Back to search

National and Kapodistrian University of Athens

Αυτόματη ανάλυση συναισθήματος σε πολυγλωσσικά κείμενα σε μέσα κοινωνικής δικτύωσης

Abstract

dc:description

This study examines multilingual sentiment analysis in social media. Sentiment analysis refers to techniques in Natural Language Processing (NLP) used to study the emotional states and subjective information (Liu, 2015, Cambria et al., 2017). In its simplest form, sentiment analysis deals with the polarity of a given text, i.e., whether the opinion expressed in it is positive or negative. While sentiment analysis research initially focused on product reviews written in English, recently, there is a lot of interest in sentiment analysis applications in social media (Pozzi et al., 2017) and multilingual texts (Liu, 2015).In this study, different sentiment analysis methods are compared in terms of their performance on datasets consisting of Greek, English, and Italian comments on Twitter, further tagged for topic ("general comments", "politics", "technology", “fashion”). The performance of different methods on different datasets is thus compared, as well as the same exactly method on different datasets. The methods examined are the following: (a) a Naive Bayes (NB) algorithm combined with n-grams, (b) Support Vector Machines (SVM) algorithm combined with n-grams and the tf-idf weighting scheme, (c) Neural networks (CNN, LSTM) with Glove and fastText word embeddings and (d), the BERT language model. The results show that BERT has achieved the best performance in all experiments except one and in all languages ​​overall compared to the rest of the methods, confirming thus the results in Devlin et al. (2019) that show the model to be highly efficient. The results also show no significant difference between the SVM and NB-based algorithms in performance, although previous research has shown conflicting results (Pang et al., 2002; cf. Wang & Manning, 2012). Similarly, the use of either bigrams or unigrams (words) was shown to achieve similar performance in this study. The results show that the use of the tf-idf weighting scheme is very effective in improving performance, confirming previous studies regarding the efficiency of the scheme (Markopoulos et al., 2015). Finally, word embeddings outperformed simple n-gram models, while LSTM-based models performed better than CNN-based models.

Degree

thesis:*
Grantor dc:publisher
National and Kapodistrian University of Athens
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Bilianos, Dimitrios
  • Μπιλιανός, Δημήτριος

Subjects

dc:subject × 8

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/54689

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Bilianos, Dimitrios; Μπιλιανός, Δημήτριος. Αυτόματη ανάλυση συναισθήματος σε πολυγλωσσικά κείμενα σε μέσα κοινωνικής δικτύωσης. National and Kapodistrian University of Athens, 2023. http://hdl.handle.net/10442/hedi/54689