E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Current Record: 46 of 824

[Add to Basket]

Identifier

000452122

Title

Structural feature extraction and engineering for sentiment analysis

Alternative Title

Εξαγωγή και αξιοποίηση δομικών χαρακτηριστικών για ανάλυση συναισθήματος

Author

Περόντση, Εύα Γ.

Thesis advisor

Πρατικάκης, Πολύβιος

Reviewer

Τσαμαρδινός, Ιωάννης
Ιωαννίδης, Σωτήρης

Abstract

The subject of this work is the sentiment analysis of Greek-speaking tweets. We use natural language processing (NLP) methods and neural networks to create three different classification models. The first model processes single, independent tweets and decides if their sentiment is positive or not, or if it is negative or not. The second model considers a tweet paired with its textual context, meaning the tweet that it responds to. With the third neural model we attempt to do sentiment analysis with the tweet, textual context and some additional, structural features, as input. These structural features are extracted from the Twitter graph, and give us information about the authors of the tweet and textual context. Our experiments show that the additional text context improves our prediction by a small percentage in some cases. However, we find no correlation between the predicted tweet sentiment and the Twitter graph structural features.

Language

English

Subject

Data analysis

Neural networks

Social network analysis

Ανάλυση δεδομένων

Ανάλυση κοινωνικών δικτύων

Νευρωνικά δίκτυα

Issue date

2022-12-02

Collection