Scholar Inbox Datasets

As part of our submission to the ACL Systems Demo Track, we provide the following datasets:

Recommendations

Explicit user ratings

rated_papers.csv
We publish explicit user ratings, as submitted via the thumbs up/ down buttons on papers displayed on scholar-inbox.com.
$N=774k$

arxiv_id	user_id	rating	time

Implicit user interactions

implicit_interactions.csv
This is a dataset containing the user-paper interactions that were not explicit up/down ratings. Such interactions include reading a paper, or clicking on any of the paper buttons to collect, share, etc.
$N=556k$

arxiv_id	user_id	time

Abstract highlighting

abstract_sentences.csv
Scholar Inbox's sentence highlighting feature was built using user-collected sentence classification data.
We used this data to train out abstract sentence highlighting model.
This dataset contains 2538 samples and four sentence labels: problem, task, idea, result

arxiv_id	abstract	start_idx	end_idx	label