LXY12581/complex_web_questions

Dataset Card for Dataset Name

Dataset Summary

A dataset for answering complex questions that require reasoning over multiple web snippets

ComplexWebQuestions is a new dataset that contains a large set of complex questions in natural language, and can be used in multiple ways:

• By interacting with a search engine, which is the focus of our paper (Talmor and Berant, 2018);
• As a reading comprehension task: we release 12,725,989 web snippets that are relevant for the questions, and were collected during the development of our model;
• As a semantic parsing task: each question is paired with a SPARQL query that can be executed against Freebase to retrieve the answer.

Supported Tasks and Leaderboards

[More Information Needed]

Languages

• English

Dataset Structure

QUESTION FILES

The dataset contains 34,689 examples divided into 27,734 train, 3,480 dev, 3,475 test. each containing:

"ID”: The unique ID of the example; 
"webqsp_ID": The original WebQuestionsSP ID from which the question was constructed; 
"webqsp_question": The WebQuestionsSP Question from which the question was constructed; 
"machine_question": The artificial complex question, before paraphrasing; 
"question": The natural language complex question; 
"sparql": Freebase SPARQL query for the question. Note that the SPARQL was constructed for the machine question, the actual question after paraphrasing
may differ from the SPARQL. 
"compositionality_type": An estimation of the type of compositionally. {composition, conjunction, comparative, superlative}. The estimation has not been manually verified,
 the question after paraphrasing may differ from this estimation.
"answers": a list of answers each containing answer: the actual answer; answer_id: the Freebase answer id; aliases: freebase extracted aliases for the answer.
"created": creation time

NOTE: test set does not contain “answer” field. For test evaluation please send email to alontalmor@mail.tau.ac.il.

WEB SNIPPET FILES

The snippets files consist of 12,725,989 snippets each containing PLEASE DON”T USE CHROME WHEN DOWNLOADING THESE FROM DROPBOX (THE UNZIP COULD FAIL)

"question_ID”: the ID of related question, containing at least 3 instances of the same ID (full question, split1, split2); "question": The natural language complex question; "web_query": Query sent to the search engine. “split_source”: 'noisy supervision split' or ‘ptrnet split’, please train on examples containing “ptrnet split” when comparing to Split+Decomp from https://arxiv.org/abs/1807.09623 “split_type”: 'full_question' or ‘split_part1' or ‘split_part2’ please use ‘composition_answer’ in question of type composition and split_type: “split_part1” when training a reading comprehension model on splits as in Split+Decomp from https://arxiv.org/abs/1807.09623 (in the rest of the cases use the original answer). "web_snippets": ~100 web snippets per query. Each snippet includes Title,Snippet. They are ordered according to Google results.

With a total of 10,035,571 training set snippets 1,350,950 dev set snippets 1,339,468 test set snippets

Source Data

The original files can be found at this dropbox link

Licensing Information

Not specified

Citation Information

@inproceedings{talmor2018web,
  title={The Web as a Knowledge-Base for Answering Complex Questions},
  author={Talmor, Alon and Berant, Jonathan},
  booktitle={Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers)},
  pages={641--651},
  year={2018}
}

Contributions

Thanks for yuancu for contributing this dataset.