CapTE: Combining Transformers and Capsule Networks for Stock Prediction

Summary

The document summarizes CapTE, a model for predicting stock movements from social media text. A Transformer encoder extracts semantic features from posts, while a capsule network is used to represent structural relationships in the text. The approach is motivated by the difficulty of capturing stock-specific meaning and associations in social media, and the paper reports better prediction performance than its benchmarks. The supplied material does not identify the datasets, benchmark definitions, metrics, or numerical results, so the strength and scale of that evidence cannot be assessed here.

The authors describe the method as text-only, with no financial data added to the model, and suggest it may generalize to other text-classification tasks. A central limitation is the daily frequency of the experiments: social media effects may be concentrated on the day a post or event occurs, making the signal less useful for next-day predictions. The summary therefore presents a modeling concept rather than evidence of a tradable strategy, and it leaves intraday prediction and market execution unaddressed.

Key ideas

• CapTE uses a Transformer encoder to extract semantic features from social media text.
• A capsule network models structural relationships in the encoded text.
• The summarized study reports improved stock movement prediction against benchmarks, but gives no metrics here.
• The model uses text without financial inputs, according to the document.
• Daily data may fail to capture shorter-lived effects from social media posts.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.