A Combined Approach for Extracting Financial Instrument-Specific Investor Sentiment from Weblogs

This work proposes a novel approach for investor sentiment extraction from blogs by combining machine-learning on the document-level and knowledge- based information extraction on the sentence-level, and suggests that the combined approach achieves a higher precision compared to a standalone knowledge-based approach.

Abstract

Investor sentiment about future returns of financial instruments is a highly relevant information source for investment managers and other stakeholders in the financial industry. Investor sentiments are abundant in financial blog texts. Making use of these sentiments constitutes a massive information management challenge when considering the millions of blog articles with everchanging and growing amounts of information that need to be acquired and interpreted. We propose a novel approach for investor sentiment extraction from blogs by combining machine-learning on the document-level and knowledgebased information extraction on the sentence-level. The proposed artifact is a financial instrument-specific investor sentiment extraction method, which we apply to a set of blog articles. The evaluation suggests that the combined approach achieves a higher precision compared to a standalone knowledge-based approach.