Integrating subjectivity into relevance measurement for conversational information retrieval

In today's fast-changing world, people want better and more advanced ways to search for information. Instead of traditional web searches, users are moving toward conversational interfaces that allow natural interactions. To address this need, Conversational Information Retrieval (CIR) was created to provide accurate and relevant results using Natural Language Processing to simulate the natural search style that humans perform. Large language models like GPT-3 and LLaMA-3 have made significant improvements in understanding and producing human-like text. However, they often produce outdated or general responses rather than ones that are relevant to a given situation. This problem becomes especially noticeable in domains where users ask open-ended questions to express their preferences, emotions, and opinions, such as when recommending books. As a result, CIR systems must be able to process both the explicit content of questions and their underlying characteristics and sentiments. Aspect-Based Sentiment Analysis (ABSA) is a key method in this context. ABSA helps extract specific aspects from user queries and determines the sentiment associated with each aspect. This detailed analysis is especially important in situations where user preferences and feelings regarding certain features need to be examined. This study proposes a way to improve CIR systems for book recommendations by adding subjectivity to the search for relevant answers. The project involves extracting aspects and sentiments from user queries and retrieving relevant books using a hybrid retrieval method that combines semantic and full-text search. By considering both semantic similarity and subjective elements of queries, the system aims to standardize document retrieval and center the process on the needs of the person seeking help rather than on the subjective opinions of others.