SPCM: A Machine Learning Approach for Sentiment-Based Stock Recommendation System

Abstract

Recommendation systems play a pivotal role in delivering user preference information. However, they often face the challenge of information cocoons due to repeated content delivery, particularly prevalent in stock recommendations that are susceptible to investor sentiment. In response to the information cocoons, we propose the Sentiment and Price Combined Model (SPCM), which leverages sentiment features and price factors to predict stock price movements. This novel framework combines collective sentiment analysis with state-of-the-art BERT transformer models and advanced machine learning techniques. Over a three-year period, we collected 40 million stock comments from the Guba platform, extracting investor sentiment conveyed in text information and investigating the impact of metrics such as homophily on stock recommendations. Experimental results indicate that both the volume of posts and the agreement index affect the effectiveness of investor sentiment, while homophily reduces the accuracy of participants’ stock price judgments. The recognition accuracy of the BERT-based sentiment analysis model reaches an impressive 84.12%, and the portfolio constructed by SPCM yields a cumulative return four times that of the industry benchmark. Furthermore, homogeneous quantitative metrics also enhance diversification in stock selection.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 24 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.