The Data Science Discipline of the School of EECS is hosting the following guest seminar:
Title: Retrieval Plans for Search Engines
Speaker: Professor Craig Macdonald (University of Glasgow)
Host: IELab group
Abstract: There are lots of different configurations for a search engine to execute a query. For instance, the user may decide to add a phrase, or the configuration may mandate proximity boost or pseudo-relevance feedback as part of the retrieval pipeline. Not all configurations work well for all queries, and hence there has been substantial work over the years towards not just developing retrieval pipelines, but also in identifying the best retrieval pipelines for a given query. In this presentation, Professor Macdonald will give an overview of past work on selective search engines approaches (changing configurations on a per-query basis); will further argue for why it's necessary to have a language to express retrieval pipelines, as surfaced in PyTerrier, and will describe the advantages this portrays for researchers, from reproducibility and verification, to pipeline optimisation and experiment execution plans.
Abstract: There are lots of different configurations for a search engine to execute a query. For instance, the user may decide to add a phrase, or the configuration may mandate proximity boost or pseudo-relevance feedback as part of the retrieval pipeline. Not all configurations work well for all queries, and hence there has been substantial work over the years towards not just developing retrieval pipelines, but also in identifying the best retrieval pipelines for a given query. In this presentation, Professor Macdonald will give an overview of past work on selective search engines approaches (changing configurations on a per-query basis); will further argue for why it's necessary to have a language to express retrieval pipelines, as surfaced in PyTerrier, and will describe the advantages this portrays for researchers, from reproducibility and verification, to pipeline optimisation and experiment execution plans.
Bio: Craig Macdonald a Professor of Information Retrieval in the Information Retrieval Group of the Information & Data Analytics Section at the University of Glasgow. His research interests are information retrieval and recommender systems, focusing on efficient yet effective retrieval, particularly encapsulating machine learning or deep learning. His team's research is published in top-tier conference venues such as ACM SIGIR, ECIR, CIKM, RecSys, WWW and WSDM, as well as in journals such as ACM Transactions on Information Systems, Information Processing & Management and the Information Retrieval Journal.
Researchers and students from all disciplines are welcome.
Researchers and students from all disciplines are welcome.
About Data Science Seminar
This seminar series is hosted by EECS Data Science.
Venue
GP-South room 343 (78-343)