Yelp Dataset Challenge Round 5 winning team
Posted: 21st January 2016
Department academics Nando De Freitas and Misha Denil were part of the winning team in the fifth round of the Yelp Dataset Challenge, for their paper ‘From Group to Individual Labels Using Deep Features’.
The Yelp dataset includes information about local businesses in 10 cities across 4 countries. The challenge awards $5,000 to research that uses this data in an innovative way and breaks new ground in research.
This paper proposes a novel approach to using group-level labels (e.g. the category of an entire review) to learn instance-level classification (e.g. the category of specific sentences inside this review). The authors designed a new objective (cost) function for training a model which uses features from a deep-learning convolutional neural network. This trained neural network can, in turn, be used as a classifier predicting which category a specific instance belongs to. Their innovative research has broad implications for a variety of fields, and not just text classification.
This entry was selected from many submissions for its technical and academic merit. A PDF copy of the winning paper can be found here http://goo.gl/n2Mzxz