Good Debt or Bad Debt: Detecting Semantic Orientations in Economic Texts

TL;DR

The paper tackles the challenge of detecting semantic orientation in financial news, where domain-specific language and event directionality complicate traditional sentiment analysis. It introduces the Linearized Phrase Structure (LPS) model, a three-stage framework that combines domain-adapted lexicons with phrase-structure information and a multi-label SVM to classify short financial texts as positive, negative, or neutral. Key contributions include a human-annotated finance phrase-bank, lexicon enhancements (including directional expressions and financial entities), and the LPS model that integrates these components with a low-dimensional, scalable representation. Empirical results show LPS outperforms traditional lexical baselines and demonstrate the value of pruning and context-aware features, suggesting practical benefits for near-real-time financial sentiment analysis and downstream decision-making.

Abstract

The use of robo-readers to analyze news texts is an emerging technology trend in computational finance. In recent research, a substantial effort has been invested to develop sophisticated financial polarity-lexicons that can be used to investigate how financial sentiments relate to future company performance. However, based on experience from other fields, where sentiment analysis is commonly applied, it is well-known that the overall semantic orientation of a sentence may differ from the prior polarity of individual words. The objective of this article is to investigate how semantic orientations can be better detected in financial and economic news by accommodating the overall phrase-structure information and domain-specific use of language. Our three main contributions are: (1) establishment of a human-annotated finance phrase-bank, which can be used as benchmark for training and evaluating alternative models; (2) presentation of a technique to enhance financial lexicons with attributes that help to identify expected direction of events that affect overall sentiment; (3) development of a linearized phrase-structure model for detecting contextual semantic orientations in financial and economic news texts. The relevance of the newly added lexicon features and the benefit of using the proposed learning-algorithm are demonstrated in a comparative study against previously used general sentiment models as well as the popular word frequency models used in recent financial studies. The proposed framework is parsimonious and avoids the explosion in feature-space caused by the use of conventional n-gram features.

Figures (4)

• Figure 1: Directional-dependence: Semantic orientation of a financial entity often depends on the favorable or unfavorable direction of events.
• Figure 2: Entity pruning rules: To reduce the number of entities and improve the information value of the remaining ones, heuristic rules are applied to merge neutral entities and take into account the effects of polarity influencers on other entities.
• Figure 3: The entity extractor is operationalized in two steps: (1) The first step is to identify all entity candidates: For every lexicon entry, we check whether a subset of leaves in the phrase-structure matches; (2) The second step is to replace the leaves with lexicon entities, and apply pruning rules to merge neutral entities and take into account the effects of polarity influencers. As final output, the extractor maps the phrase into a sequence of entities, which are positioned in their order of appearance.
• Figure 4: The figure shows the distribution of entity-sequence lengths (before pruning) in the financial phrase-bank, which consists of 5000 sentences. An entity-sequence is a linear representation of the phrase structure produced by the entity-extractor.

Theorems & Definitions (1)

• Example 2