Efficient and Effective Methods for Conversational Search and Summarization

The primary objective of this thesis is to investigate and enhance the role of Natural Language Processing (NLP) and Large Language Models (LLMs) in Information Retrieval (IR), with a particular focus on Conversational Search systems. The work aims to improve the interaction between users and IR systems by developing methods for query rewriting, optimizing the retrieval process, and summarizing retrieved information. The first part of the thesis explores methods to enhance user queries by incorporating external knowledge sources and leveraging advanced Large Language Models. The objective is to mitigate limitations in query understanding by enriching user inputs with structured commonsense knowledge. Additionally, the research investigates the potential of instruction-tuned language models, ranging from large state-of-the-art systems to computationally efficient smaller models, to automatically reformulate queries and improve the effectiveness of conversational information retrieval. The final part of the thesis explores methods for summarizing retrieved content across different domains and modalities. Emphasis is placed on effectively condensing lengthy textual information, particularly in complex and specialized fields such as legal and regulatory contexts, by employing various NLP techniques and language models. Additionally, the research investigates approaches to multimodal summarization, proposing methods to generate meaningful summaries from visual data, including geospatial imagery. These efforts highlight the adaptability and wide applicability of NLP and LLM-based methods. In summary, the thesis introduces innovative approaches aimed at enhancing the intelligence, efficiency, and adaptability of Information Retrieval systems. It establishes a solid foundation for future research in low-resource NLP, domain-specific retrieval applications, and multimodal content summarization, facilitating interdisciplinary collaboration.