Cross-language opinion lexicon extraction using mutual-reinforcement label propagation

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Growing interest in automatic opinion lexicon construction from sources like product reviews.
• Existing methods often rely on external resources (e.g., WordNet), limiting their applicability.
• Unsupervised/semi-supervised learning offers solutions but faces challenges with imbalanced datasets and scarce corpora in different languages.

Purpose of the Study:

• To address limitations in multilingual opinion lexicon extraction, specifically data scarcity and imbalance.
• To develop a framework that enables cross-lingual knowledge transfer for lexicon building.
• To improve the accuracy and applicability of automatic opinion lexicon generation.

Main Methods:

• Exploration of a mutual-reinforcement label propagation framework.
• Application of a label propagation algorithm to a word relation graph for each language.
• Utilization of a bilingual dictionary as a bridge for inter-lingual information transfer.

Main Results:

• The proposed framework facilitates mutual learning and reinforcement between languages.
• Experimental results demonstrate significant outperformance compared to baseline methods.
• The approach effectively addresses challenges of data imbalance and scarcity in multilingual settings.

Conclusions:

• The mutual-reinforcement label propagation framework is effective for multilingual opinion lexicon extraction.
• The method successfully enables cross-lingual knowledge transfer, boosting performance.
• This approach enhances the development of opinion lexicons, particularly for low-resource languages.