YoGosling is an automatic push notification and daily digest system that monitors the Twitter stream in real-time and timely delivers interesting contents to social media users. Interests of users are expressed in the form of what we call as "interest profiles," which are hierarchical representation of user queries in natural human language. Built on the Anserini project, a Lucene-based search framework at the University of Waterloo, YoGosling served as a track-wide baseline system for the Text Retrieval Conference (TREC) 2016 Real-Time Summarization (RTS) track. YoGoslingLMGTFY, rooted on YoGosling, leverages external sources of evidence to support a finer-grained push notification scenario. YoGoslingLMGTFY participated in the 2016 TREC RTS track as an individual system run. For more details about YoGosling, please refer to this paper at section 4, the track guideline or the track mailing list.
In this project, we implemented an interleaving strategy for retrospective summarization and prospective notification tasks and validated the technique in comparison to a set of TREC 2014 & 2015 Microblog track evaluation metrics. As a pilot study, we also designed an A/B testing interface to investigate the interactiveness and effectiveness of interleaved evaluation. For more details, please refer to this paper.
In this project, we implemented three palmprint recognition algorithms including the state-of-art BLPOC (Phase-Based Correspondence Matching) from CVPR Biometrics Workshop 2014, NMRT (Neighbourhood Matching Radon Transform) and RLOC (Robust Line Orientation Code) in MATLAB. We proposed an improved reference point selection approach and improved the palmprint verification performance to 18.01% on right hand to left hand matching. We also evaluated the flipped palmprint matching performance with CASIA contactless, PolyU and IITD palmprint database.