The Glasgow Information Retrieval Group within the School of Computing Science at the University of Glasgow was founded 32 years ago in 1986 by Professor C. J. ‘Keith’ van Rijsbergen, often considered one of the founders of modern Information Retrieval (IR). From its outset, the Glasgow IR group has focused on improving the effectiveness of IR systems, inventing new logic & probabilistic retrieval models in the 90's and early 2000's, followed by the development of adaptive query expansion techniques, interactive multimedia models, the Divergence From Randomness framework, as well as leading research into quantum, expertise search and search result diversification models in the late 2000's. Since then, the Information Retrieval group embraced emerging machine learning and deep learning technologies for very large corpora and data streams, and have been at the forefront of research, development and application of those technologies for search and recommendation use-cases in a manner that ensures both effectiveness and efficiency.
The Glasgow IR Group has a strong research track record. Indeed, the ACM Digital Library shows that the group is ranked first by number of papers (429) at the SIGIR conference (the top CORE A* conference in the IR field). Meanwhile, a recent study by Microsoft Research of the 40 years of SIGIR showed the University of Glasgow as the 5th most cited university at the conference and the 1st in Europe. The group is also renowned for developing the popular open source IR platform, Terrier.org, which has been downloaded over 60,000 since its first release in 2004 and is cited by over 3500 research papers. Furthermore, the group has a long history of engagement with the public and industry sectors from small SMEs to multinational corporations.
The Informer magazine of BCS's Information Retrieval Specialist Group carried a recent profile on the Glasgow Information Retrieval Group.
As the most active Information Retrieval group by publications in Europe and one of the longest running, our research covers the full-spectrum of topics that are relevant to the development of IR systems:
IR & Recommender Systems Models
- Theoretical modelling of IR systems
- Machine learning and deep learning for information retrieval and recommender systems
- Interactive information retrieval (personalised IR, emotion based search, user modelling for IR, gestural IR)
- User modelling and personal information access
- Topic modeling; Entity search; Natural language processing for IR
- Recommender systems; Context-aware venue suggestion
Large-scale IR & Efficient IR
- Web information retrieval; Big data and information retrieval
- Efficient architecture for large-scale IR systems; Data stream processing architectures
Data Streams & IR
- Real-time information retrieval
- Search in social and sensor networks
Artificial Intelligence & IR
- Conversational information seeking and dialogue systems
- Information credibility, transparency, explainability and verification in IR systems
- Fairness in information retrieval & recommender systems
Natural Language Processing & IR
- Information extraction including entity and relation extraction
- Automatic knowledge graph construction
- Multi-task models, joint models and summarization
- Multimedia information retrieval
- Domain-specific information retrieval: smart cities; health; news; eDiscovery; sensitivity review
- Emergency management and crisis informatics
- Politics and Media
- Test collections and evaluation metrics
- Evaluation of IR systems and crowdsourcing for IR
- Online and Offline Evaluation of IR and Recommender Systems
- Eye-tracking and physiological approaches, such as fMRI
Current staff and students
Current Research Assistants and Research Students:
- Ting Su
- Xi Wang
- Xiao Wang
- Siwei Lu
- JingMin Huang
- Yashon Wu
- Xin Xin
- Carlos Gemmel
- Federico Rossetto
- Sarawoot Kongyoung
- Alexander Hepburn
- Ian Mackie
- Jun Choi Hyun
- Hitarth Narvala
- Jarana Manotumruksa (2019), University College London, Researcher
- Anjie Fang (2019), Amazon, Applied Scientist
- Jorge David Gonzalez Paule (2019), Jobandtalent Espana, Data Scientist
- Colin Wilkie (2019), Siemens, Data Engineer
- David Maxwell (2019), University of Deft, Data Engineer
- Graham McDonald (2019), University of Glasgow, Lecturer
- James McMinn (2018), ScoopAnalytics, Co-Founder
- Stuart Mackie (2018), BiP Solutions/Strathclyde Uni, Data Scientist
- Horatiu Bota (2018), Prodsight, Data Scientist
- Jesus Alberto Rodriquez Perez (2018), University of Glasgow, Postdoctoral Researcher
- Fajie Yuan (2018), Tencent, Senior Researcher
- Ryen White (Research Manager, Microsoft Research AI)
- Mark Sanderson (Professor, Royal Melbourne Institute of Technology)
