skonan/PaperLens-arXiv-embeddings
收藏资源简介:
PaperLens arXiv per_venue检索索引数据集是一个用于PaperLens arxiv搜索和重排部署的索引包和p_accept缓存种子。它覆盖了来自LitSearch重排实验的约80,000篇per_venue arXiv语料库。数据集包含以下组件:Lucene BM25索引(基于pyserini)、Qwen3-Embedding-0.6B密集FAISS(Flat)索引、语料库文件(用于将FAISS位置解析为arxiv_id),以及预计算的p_accept值(分别用于3B和7B重排器,覆盖约28,664篇论文)。这些资源支持检索系统的快速部署和高效运行,但排除了仅用于重建的嵌入内存映射和完整元数据。
PaperLens arXiv per_venue retrieval indexes is an index bundle and p_accept cache seed for the PaperLens arxiv search/rerank deployment. It covers the ~80K per_venue arXiv corpus from the LitSearch reranking experiment. The dataset includes: a Lucene BM25 index (pyserini) over the per_venue corpus, a Qwen3-Embedding-0.6B dense FAISS (Flat) index, the per_venue corpus file (resolves FAISS position to arxiv_id), and precomputed p_accept for the 3B and 7B rerankers (covering ~28,664 papers). These components facilitate retrieval and reranking operations, excluding rebuild-only embedding memmap and full metadata.




