Reproduce “Taxonomy-guided Semantic Indexing for Academic Paper Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
SeongKu Kang, Yunyi Zhang, Pengcheng Jiang, Dongha Lee, Jiawei Han, Hwanjo Yu
The TaxoIndex framework enhances academic paper search by enabling dense retrievers to leverage a taxonomy-guided semantic index that captures both topic-level and phrase-level academic concepts. It consistently improves retrieval accuracy, particularly for complex queries, and demonstrates robust performance even with limited training data, while providing interpretability for search results.
Key insights
- Explicitly integrating structured domain knowledge, such as an academic taxonomy, through a multi-granularity index (topics and phrases) significantly improves the ability of dense retrievers to match specific academic concepts.
- An inference-based, plug-and-play add-on module can effectively guide a frozen backbone dense retriever by learning to infer and fuse academic concept representations, offering flexibility and parameter efficiency.
- The framework's effectiveness is robust to limited training data and partial taxonomy completeness, suggesting its practical applicability in specialized domains where resources might be scarce.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.