Domain-Specific Datasets

[DelaramRajaei]

Hey @michelecatani
This is the issue where you can log your findings about datasets in specific domains like law and medicine, which include query and relevant document pairs.

Additionally, please check the availability of temporal data and user-specific information for each query within these datasets.

[michelecatani]

Thanks Delaram for creating the issue!

I have found the following two options for medical:

NFCorpus

Link: https://www.cl.uni-heidelberg.de/statnlpgroup/nfcorpus/

This dataset (downloadable from the link) contains queries and relevant documents, but does not seem to contain any sort of time/user information. It's free to use for academic purposes and relatively small.

TREC Biomedical Tracks

Link: https://www.trec-cds.org

There are quite a few datasets that you can go into on that link. Most datasets in this link contain topics, qrels, documents, and some time based data. As far as user data, there are actors associated with the studies but I'm not sure if that's what you were looking for. Some datasets appear not to contain qrels.

For law, I have found the following:

CLEF-IP 2009 Test Collection

Link: https://researchdata.tuwien.at/records/a2svx-p1y38

This is specifically in the Intellectual Property domain. It contains topics, qrels, and documents. The patents contain several different dates, so there is temporal information.

[michelecatani]

@DelaramRajaei @hosseinfani I do have a few questions though:

1. By temporal data, do you mean times referenced in the documents? Or times related to the queries themselves?

2. What exactly are you looking for when you ask for user data? Ie, who is the user in this situation?

[hosseinfani]

Hi @michelecatani

1. the time that the query has been submitted. we argue that the same query but submitted in different time convert different semantic, like if you put coffee in the morning means you are looking for coffee shop but in the evening may mean you're looking for different coffee makers

2. the user who has submitted the query. we argue that user matters when interpreting the query. For now, userid suffice for our work but later we can add more info about the user like the user's interests or etc.

[michelecatani]

Hi @hosseinfani thanks for the reply and clarification! I will look for some different datasets then, as the aforementioned don't seem to contain anything like you're describing.

[michelecatani]

As of yet, I have been unable to find any domain specific datasets that contain user-id or temporal data for the queries. I have been able to find a dataset that contains user-id for the queries, but it is not domain specific.

Link: https://trec.nist.gov/data/session2014.html

This dataset also contains temporal information, but it seems to be transformed. The queries are given a "starttime" that starts from 0.

I will continue looking for domain specific datasets with user id and temporal information