- Mounia Lalmas (Head of Tech Research, Spotify)
- Ian Ruthven (Professor, Strathclyde University)
- Fabio Crestani (Professor, University of Lugano)
- Vassilis Plachouras (Software Engineering, Facebook)
- Leif Azzopardi (Chancellor's Fellow, Strathclyde University)
- Rodrygo Santos (Assistant Professor, Federal University of Minas Gerais)
- Eugene Kharitonov (Research Engineer, Facebook)
- Saul Vargas (Senior Machine Learning Scientist, ASOS)
- Dyaa Albakour (Lead Data Scientist, Signal Media)
- Nut Limsopatham (Senior Researcher, Microsoft AI)
- Amir Jadidinejad (AI Engineer, Glaxo Smith Kline)
- Zaiqiao Meng (Researcher, Cambridge University)
Terrier IR platform
Terrier is a highly flexible, efficient, and effective open source search engine, readily deployable on large-scale collections of documents developed by the IR group. Terrier implements state-of-the-art indexing and retrieval functionalities, and provides an ideal platform for the rapid development and evaluation of large-scale retrieval applications. Indeed, Terrier is used internationally, with over 60,000 downloads since its first release in 2004. Terrier is is used widely by the research community, with over 3700 citations in research papers according to Google Scholar.
Visit the website at http://terrier.org to learn more and download Terrier for free.
For those new to the Information Retrieval field, the group maintains a useful set of common resources for researchers and practitioners:
- Information Retrieval Test Collections: On this page are a list of publically available IR test collections. Some are held locally and some are pointers to remote sites.
- Collections of text and corpora: What's the difference between a test collection and a text collection? Well a test collection has to have associated queries and relevance judgements. The things in here are simply document collections.
- Language reference works: This page contains links to online language reference works, such as dictionaries, thesauri etc.
- IR systems: A list of links to some sites that have information about IR systems.
- Linguistic utilities: Bits of IR language related utilities like stemmers, stop words lists, morphological taggers, etc.
- IR Journals: Various table of contents and abstracts of the papers in a number of well known IR journals.
- IR Organisations: Various IR groups and more formal organisations.
- Books: Supplements of books or whole books online.
Group: Information Retrieval (IR)
Speaker: Omar Khattab, Stanford University
Date: 30 November, 2020
Time: 16:00 - 17:00
Deep pre-trained language models (LMs) have become central to many recent neural IR models, greatly improving effectiveness but also inflating computational cost. In practice, such computation cost imposes high latency and confines these models to re-ranking the output of term-based retrieval. In this talk, I will discuss approaches for tackling these problems and present ColBERT, a novel ranking model that adapts deep LMs for efficient end-to-end retrieval. ColBERT learns to encode queries and documents independently at a fine granularity and introduces a scalable late-interaction architecture that compares these granular representations for scoring. Doing so enables accurate retrieval from millions of documents and achieves competitive effectiveness with expensive re-rankers in 10–100s of milliseconds instead of multi-second latencies, while markedly boosting recall. Building on this work, I will discuss how supervising an end-to-end retriever differs from training a re-ranker and present ColBERT-QA, an open-domain question answering system that uses only weak supervision to train a task-specific retriever to state-of-the-art quality. I will conclude with thoughts on some open problems at the intersection of IR and NLP.
Omar Khattab is a Ph.D. student at Stanford University, working with Matei Zaharia and Chris Potts. He is interested broadly in Natural Language Understanding at scale, where systems capable of retrieval and multi-hop reasoning can leverage massive text corpora to make knowledgeable predictions. His current projects tackle the tasks of passage retrieval, open-domain question answering, and claim verification. Before joining Stanford, Omar earned his B.S. degree in Computer Science from Carnegie Mellon University in Qatar, where he worked with Mohammad Hammoud on large-scale data analytics.
IR Group in a nutshell
- #1 Information Retrieval group in Europe (ACM SIGIR publications)
- Creator of world-famous Terrier.org IR platform
- Leader in next generation Big Data processing technologies
- Leading international data challenges (TREC CARS, TREC Incident Streams)
- Driving innovate intelligent systems for the home, public and commercial sectors