Generating appropriate answers in dialogue systems using Large Language Models

نوع: Type: Thesis

مقطع: Segment: masters

عنوان: Title: Generating appropriate answers in dialogue systems using Large Language Models

ارائه دهنده: Provider: shaghayegh varasteh kazemi

اساتید راهنما: Supervisors: Muharram Mansoorizadeh and Hassan Khotanlou

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: yousef sanati and reza mohammadi

زمان و تاریخ ارائه: Time and date of presentation: 2026

مکان ارائه: Place of presentation: seminar

چکیده: Abstract: Conversational search is an important area of information retrieval that aims to support users through multi-turn interactions and to progressively understand their information needs. In such systems, user utterances often involve language dependencies, implicit references, and varying levels of complexity, making their representation as a single standalone query not always effective. Therefore, approaches based on generating multiple queries have received considerable attention as a means of covering different aspects of the user’s information need. However, in many existing methods, the number of generated queries is fixed, which may result in unnecessary processing and increased computational costs for simple utterances. In this study, an adaptive approach is proposed to determine the optimal number of multi-aspect queries in conversational search. In the proposed method, a set of linguistic features is extracted from each input utterance to estimate its level of complexity. Based on this estimation, an appropriate number of queries is then determined within a predefined range, and the corresponding queries are generated using a large language model. Each generated query is independently submitted to the retrieval process, and the resulting outputs are merged into a single final list. The evaluation results demonstrate that dynamically adjusting the number of queries can improve system efficiency. Specifically, processing time is reduced for simple utterances, while retrieval quality is maintained for complex utterances. These findings indicate that considering the linguistic complexity of user utterances plays an important role in designing more efficient conversational search systems and can provide a foundation for the development of intelligent systems for real-world applications.