Recommender Systems for Sector-Specific Stock Analysis

Recommender systems are essential tools that assist users in making informed decisions by providing suggestions based on their previous actions and preferences. Traditional methods like content based and collaborative filtering work well for smaller datasets but often fail to extract information from large datasets due to their limited ability to capture complex patterns. This paper introduces a novel approach for stock recommendation that leverages a large language model, Llama 3.1, fine-tuned on financial news data. Financial news headlines and summaries were collected using the FinnHub API which served as the primary dataset. The system integrates stock price forecasting, sentiment analysis, and performance indicators to generate an effective analysis for decision making. The forecasting algorithm generates the stock price predictions which were integrated with sentiment analysis and stock performance indicators to create an informative prompt-response dataset for fine-tuning. The proposed system ranks stocks based on positive developments, potential concerns, and forecasted closing prices, providing Buy or Sell recommendations. Experimental results demonstrated moderate performance for predicting Buy recommendations, while Sell predictions exhibited lower accuracy comparatively. Analysis based on sectors revealed that the consumer cyclical and healthcare sectors yielded the best performance for Buy and Sell recommendations, respectively.