In the age of information overload, understanding and synthesizing a rich amount of knowledge from text becomes increasingly challenging. This thesis focuses on how we can apply natural language processing (NLP) techniques, especially those powered by large language models (LLMs), to reduce the human effort dealing with extensive corpora. The research directions are: (1) understanding human decision-making through text analysis, (2) decision-focused summarization, and (3) study organization for literature review. First, we develop a novel method to predict physician fatigue using physician note characteristics, with an LLM-based note unpredictability score as the most crucial feature. This method not only provides insights into the potential impact of fatigue on doctor decision-making but also raises important concerns about the integration of AI tools in healthcare settings. Second, we present decision-focused summarization (DecSum) approach which produces summaries to support human decision-making, using restaurant future rating prediction task as a testbed. We demonstrate the effectiveness of DecSum in supporting more accurate decision-making in a human study. Lastly, we propose a hierarchical category generation pipeline using LLMs to facilitate the literature review process. Using the generated hierarchies on selected research topics, including human expert annotations on a subset, we curate a study organization dataset as a valuable resource for future research on assistive tools for literature review. Through these investigations, this thesis shows how LLM-empowered NLP techniques can enhance our understanding of human decision-making processes and provide effective support for information synthesis and decision-making in various domains.