Context ICLR (International Conference on Learning Representations) is a premier machine learning conference. Unlike the other two flagship machine learning conferences ICML and NIPS, ICLR chooses a single-blind public review process in which the reviews and their rebuttals are both carried out transparently and in the open. This dataset was created by crawling the public [ICLR 2017 paper review site][1]. It seems ICLR is going double-blind from 2018, so my guess is that authors will remain anonymous during the review process. So, this dataset is unique because it captures a public academic review process with academic affiliations and all paper decisions including rejections. Content The dataset consists of two CSV files: - **iclr2017_papers.csv**: This file has a row per submission. It includes the paper title, authors, author conflicts, abstracts, tl;dr (a simplified abstract), and final decision (Accept/Oral, Accept/Poster, Accept/InviteToWorkshop, Reject). Each row has a unique identifier key called the "paper_id." - **iclr2017_conversations.csv**: This file has a row per textual review, rebuttal, or comment. It is related to the previous papers dataset using the secondary key "paper_id." All rows talking about a single paper share the same "paper_id." The conversations associated with each paper can be thought of as a forest. Each tree in the forest begins with a review followed by rebuttals and further comments/conversation. Each such textual entry composed by an individual is listed in its own row. The nodes of the tree are connected using the fields "child_id" and "parent_id" which can be used to construct the entire conversation hierarchy. Acknowledgements All rights for abstracts rest with the paper authors. Reproduction of abstracts here is solely for purposes of research. Thanks to the authors of Beautiful Soup 4 Python package which considerably simplified the process of curating this dataset. Inspiration This dataset was created to understand gender disparities in paper submissions and acceptances. Annotating each author with a binary gender is a pending task. The dataset can also be used to model communication processes employed in negotiation, persuasion, and decision-making. Another use of this dataset could be in modeling and understanding textual time-series data. [1]: https://openreview.net/group?id=ICLR.cc/2017/conference