📝 Abstract:
Assigning suitable academic reviewers to research projects is a complex task that requires understanding both the semantic content of project descriptions and the diverse research backgrounds of academicians. Manual reviewer assignment or simple keyword-based methods often fail to capture these relationships effectively. This thesis proposes a semantic, data-driven framework that integrates embedding-based retrieval, clustering, and large language model (LLM) based reasoning to improve project–academician matching. Academic profiles consisting of publications, supervised theses, and research keywords, along with target project descriptions, are encoded using a pretrained embedding model to produce dense semantic representations. The resulting embeddings are then grouped using clustering techniques to organize academicians according to their primary research domains. Given a target project, candidate reviewers are retrieved from relevant clusters, enabling the selection of reviewers across different research areas. This approach is particularly beneficial for interdisciplinary projects, ensuring that different domains are both adequately represented. The top candidates are subsequently provided to an LLM, which reranks them according to their suitability and generates explicit reasoning to justify each assignment decision. The framework is evaluated through an ablation study analyzing the effects of different textual feature combinations, embedding model selection, and LLM choices on matching performance. Overall, the proposed approach aims to enhance reviewer assignment accuracy and interpretability by combining quantitative semantic similarity with qualitative LLM-driven reasoning.