The 34th European Conference on Information Retrieval (ECIR) was held in Barcelona, Spain in the pre-Easter week. A strong conference program was complemented by breathtaking architectural scenery, good food and one of Europe’s best ranges of night life activities (well, and rain 😉 ). The conference was kicked-off by a great keynote on “Studying Network Structures for IR: the Impact of Size” in which Paolo Boldi elaborated on the pitfalls of large-scale network analysis and why the betweenness measure is not the last word in the field. My personal highlights from the accepted full paper presentations include:
- Edward Bortnikov et al. – Modeling Transactional Queries via Templates
The authors categorize transactional queries by use case, e.g., booking a flight, and associate the resulting query profiles with a range of generic patterns for each use case. Subsequent processing steps are envisioned to exploit this information in the form of dedicated support for the use case at hand.
- Abdigani Diriye et al. – Interactive Search Support for Difficult Web Queries
The paper discribes an interactive search support system that predicts query quality and identifies potential weaknesses in the query formulation step. The searcher is interactively informed about ways of increasing the estimated likelihood of search success via query reformulations. The authors report usage statistics and show the beneficial effect of their system.
- Tamara Polajnar et al. – Detection of News Feeds Items Appropriate for Children
Based on a wide range of textual features, the authors propose a classification scheme that detects child-friendly and suitable news streams, taking into consideration notions such as readability and resource credibility.
- Morgan Harvey et al. – Comparing Tweets and Tags for URLs
Starting from the TREC 2011 Microblog Track dataset, the authors work towards increasing the tagging coverage of tweeted links. Collaborative bookmarking platforms such as del.icio.us showed very low coverages of ~4%. Using rigorous filtering and cleaning steps, high-quality tags can be generated from tweet contents.
- Jin Young Kim and Bruce Croft – A Field Relevance Model for Structured Document Retrieval
The authors propose a dynamic field weighting scheme for structured document retrieval. Based on corpus-wide term (co-)occurrence statistics, field weights are able to better capture the underlying distribution of relevance than would be possible given static weighting schemes.
- Rao Muhammad Adeel Nawab et al. – Retrieving Candidate Plagiarised Documents Using Query Expansion
The authors phrase the problem of identifying plagiarised documents as a retrieval task. Given this setting, they apply query expansion using thesauri and ontologies as a starting point in order to identify sophisticated examples of plagiarism that go beyond verbatim copies.
- Stephen Robertson – On Smoothing Average Precision
IR document test collections are often considered as being sampled from a global population of documents. Current performance metrics, however, do not appropriately account for this fact. The author introduces a smoothed version of average precision that attempts to remedy the statistical flaws of applying raw metrics on sampled collections.
- Shima Gerani et al. – Score Transformation in Linear Combination for Multi-criteria Relevance Ranking
Retrieval model fusion approaches typically rely on weighted linear combinations of constituent scores or ranks in order to reach a global ranking. In order to enable such score combinations, comparability of scores has to be ensured. The authors propose two non-linear score transformation methods that can replace fixed component weights in the aggregation process.
- Guido Zuccon et al. – Top-k Retrieval Using Facility Location Analysis (Best Paper)
The authors draw an analogy between the economical theory of facility location analysis and search engine result diversification. Based on this metaphor, a greedy local search algorithm is used in order to explore potential result set compositions.
- Sadiye Alici et al. – Adaptive Time-to-Live Strategies for Query Result Caching in Web Search Engines
Large scale search engines invest significant consideration into their caching mechanisms in order to cope with peek request loads. The authors propose a dynamic approach for determining the staleness of cached result sets taking into consideration the current back-end load.
Next stop: ECIR 2013, Moscow, Russia