We live in a society where a massive quantity of data is generated daily on online social network platforms. This enormous data contains vital opinion-related information that many companies and other scientific and commercial industries are trying to exploit for their benefits. For that purpose, sentiment analysis is required. Sentiment analysis or opinion mining is the branch of data analytics for extracting sentiments from messages expressed by users on a particular subject. Although, in the past years a considerable research has been made for the English language, the works of Sentiment Analysis in Greek language is not so popular, due to smaller user base. In this work, we provide a method to create domain-specific dictionaries given a corpus of tweets in the Greek language. In those Lexicons, we take into consideration the significance of each word for the specific domain, by introducing a new attribute Weightw. Also, we deploy a hybrid framework which utilizes the newly created domain-specific Lexicon with the Naïve Bayes classifier to analyze and predict the sentiment of each tweet. Our framework has the ability to merge the better of the two basic concepts, the Lexicon and Machine Learning method, and demonstrates the significance of the words for domain-specific Lexicon, for achieving optimal results when performing Sentiment Analysis.
5 February 2026 in Social Network Analysis and Mining
This paper presents a hybrid sentiment analysis framework tailored for the Greek language, combining a novel lexicon construction methodology with an Active Learning (AL) strategy to enhance performance and minimize annotation effort. A domain-specific lexicon was developed using two original metrics—Score w , capturing the sentiment polarity of words, and Weight w , reflecting their combined emotional and statistical importance. These metrics enabled the transformation of tweet-level annotations into semantically rich word-level representations. To evaluate the framework, four classification models (Naïve Bayes, Logistic Regression, SVM, and Decision Trees) were trained with and without lexicon-derived features. The results showed substantial improvements in accuracy and F1-score when incorporating the lexicon, especially for Logistic Regression and Decision Tree models. Furthermore, a dynamic Active Learning approach was employed, where the lexicon was reconstructed in each cycle using the evolving labeled dataset. This strategy achieved the same levels of accuracy and Macro F1-score, as with full-supervision framework, while using only 30% of the data. The findings demonstrate the efficacy of combining lexicon enhancement with uncertainty-based sampling, offering a cost-effective and scalable solution for sentiment analysis in low-resource languages such as Greek.
Sentiment analysis for Greek social media texts remains challenging because of limited annotated resources, linguistic variation, and domain-dependent sentiment expression. This study presents a topic-aware, lexicon-guided framework for sentiment classification across five reference domains in Greek social media. Domain-specific sentiment lexicons are activated according to the relevant domain and transformed into a structured representation comprising a token-level multi-channel lexical matrix and aggregate lexical descriptors. A fusion convolutional neural network combines these complementary components to classify sentiment while retaining explicit lexical evidence for inspection. The evaluation follows a leakage-free protocol: lexicons are constructed exclusively from the sentiment inner-training subset, validation data are used for model selection, and a held-out test set is reserved for final evaluation. The proposed fusion CNN achieved the strongest held-out sentiment result among the evaluated models, with an Accuracy of 0.8029 and a Macro-F1 of 0.7883, exceeding TF–IDF + Linear SVM and fine-tuned GreekBERT baselines in the present experimental setting. Ablation results show that the token-level lexical matrix and global descriptors provide complementary information. For domain routing, GreekBERT late fusion achieved an Accuracy of 0.9162 and a Macro-F1 of 0.9116. When lexicon activation used predicted rather than reference domains, the end-to-end sentiment pipeline achieved a Macro-F1 of 0.7569. These findings indicate that explicit domain-specific lexical knowledge can support an interpretable sentiment representation while making the effects of lexical coverage and topic-routing uncertainty visible.