SOTA Search Over Academic Publications
Blog post from Exa
Exa, a modern AI search engine, has launched a state-of-the-art search feature for academic publications, creating a dedicated index of approximately 350 million publications and 30 million authors. This new feature allows users to search scientific literature using natural language queries, even when the queries are vague or based on imperfect memories. Traditional academic searches rely on exact titles, authors, or keywords, but Exa's approach focuses on the meaning of queries, enabling retrieval of specific publications based on factual clues or incomplete recollections. Two benchmarks were developed to test Exa's effectiveness: known-item retrieval and tip-of-the-tongue retrieval, with Exa achieving an 82.8% success rate for queries. Exa outperformed other systems like Google Scholar in recall and mean reciprocal rank (MRR) while maintaining low latency. The search engine processes complex documents by converting PDFs into searchable text and integrating metadata such as authorship and citations, combining results from its web index and publication index to offer comprehensive coverage and structure.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.