Advances in Information Retrieval: 31th European Conference by W. Bruce Croft (auth.), Mohand Boughanem, Catherine Berrut,

By W. Bruce Croft (auth.), Mohand Boughanem, Catherine Berrut, Josiane Mothe, Chantal Soule-Dupuy (eds.)

This booklet constitutes the refereed court cases of the thirtieth annual eu convention on info Retrieval learn, ECIR 2009, held in Toulouse, France in April 2009.

The forty two revised complete papers and 18 revised brief papers offered including the abstracts of three invited lectures and 25 poster papers have been conscientiously reviewed and chosen from 188 submissions. The papers are equipped in topical sections on retrieval version, collaborative IR / filtering, studying, multimedia - metadata, professional seek - ads, evaluate, opinion detection, internet IR, illustration, clustering / categorization in addition to allotted IR.

Show description

Read or Download Advances in Information Retrieval: 31th European Conference on IR Research, ECIR 2009, Toulouse, France, April 6-9, 2009. Proceedings PDF

Similar european books

Grand Designs and Visions of Unity: The Atlantic Powers and the Reorganization of Western Europe, 1955-1963

Within the overdue Fifties, opposed to the unfolding backdrop of the chilly conflict, American and eu leaders begun operating to reshape Western Europe. They sought to evolve the area to a altering global within which ecu empires have been swiftly disintegrating, Soviet impression used to be spreading, and the USA might not shoulder the complete political and financial burden of the West but hesitated to proportion it with Europe.

Computation and Comparison of Efficient Turbulence Models for Aeronautics — European Research Project ETMA

The computation of advanced turbulent flows by way of statistical modelling has already a protracted background. the preferred two-equation types at the present time have been brought within the early sev­ enties. despite the fact that those versions were usually proven in fairly educational situations. The develope­ ment of desktops has ended in progressively more acurate numerical equipment.

Additional info for Advances in Information Retrieval: 31th European Conference on IR Research, ECIR 2009, Toulouse, France, April 6-9, 2009. Proceedings

Sample text

1 [18]. , 9 out of 24 results are statistically significant, 22 out of 24 improvements are positive, and the highest improvement on MRR is over 25%, showing that risk adjustment can dramatically increase the chance of returning one relevant document close to the top of a ranked list. Our approach also outperforms the linear smoothing approach on MAP for all four collections. Comparing Table 3 (b) with Table 3 (a), we can see that our approach combined with the linear smoothing performs better than our approach without background smoothing in 18 out of 24 results, and better than the Dirichlet smoothing in 18 out of 24 results.

Therefore, our risk adjustment complements background smoothing in performance improvement. 6 Conclusion and Future Work Uncertainty is an intrinsic part of document ranking, but has not generally been considered in current IR models. Current models usually provide a scalar estimate of the mean of a document’s posterior probability distribution. However, the probability distribution is better described by both its mean and variance. e. some results may be very good while others may be very poor.

For each rank position, the objective is to select a document that has the maximum increase of the objective function. Notice that such a sequential update may not necessarily provide a global optimization solution, but it provides an excellent trade-off between accuracy and efficiency. , n}. The final expression is derived as k−1 Ok − Ok−1 = wk (E[rk ] − αwk σk 2 − 2α wi σi σk ρi,k ) i=1 (9) Mean-Variance Analysis: A New Document Ranking Theory in IR 11 Because wk is a constant for all documents, we can safely drop it when using the increase to rank documents.

Download PDF sample

Rated 4.37 of 5 – based on 50 votes