bio-protocol/bio-faiss-microbiome-v1

Bio-FAISS微生物组索引v1是一个用于科学检索的FAISS索引和元数据集。它包含科学论文的文本数据，适用于文本检索和问题回答任务。数据集按照章节、段落进行分层切片，每个切片约为380个token，有约15%的重叠。使用bio-protocol的科学检索器作为嵌入器，通过内积计算余弦相似度。数据集的索引类型为IndexFlatIP。该数据集包含的字段有：切片ID、切片哈希值、文本、论文ID、标题、章节、小节、段落索引、关键词和提升值。

Bio-FAISS Microbiome Index v1 is a FAISS index and metadata set for scientific retrieval. It contains text data from scientific papers, suitable for text retrieval and question answering tasks. The dataset is chunked hierarchically by sections and paragraphs into ~380-token chunks with ~15% overlap. The bio-protocol scientific retriever is used as the embedder, computing cosine similarity via inner product. The index type of the dataset is IndexFlatIP. The fields included in the dataset are: chunk_id, chunk_hash, text, paper_id, title, section, subsection, paragraph_index, keywords, and boost.