The introduction of Large Language Models (LLMs) has been a significant advancement in Artificial Intelligence. These models face unique challenges in the finance industry but have seen progress in financial text summarization, stock price predictions, financial report production, news sentiment analysis, and financial event extraction. However, in the Chinese financial market, LLMs lack an in-depth understanding of the industry. To address this, researchers have introduced DISC-FinLLM, a comprehensive approach to creating Chinese financial LLMs. This method aims to enhance the LLMs’ skills in generating and comprehending financial text, having multi-turn conversations, and assisting financial modeling. DISC-FinLLM is built using a Multiple Experts Fine-tuning Framework and has been evaluated to outperform the base model in various financial tasks.
The Advancement of Large Language Models in the Finance Industry
The field of Artificial Intelligence has witnessed a significant breakthrough with the introduction of Large Language Models (LLMs). These models, based on Natural Language Processing (NLP), are capable of handling large and complex datasets. In the finance industry, LLMs have revolutionized various areas such as financial text summarization, stock price prediction, financial report production, news sentiment analysis, and financial event extraction.
Challenges Faced by LLMs in Finance
As the volume and complexity of financial data continue to increase, LLMs encounter several challenges. These include:
- Lack of human-labeled data
- Lack of finance-specific expertise
- Difficulty in multitasking
- Constraints of numerical computing
- Inability to handle real-time information
While LLMs like GPT-4 excel in dialogue abilities, command comprehension, and following directions, they lack a deep understanding of the financial industry, especially in markets like China.
Introducing DISC-FinLLM: A Comprehensive Approach
To address this issue, a team of researchers has developed DISC-FinLLM, a comprehensive approach for creating Chinese financial LLMs. The main objective of this method is to equip LLMs with the ability to generate and comprehend financial text, engage in multi-turn conversations about financial issues, and assist with financial modeling and knowledge-enhanced systems.
The DISC-FIN-SFT Instruction Dataset
The researchers have also created a supervised instruction dataset called DISC-FIN-SFT, which serves as the foundation for constructing DISC-FinLLM. This dataset includes the following primary categories:
- Financial Consulting Instructions: Answering inquiries and offering guidance on financial matters.
- Financial Task Instructions: Assisting with various financial chores.
- Instructions on Financial Computing: Addressing financial statistical, computational, and modeling issues.
- Retrieval-Enhanced Instructions: Facilitating knowledge retrieval through questions, references, and answers.
The MEFF Architecture and Assessment Benchmarks
The construction of DISC-FinLLM involves a Multiple Experts Fine-tuning Framework (MEFF) and four Low-rank adaptation (LoRA) modules trained on different dataset segments. This architecture enables the model to perform well in various financial scenarios and jobs, catering to different user groups.
Multiple assessment benchmarks have been conducted to evaluate the performance of DISC-FinLLM. The results demonstrate its superiority over the base foundation model in all downstream tasks.
Practical AI Solutions for Middle Managers
If you want to leverage AI to evolve your company and stay competitive, consider implementing DISC-FinLLM. It offers practical solutions for:
- Identifying automation opportunities
- Defining measurable KPIs
- Selecting customized AI tools
- Implementing AI gradually through pilot projects
To learn more about AI solutions and how they can redefine your sales processes and customer engagement, visit itinai.com.
