Research on Complex Content Understanding and Intelligent Applications of Dialogue Platforms Powered by Generative Large Language Models

1. Research Background

With the explosive growth of data in the financial field and customers' continuous pursuit of intelligent services, the application of pre-trained large language models in the securities industry has become an important research direction. Large language models have a wide range of application prospects in various businesses such as wealth management, compliance risk control, etc. They can be applied to intelligent customer service, intelligent investment advisor, and intelligent investment research scenarios to improve customer satisfaction and employee work efficiency.

This topic explores the application of large language models in various business scenarios in the securities industry by using the private professional data of securities firms, and quickly promotes the application of large language models by using mature docking tools. These studies will help to solve the pain point problems of the securities industry and promote the intelligentization process of the securities industry.

2. Content of the Research

This topic privatizes and deploys billion-to-ten-billion-level parameter scale generative large language models to provide intelligent Q&A, intelligent stock diagnosis, product introduction, and other services for wealth, investment, investment research, risk control, and other businesses. In terms of wealth business, the intelligent dialog platform can assist account managers and customer service personnel in customer service and improve service efficiency. In terms of customer service, it provides information broadcasting, wealth assistant, product introduction and other services through the company's trading terminal software, reaches customers in a vivid form, realizes personalized value output in business, achieves rapid creation and reduces operating costs through technical empowerment; in terms of investment research and investment banking business and internal application, it provides intelligent downgrading comprehension of complex contents such as research reports, announcements, and offering brochures and reaches business personnel directly through dialogues, improving their work. In investment research, investment banking business, and internal applications, intelligent dimensional understanding of complex content, such as research reports, announcements, and offering prospectuses, can be directly accessed through dialogue to enhance the efficiency of business personnel.

Relying on the company's data governance system and professional high-quality financial data, this topic utilizes the core capabilities of the large language model to build general large language model content understanding and intelligent dialogue capabilities, which realizes the empowerment of business personnel.

3. Research Achievements

The core technology and innovation points of this topic have gained good results in many business scenarios of the company:

Customer service scenarios based on intelligent dialogue capabilities: customer service Q&A is one of the common scenarios in the field of financial services, the large language model is based on a professional financial knowledge base that has accumulated a lot of experience, and by combining it with a distributed data search engine, it carries out the optimization of the knowledge base in terms of subword optimization and indexing design, which allows customer service personnel to retrieve information about the questions quickly and summarize the relevant knowledge to give a more comprehensive and all-rounded answer, which significantly improves the efficiency and quality of customer service. Customer service efficiency and quality.

Intelligent investment advisor based on investment underlying volume and price analysis: This topic innovatively integrates the theoretical framework of finance with large language models and applies it to retail customer service. Based on the investment volume and price strategy, multi-factor model and time series analysis technology, the large language model can be used to predict the future volume and price movements of the investment underlying and the possible investment opportunities and risks, and combined with the results of the volume and price model of the investment underlying and the analysis of the client's positions, it can reveal the risks and problems of the client's positions in a timely manner, and provide the client with more professional investment guidance and advice.

Intelligent recommendation and search based on customer intent recognition: Using generative large language model technology, we accurately analyze video, text, pictures and other content to build a multimodal content labeling system. Combined with customer profiles, customer behavior and other relevant data, through a combination of explicit and implicit recall algorithms, it accurately matches content with customers, accurately locates the interest groups of relevant content, and builds intelligent recommendation, intelligent Q&A, intelligent search, and intelligent push engines for efficient and personalized distribution of content through Q&A, search, recommendation, push and other means.

Relevant data show that after using the algorithm of this topic, the click rate of customer information content has increased by 21%, the number of transactions has increased by 9.6 times, the amount of transactions promoted after information reading is 1.44 billion yuan, and the penetration rate of information customers has reached 10.6%.

Investment assistant based on content-intelligent dimensional reduction technology: Securities industry information is highly complex and specialized, including a large number of research reports, investment education knowledge, announcements, market data, public opinion information, and laws and regulations in terms of data and information. Practitioners in the securities industry have rich business knowledge and experience, but it is difficult to realize rapid processing of diversified and massive information by relying on human resources. This project makes full use of the advantages of the large language model to carry out timely and accurate intelligent downgrading and interpretation of high-quality and complex professional content with multi-dimensional heterogeneity, and distills highly generalized information on business knowledge, economic conditions, industry trends, product features, and regulatory points, so as to provide more accurate information services for account managers, customer service personnel, and other professional business personnel.

4. Summary & Outlook

In recent years, with the rapid development of big data technology, it has provided technical support for solving the pain points and difficulties in the securities industry. By privatizing the deployment and application of large language models, this project has actively explored and practiced in customer service, customer analysis, intelligent recommendation, professional article analysis, etc., and has achieved some results. In the future, with the development of large language model technology, smarter, more reliable, and lower-cost large language models will appear one after another, which will more effectively solve the many problems and challenges faced by the current industry.

Note: This project was awarded the Third Prize of the 2023 Industry Co-Research Project by the Securities Information Technology Research and Development Center (Shanghai).