|Xiaomu Questioning||237.7 MB||Bandit Learning, Implicit Feedback||data||NeurIPS'18|
Xiaomu Questioning is an innovative feature which actively and adaptively displays banner containing questions in lecture videos, aiming at improving learners' engagement.
Learner is allowed to click on the displayed banner so as to see answer to the question.
Random selection strategy on positions and questions are used and the data of displayed question/clicks/feature/candidate items are collected.
The data is in .pkl form and must be deserialized by the package "pickle" of python3.
> data = pickle.load(open(“rawData.pkl”, ‘rb’))
The data is collected from Xiaomu Questioning, consisting of 7 columns.
The description of each column:
1st column: vidx - integer, the ID of video;
2nd column: aid - integer, the ID of recommended question;
3rd column: reward - integer, the observed feedback(1 for click, 0 for non-click)
4th column: candidate list of arms - list of integer, the available candidates at this decision point;
5th column: feature of relevance - two dimensional numpy array, each row representing the corresponding arm in the candidate list;
6th column: feature of examination - two dimensional numpy array, each row representing the corresponding arm in the candidate list;
7th column: position of the recommended arm - integer;