sample_datatset
收藏资源简介:
该数据集旨在支持查询重写任务,包含三个主要部分:1)5个原始PDF文件;2)一个parquet文件,其中包含从上述5个PDF以及额外30个PDF中提取的文本块,每个文本块包括使用bge-m3模型生成的嵌入向量、元数据、文本内容和唯一ID;3)一个testdata JSON文件,定义了第一版查询重写任务所涵盖的所有测试场景,每个场景包含相关的查询问题、关联的文本块ID以及查询重写的预期行为。数据集的核心是parquet文件中的文本块及其嵌入表示,而JSON文件提供了针对性的任务定义和评估基准。
This dataset is designed to support query-rewriting tasks. It consists of three main parts: 1) 5 original PDF files; 2) a parquet file containing text chunks extracted from the aforementioned 5 PDFs and an additional 30 PDFs, with each chunk including an embedding vector generated using the bge-m3 model, metadata, text content, and a unique ID column; 3) a testdata JSON file that defines all test scenarios covered in the first version of the query-rewriting task, with each scenario containing relevant query questions, associated text chunk IDs, and expected behavior for query rewriting. The core of the dataset is the text chunks and their embeddings in the parquet file, while the JSON file provides targeted task definitions and evaluation benchmarks.




