HAMNET-AI / PDFTriage

Reproduction paper --- PDFTriage : Question Answering over Long, Structured Documents
MIT License
40 stars 3 forks source link

问题:论文第二页主要讲什么 效果很差 #2

Open Leizhenpeng opened 10 months ago

Leizhenpeng commented 10 months ago

请输入查询:What is the second page of the paper mainly about? 查询问题: What is the second page of the paper mainly about? 0 $.data[1].boxes[0].text // boxes 应该是范围 The second page of the paper is mainly about related work. 请输入查询:What is the third page of the paper mainly about? 查询问题: What is the third page of the paper mainly about? 2 // 应该是0 Fetching figure 请输入查询:What is the third page of the paper mainly about? 查询问题: What is the third page of the paper mainly about? 2 // 应该是0 / Fetching figure 请输入查询:

 query_prompt = f"What contents to the number of pages mentioned in this question : {query}"
    path = query_engine.query(query_prompt).metadata['json_path_response_str'].replace("&&", "&")
    print(path)

JSONQueryEngine效果并不好,最后查询的方式最好手动实现函数查询,不应该talk to json

Aeemforst commented 10 months ago

查询的是手动查询的,JSONQueryEngine是通过问题和josn的格式获取josn的查询路径

Aeemforst commented 10 months ago

请输入查询:What is the second page of the paper mainly about? 查询问题: What is the second page of the paper mainly about? 0 $.data[1].boxes[0].text // boxes 应该是范围 The second page of the paper is mainly about related work. 请输入查询:What is the third page of the paper mainly about? 查询问题: What is the third page of the paper mainly about? 2 // 应该是0 Fetching figure 请输入查询:What is the third page of the paper mainly about? 查询问题: What is the third page of the paper mainly about? 2 // 应该是0 / Fetching figure 请输入查询:

 query_prompt = f"What contents to the number of pages mentioned in this question : {query}"
    path = query_engine.query(query_prompt).metadata['json_path_response_str'].replace("&&", "&")
    print(path)

JSONQueryEngine效果并不好,最后查询的方式最好手动实现函数查询,不应该talk to json

我这边测试问题的查询结果没问题呀