TY - GEN
T1 - Joint Local and Global Sequence Modeling in Temporal Correlation Networks for Trending Topic Detection
AU - Shu, Kai
AU - Li, Liangda
AU - Wang, Suhang
AU - Zhou, Yunhong
AU - Liu, Huan
N1 - Funding Information: This material is in part supported by the NSF award # 1614576. Publisher Copyright: © 2020 ACM.
PY - 2020/7/6
Y1 - 2020/7/6
N2 - Trending topics represent the topics that are becoming increasingly popular and attract a sudden spike in human attention. Trending topics are critical and useful in modern search engines, which can not only enhance user engagements but also improve user search experiences. Large volumes of user search queries over time are indicative aggregated user interests and thus provide rich information for detecting trending topics. The topics derived from query logs can be naturally treated as a temporal correlation network, suggesting both local and global trending signals. The local signals represent the trending/non-trending information within each frequency sequence, and the global correlation signals denote the relationships across frequency sequences. We hypothesize that integrating local and global signals can benefit trending topic detection. In an attempt to jointly exploit the complementary information of local and global signals in temporal correlation networks, we propose a novel framework, Local-Global Ranking (LGRank), to both capture local temporal sequence representation with adversarial learning and model global sequence correlations simultaneously for trending topic detection. The experimental results on real-world datasets from a commercial search engine demonstrate the effectiveness of LGRank on detecting trending topics.
AB - Trending topics represent the topics that are becoming increasingly popular and attract a sudden spike in human attention. Trending topics are critical and useful in modern search engines, which can not only enhance user engagements but also improve user search experiences. Large volumes of user search queries over time are indicative aggregated user interests and thus provide rich information for detecting trending topics. The topics derived from query logs can be naturally treated as a temporal correlation network, suggesting both local and global trending signals. The local signals represent the trending/non-trending information within each frequency sequence, and the global correlation signals denote the relationships across frequency sequences. We hypothesize that integrating local and global signals can benefit trending topic detection. In an attempt to jointly exploit the complementary information of local and global signals in temporal correlation networks, we propose a novel framework, Local-Global Ranking (LGRank), to both capture local temporal sequence representation with adversarial learning and model global sequence correlations simultaneously for trending topic detection. The experimental results on real-world datasets from a commercial search engine demonstrate the effectiveness of LGRank on detecting trending topics.
UR - http://www.scopus.com/inward/record.url?scp=85088401982&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85088401982&partnerID=8YFLogxK
U2 - 10.1145/3394231.3397924
DO - 10.1145/3394231.3397924
M3 - Conference contribution
T3 - WebSci 2020 - Proceedings of the 12th ACM Conference on Web Science
SP - 335
EP - 344
BT - WebSci 2020 - Proceedings of the 12th ACM Conference on Web Science
PB - Association for Computing Machinery, Inc
T2 - 12th ACM Conference on Web Science, WebSci 2020
Y2 - 6 July 2020 through 10 July 2020
ER -