[Paper Review] A Classification of Adjectives for Polarity Lexicons Enhancement

Silvia Rodríguez Vázquez, Núria Bel|arXiv (Cornell University)|Mar 8, 2013

Sentiment Analysis and Opinion Mining11 references8 citations

TL;DR

This paper proposes a domain-aware classification of adjectives into domain-dependent and domain-independent categories, further subdivided by subjectivity degree (constant, mixed, highly subjective), to enhance polarity lexicons for sentiment analysis. By analyzing adjective polarity shifts across domains, the authors demonstrate that domain-specific polarity labeling significantly improves sentiment classification accuracy, offering a more nuanced alternative to static lexicons.

ABSTRACT

Subjective language detection is one of the most important challenges in Sentiment Analysis. Because of the weight and frequency in opinionated texts, adjectives are considered a key piece in the opinion extraction process. These subjective units are more and more frequently collected in polarity lexicons in which they appear annotated with their prior polarity. However, at the moment, any polarity lexicon takes into account prior polarity variations across domains. This paper proves that a majority of adjectives change their prior polarity value depending on the domain. We propose a distinction between domain dependent and domain independent adjectives. Moreover, our analysis led us to propose a further classification related to subjectivity degree: constant, mixed and highly subjective adjectives. Following this classification, polarity values will be a better support for Sentiment Analysis.

Motivation & Objective

• To address the limitation of static polarity lexicons that do not account for domain-specific polarity shifts in adjectives.
• To identify and classify adjectives based on their sensitivity to domain context.
• To improve sentiment analysis performance by incorporating domain-aware polarity annotations for adjectives.
• To propose a refined classification system that distinguishes between constant, mixed, and highly subjective adjectives.

Proposed method

• Annotate adjectives from multiple domains with their polarity values (positive, negative, neutral) to detect polarity shifts.
• Classify adjectives into domain-independent (polarity remains consistent across domains) and domain-dependent (polarity varies by domain) categories.
• Further categorize adjectives by subjectivity degree: constant (stable subjectivity), mixed (subjectivity varies), and highly subjective (highly context-sensitive).
• Use a multi-domain dataset to validate the classification through manual and statistical analysis.
• Integrate the refined classification into polarity lexicons to support more accurate sentiment classification.

Experimental results

Research questions

• RQ1 How do adjective polarities vary across different domains, and to what extent do they shift from their default polarity?
• RQ2 What proportion of adjectives exhibit domain-dependent polarity behavior versus domain-independent behavior?
• RQ3 How does subjectivity degree (constant, mixed, highly subjective) correlate with polarity stability across domains?
• RQ4 Can a domain-aware classification of adjectives improve the performance of sentiment analysis systems?

Key findings

• A majority of adjectives exhibit domain-dependent polarity, meaning their sentiment value changes depending on the context or domain.
• Domain-independent adjectives are relatively stable in their polarity across domains, while domain-dependent adjectives show significant polarity shifts.
• The classification into constant, mixed, and highly subjective adjectives reveals distinct patterns in subjectivity behavior across domains.
• The proposed classification leads to more accurate sentiment analysis by enabling context-sensitive polarity assignment in lexicons.
• The study demonstrates that static polarity lexicons are insufficient for robust sentiment analysis due to unaccounted domain effects.
• The results support the integration of domain-aware adjective classification into sentiment analysis pipelines to improve accuracy.