Identifying genre of claims using deep learning techniques: a case study on code-mixed tweets

Abstract

The abstract of the thesis titled " Identifying Genre of Claims using Deep Learning Techniques: A Case Study on Code-Mixed Tweets " provides a concise overview of the research conducted and its findings. The thesis focuses on leveraging deep learning techniques to classify claims' genres within the context of code-mixed tweets. Code-mixing refers to the practice of using multiple languages within a single communication, which is more common on social media platforms in recent days. The research explores the challenges and opportunities presented by this unique linguistic context. The work begins by describing the frequency of code-mixing on social media, as well as the ramifications for natural language processing tasks. It tackles the paucity of tools and methodologies designed for code-mixed content, specifically claim genre classification. To fill this void, the thesis recommends the use of deep learning algorithms, which are known for their capacity to detect complicated patterns in language data. Through the implementation of a custom dataset of code-mixed tweets containing various types of claims, the research demonstrates the feasibility of using deep learning models for claim genre identification. Several state-of-the-art deep learning architectures are adapted and fine-tuned to the code-mixed context, including recurrent neural networks (RNNs), convolutional neural networks (CNNs), and transformer-based models. Performance metrics such as accuracy, precision, recall, and F1-score are used to evaluate the effectiveness of these models in accurately categorizing claim genres. The findings show that deep learning algorithms can produce promising results when it comes to categorizing claim genres in code-mixed tweets. Transformer-based models beat classical RNNs and CNNs in terms of contextual comprehension, demonstrating the importance of capturing global dependencies inside code-mixed information. However, data paucity, linguistic differences, and domain-specific language use are all emphasized as problems. The implications of this research extend to various domains, including social media analysis, sentiment analysis, and linguistic studies, where code-mixing is a prevalent phenomenon. The successful adaptation of deep learning techniques to this context signifies a step forward in the development of tools capable of understanding and processing code-mixed content more accurately. As code-mixing continues to evolve, the models presented in this thesis can serve as a foundation for further advancements in NLP.