Stock Prediction via Sentimental Transfer Learning
Bibliographic record
Abstract
Stock prediction is always an attractive problem. With the expansion of information sources, news-driven stock prediction based on sentiments of social media, such as sentiment polarities in financial news, becomes more and more popular. However, the distributions of news articles among different stocks are skewed, which makes stocks with few news have few training samples for their prediction models, and thus leads to low prediction accuracy in the stock predictions. To address this problem, we propose sentimental transfer learning, which transfers sentimental information learned from news-rich stocks (source) to the news-poor ones (target), and prediction performances of the later ones are, therefore, improved. In this approach, the financial news articles of both the source and target stocks are first mapped into the same feature space that is constructed by sentiment dimensions. Second, we develop three different transfer principles in order to explore different transfer scenarios: 1) the source and target stocks’ historical price time series are highly correlated; 2) the source and target stocks are in the same sector and the former is the most news-rich one in the sector; and 3) the source stock has the highest prediction performance in validation data set. Third, a majority voting mechanism is designed based on the principles. The voting mechanism is to select the most proper source stock from the candidate stocks that are generated by different principles. Stock predictions are finally made based on the prediction models trained on the selected stocks. Experiments are conducted based on the data of Hong Kong Stock Exchange stocks from 2003 to 2008. The empirical results show that sentiment transfer learning can improve the prediction performance of the target stocks, and the performances are better and more stable with the source stocks selected by the voting mechanism.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".