SemanticRec | Recommender Systems for Literature Discovery

A novel AI framework for personalized research assistance and academic literature discovery, published in Engineering Reports (Wiley).

SemanticRec is a research framework designed to enhance personalized research assistance and academic literature discovery. It addresses key limitations of existing recommender systems—such as reliance on keyword matching and a lack of contextual understanding—by integrating traditional similarity measures (cosine and Jaccard similarity) with contextual semantic embeddings from SciBERT. The system was evaluated on large-scale datasets including DBLP and ACM records, achieving significant improvements in retrieval performance, precision, and F1-scores.

Tech Stack

PythonPyTorchSciBERTNLPRecommender SystemsCosine / Jaccard Similarity

Features

Hybrid recommendation combining classical similarity measures with contextual embeddings
Semantic understanding of academic text using SciBERT
Evaluated on large-scale DBLP and ACM datasets for robustness and generalizability
Improved precision and F1-scores for academic search systems
Research published in Engineering Reports, Wiley (2026)

Challenges

Moving beyond keyword-based matching to capture contextual relevance.
Scaling evaluation across large-scale bibliographic datasets.
Balancing classical retrieval metrics with semantic quality.

Learnings

Deepened understanding of hybrid recommender system design.
Gained hands-on experience with transformer-based embeddings for retrieval.
Learned to rigorously evaluate retrieval systems using precision and F1 metrics.

Links

Live Github