2.2 Personalizing Asker Satisfaction Prediction
as many users try the service once and never come
We now describe our initial attempt at personalizing
back. However, for personalized satisfaction, at least
the ASP framework described above to each asker:
some prior history is needed. Therefore, in this early
• ASP Pers+Text: We first consider the naive per-
version of our work, we focus on users who have
sonalization approach where we train a separate
posted at least 2 questions - i.e., have the minimal
classifier for each user. That is, to predict a par-
history of at least one prior question. In the future,
ticular asker’s satisfaction with the provided an-
we plan to address the “cold start” problem of pre-
swers, we apply the individual classifier trained
dicting satisfaction of new users.
solely on the questions (and satisfaction labels)
Methods compared:
provided in the past by that user.
• ASP: A “one-size-fits-all” satisfaction predictor
• ASP Group: A more robust approach is to train a
that is trained on 10,000 randomly sampled ques-
classifier on the questions from the group of users
tions with only non-textual features (Section 2.1).
similar to each other. Our current grouping was
• ASP+Text: The ASP classifier with text features.
done simply by the number of questions posted,
• ASP Pers+Text and ASP Group: A personal-
essentially grouping users with similar levels of
“activity”. As we will show below, text features
ized classifiers described in Section 2.2.
only help for users with at least 20 previous ques-
Bạn đang xem 2. - TÀI LIỆU BÁO CÁO KHOA HỌC YOU’VE GOT ANSWERS TOWARDS PERSONALIZED MODELS FOR PREDICTING SUCCESS IN COMMUNITY QUESTION ANSWERING DOC
