Building and evaluating resources for sentiment analysis in the Greek language

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Social Media Analysis

Background:

• Sentiment analysis relies on resources like lexicons and word embeddings, which are underdeveloped for Greek.
• Existing resources are insufficient for effective sentiment analysis in the Greek language.

Purpose of the Study:

• To address the lack of resources for Greek sentiment analysis.
• To develop and release publicly available resources for analyzing Greek online text.
• To improve performance on sentiment analysis and related tasks in Greek.

Main Methods:

• Creation of a manually annotated sentiment lexicon.
• Development of semi-supervised word embedding vectors for Greek.
• Annotation of datasets for various NLP tasks.
• Experimental evaluation using different algorithms and parameters.

Main Results:

• Achieved a 24.9% average relative improvement in F-score for cross-domain sentiment analysis.
• Demonstrated the effectiveness of the new resources over traditional n-gram features.
• Showcased promising results on related tasks like emotion analysis and sarcasm detection.

Conclusions:

• The developed resources significantly enhance sentiment analysis for the under-resourced Greek language.
• The resources are valuable for cross-domain sentiment analysis, emotion analysis, and sarcasm detection.
• Publicly releasing these resources facilitates further research and development in Greek NLP.