24 January 2013
Bias-Variance Analysis in Information Retrieval
12:30pm - 13:45pm
, Peng Zhang - Visiting Student from Robert Gordon University, Aberdeen
The estimation of query language model is an important task in information retrieval (IR). The ideal estimation is expected to be not only effective in terms of mean retrieval performance over all queries, but also stable in the sense that the performance is stable for different individual queries. In practice, however, improving effectiveness could sacrifice the stability, and vice versa. In this talk, I will present how to investigate this trade-off from a new perspective, i.e., the bias-variance trade-off, which is a fundamental theory in statistical estimation. Our analysis includes a number of factors (e.g., query model complexity, query model combination, document weight smoothness and irrelevant documents removal) that can influence the query language modelling. The bias-variance analysis will potentially form an analysis framework and a novel evaluation strategy for IR, especially for the query language modelling.
Mr. Peng Zhang is now a visiting research student at OU, and a PhD student at Robert Gordon University, Aberdeen, UK. His PhD research is to build theoretical frameworks/models for Information Retrieval (IR), in both classical and quantum perspectives. His thesis title is approximating the true relevance model in relevance feedback. He has published respected journal papers in ACM Transactions on Asian Language Processing, IEEE Transactions on Neural Networks and IEEE Transactions on Knowledge and Data Engineering, as well as top conference papers in SIGIR 2009 and ACL 2008, 2010. He won the Best Poster Award at ECIR’2011 based on a quantum-inspired ranking approach. He was the local co-Chair at QI’2011 and the proceedings chair at AIRS’2012.
Save to your Calendar