PKUICST at TREC 2017 Real-Time Summarization Track: Push Notifications and Email Digest

Jizhi Tang, Chao Lv, Lili Yao, Dongyan Zhao
2017 Text Retrieval Conference  
In this paper, we describe our approaches and corresponding results in the Real-Time Summarization(RTS) track at the 2017 Text Retrieval Conference(TREC). The main idea is to build a two-stage filter system for both scenario A and B. In the first stage, tweets are filtered according to its relevance score to a particular topic, while in the second stage, they are filtered according to its novelty score to previous submitted tweets. We tried several approaches to model the text similarity, such
more » ... s negative KL-divergence and cosine distance, as well as blending models. Especially, in scenario A, the push notification scenario, we designed a decoupled system that can maintain high availability in order to meet the real-time requirements. The experiment results show that our methods reach good performance with respect to several metrics such as EG-p and nDCG-p.
dblp:conf/trec/TangLYZ17 fatcat:x565fygblbdvnfnzvlzsnzn7xm