Efficient Semantic Search for Detecting Redundant Student Project Proposals

Academic institutions face persistent challenges with redundant student project proposals, leading to duplicated efforts and diminished research originality. These challenges are exacerbated by limited visibility into prior submissions, varying terminologies, and the growing volume of archived proposals, which render manual review processes ineffective. To address these limitations, this paper introduces an automated Project Similarity Detection System leveraging advanced natural language processing. The solution combines Sentence-BERT for generating multilingual semantic embeddings of project proposals and Facebook AI Similarity Search for efficient high-dimensional similarity retrieval. A domain-optimized pipeline processes academic text through tokenization, stopword removal, and stemming before converting titles/descriptions into dense vectors. The system enables real-time comparison against institutional archives using cosine similarity, providing immediate feedback during proposal submission. Evaluations confirm the system's effectiveness in identifying semantically similar projects with high relevance and practical response times. This approach supports originality, enhances institutional knowledge reuse, and opens pathways for future expansion to broader academic text inputs and user-interactive refinement.