provide query of key indicator data, and provide a unified packaged interface for the long memory investment research assistant Agent module to call; The long memory investment research assistant Agent module includes: A summary module, wherein the summary module generates the latest memorable information and data according to the latest information and data, and stores the latest memorable information and data in the hierarchical long-term memory module; An observation module, the observation module is used to monitor and query the latest and most recent key indicator data as key financial indicator data; A reflection module, which, based on its own professional financial knowledge and experience, combines the latest memorable information and data with the key financial indicator data, and according to the investment targets and parameters input by the user, instantly realizes investment research output, including investment research briefs, reasoning details and investment decisions.

2. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 1 is characterized by: The reflection module includes an immediate reflection module and an extended reflection module. The immediate reflection module generates the investment research briefing based on the professional financial knowledge and experience it possesses, the latest memorable information and data, and the key financial indicator data, and the investment targets and parameters input by the user; The extended reflection module analyzes and generates the reasoning details and the investment decision based on the historical data of the investment target and the key financial indicator data, combined with the messages that are likely to cause emotional positive and negative effects, and based on the investment target and parameters input by the user.

3. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 1 or 2, characterized in that: The investment targets and parameters include stock codes and transaction dates.

4. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 1 or 2, characterized in that: The latest information and data include hot discussions in WeChat groups, hot public opinions on Weibo, news on listed companies, credit approvals of listed companies and annual reports of listed companies.

5. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 1 or 2, characterized in that: The key financial indicator data include daily stock prices, important fund indices, important currency exchange rates and gold prices.

6. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 4 is characterized by: The hierarchical long-term memory module stores the latest memorable information and data received according to the timeliness of the latest memorable information and data in: A shallow processing unit for storing the latest memorizable information and data released within one month; A middle-level processing unit for storing the latest memorizable information and data released for more than one month and within one year; The deep processing unit is used to store the latest memorizable information and data released more than one year ago.

7. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 6 is characterized by: The shallow processing unit, the middle processing unit and the deep processing unit self-update the latest information and data stored in each according to time, including: If the storage time of the latest memorizable information and data in the shallow processing unit exceeds one month, it is transferred to the middle processing unit; If the storage time of the latest memorizable information and data in the middle-level processing unit exceeds one year, it will be transferred to the deep-level processing unit.

8. The method for implementing a financial large model intelligent investment research assistant Agent according to claim 7 is characterized by: In the deep processing unit, a long-term data and information processing threshold is set according to the time and quantity of the latest memorable information and data stored, and the latest memorable information and data exceeding the long-term data and information processing threshold is used as long-term data and information, and fine-tuning training of the basic large model is started. After the training is completed, the long-term data and information are eliminated.

Images