Umair, Areeba (2023) Devising Artificial Intelligence Tools For Complex Data. [Tesi di dottorato]

Anteprima

Testo Umair_Areeba_36.pdf Download (9MB) | Anteprima

Abstract

Complex data, in the context of data science, refers to information with complex structures, such as high-dimensionality, mixed data types, temporal or spatial dependencies, graph formats, unstructured content, missing values, nonlinear relationships, hierarchical organization, or dynamic changes over time. Social media data is considered complex due to its diverse formats, high volume, unstructured nature, temporal dynamics, network structure, noise, missing data, sentiment, and contextual challenges. Social media dataset encompasses a range of information, including public reactions, opinions, news, and discussions. Analyzing this data provides insights into public sentiment, and thus also help to adopt strategies and campaigns according to people’s need. Various techniques have been employed for COVID-19 sentiment analysis including Lexicon-based methods, supervised machine learning models, Deep learning approaches, transformer-based models, Ensemble methods, and aspect-based analysis. In the first phase of this thesis, we used freely available X app (former Twitter) complex data and proposed BERT+NBSVM for classifying negative and positive tweets regarding COVID-19 vaccines, after applying necessary pre-processing steps. In second phase of thesis, we proposed sentiment analysis based recommender system for COVID-19 vaccines. For this purpose, we proposed an ensemble of random forest with CT-BERT_CONVLayerFusion model, for classifying the tweets into seven different categories of sentiments. We also utilized some of the Geo-Spatial approaches to geographically analyse the peoples sentiments. The proposed techniques have shown encouraging results from both a qualitative and quantitative point of view. All the results are published in reputed Journals and International Conferences.

Downloads

Actions (login required)

Modifica documento