Sentiment Analysis of Textual Content in Social Networks From Hand-Crafted to Deep Learning-Based Models [Problem Statement]

 

.

...

 Most of the existing systems on sentiment analysis rely heavily on a rich set of sentiment resources (such as text corpora with manually annotated sentiment polarity, sentiment lexicons, and word embeddings). However, the distribution of sentiment resources is very imbalanced among languages. Thus, building a sentiment analysis system in low-resource languages requires tremendous human effort to construct such resources, which is a time-consuming and expensive task.

.

Although opinions are the most common shared content in online social networks and forums, people also tend to share their emotions which are the keys to their feelings and thoughts. Emotion analysis is the task of determining the attitude towards a target or topic. The attitude can be the polarity (positive or negative), or an emotional state such as joy, anger or sadness [101–103].

.

1.2 Objectives

In this thesis, we aim to develop methods to automatically analyse textual content shared on social networks and identify people’ opinions, emotions and feelings at different level of analysis and in different languages. As such, we introduce the following set of goals:

.

• To develop efficient sentiment analysis systems based on new features that can be used with traditional Machine Learning (Machine Learning) methods. Towards this objective, we aim to use pre-built resources such as sentiment lexicons and word embeddings to extract new features.

.

• To move sentiment analysis beyond sentence-level and polarity-based analysis. Towards this objective, we are interested in:

– Proposing ensemble systems that combine classical Machine Learning models with Deep Learning to analyse emotions expressed on Twitter.

– Developing a Deep Learning based model to solve the problem of multi-label emotions classification.

– Utilising Deep Learning based models to analyse opinions at the aspect level.

.

• To move sentiment analysis beyond a single language.

We aim to utilise the concept of transfer learning to develop a system that can transfer sentiment knowledge from high resources languages to low resources languages. The final goal of this objective is to obtain a universal sentiment analysis system that works with low resource languages and does not require machine translation.

.

• To test the developed systems on real cases of analysis

.

• The idea of this objective is to test our developed systems in real applications and different domains. Hence, towards this objective, we define the following sub-goals:

– To collect tweets of local people, visitors and official brand destination offices from different tourist destinations and analyse the opinions shared in these tweets.

– To combine the aspect-based sentiment analysis with multi-criteria decision aid systems to improve the decision-making process.

.

https://deim.urv.cat/~itaka/itaka2/PDF/acabats/PhD_Thesis/TESI_Mohammed_Jabreel.pdf

.

.