1. Example: Academic Search
PaperID: 1770
Cross-domain Ranking via Latent Space Learning
Jie Tang *+ and Wendy Hall ‡
*Department of Computer Science and Technology, Tsinghua University †Tsinghua National Laboratory for Information Science and Technology (TNList)
‡Electronics and Computer Science, University of Southampton, UK
Cross-domain Ranking: learning to rank objects from multiple interrelated domains.
Also called heterogeneous cross-domain ranking, when the domains are different.
Problem Formulation
Example: Academic Search
Input: Given T domains, with each having a training dataset—i.e., ranking pairs {(xt1q, yt1q), …, (xtnq, ytnq)}, where
xt1q is K-dimensional feature vector to represent the relevance between an object in domain t and the query q at time t;
ytnq ∈{r1, …, rlt}represents the ranking level of the object w.r.t. the query q;
Learning: learn T ranking functions {ft}Tt=1 simultaneously, with each ft for predicting the rank level of unlabeled objects in the corresponding domain.
BayCDR: Cross-domain Ranking via Latent Space and Bayesian Learning
Basic model
The proposed model
Learning algorithm
Results
Accuracy performance
Scalability performance
Dataset:
LETOR 2.0 (Liu et al. 2007): a public dataset for learning to rank research. LETOR consists of three homogeneous sub datasets (i.e., TREC2003, TREC2004, and OHSUMED), with 50, 75, and 106 queries, respectively.
AMiner (Tang et al. 2008): The dataset contains 14,134 authors, 10,716 papers, and 1,434 conferences. Given a query, the goal is to find experts, top conferences, and authoritative papers for the query.
Jie Tang and Wendy Hall. Cross-domain Ranking via Latent Space Learning. In Proceedings of the 31st AAAI Conference on Artificial Intelligence (AAAI'17).