Swarming to rank for information retrieval

verfasst von: Ernesto Diaz-Aviles, Wolfgang Nejdl, Lars Schmidt-Thieme
Abstract: This paper presents an approach to automatically optimize the retrieval quality of ranking functions. Taking a Swarm Intelligence perspective, we present a novel method, Swarm-Rank, which is well-founded in a Particle Swarm Optimization framework. SwarmRank learns a ranking function by optimizing the combination of various types of evidences such content and hyperlink features, while directly maximizing Mean Average Precision, a widely used evaluation measure in Information Retrieval. Experimental results on well-established Learning To Rank benchmark datasets show that our approach significantly outperformed standard approaches (i.e., BM25) that only use basic statistical information derived from documents collections, and is found to be competitive with Ranking SVM and RankBoost in the task of ranking relevant documents at the very top positions.
Organisationseinheit(en): Forschungszentrum L3S
Externe Organisation(en): Stiftung Universität Hildesheim
Typ: Aufsatz in Konferenzband
Seiten: 9-15
Anzahl der Seiten: 7
Publikationsdatum: 08.07.2009
Publikationsstatus: Veröffentlicht
Peer-reviewed: Ja
ASJC Scopus Sachgebiete: Theoretische Informatik und Mathematik, Theoretische Informatik
Elektronische Version(en): https://doi.org/10.1145/1569901.1569904 (Zugang: Geschlossen)

BibTeX

@inproceedings{5c73bffed7574d09b0b78c6274ec1c0b,
title = "Swarming to rank for information retrieval",
abstract = "This paper presents an approach to automatically optimize the retrieval quality of ranking functions. Taking a Swarm Intelligence perspective, we present a novel method, Swarm-Rank, which is well-founded in a Particle Swarm Optimization framework. SwarmRank learns a ranking function by optimizing the combination of various types of evidences such content and hyperlink features, while directly maximizing Mean Average Precision, a widely used evaluation measure in Information Retrieval. Experimental results on well-established Learning To Rank benchmark datasets show that our approach significantly outperformed standard approaches (i.e., BM25) that only use basic statistical information derived from documents collections, and is found to be competitive with Ranking SVM and RankBoost in the task of ranking relevant documents at the very top positions.",
keywords = "Learning to rank, Particle swarm optimization, Ranking function, Swarm intelligence",
author = "Ernesto Diaz-Aviles and Wolfgang Nejdl and Lars Schmidt-Thieme",
year = "2009",
month = jul,
day = "8",
doi = "10.1145/1569901.1569904",
language = "English",
isbn = "9781605583259",
series = "Proceedings of the 11th Annual Genetic and Evolutionary Computation Conference, GECCO-2009",
pages = "9--15",
booktitle = "Proceedings of the 11th Annual Genetic and Evolutionary Computation Conference, GECCO-2009",
note = "11th Annual Genetic and Evolutionary Computation Conference, GECCO-2009 ; Conference date: 08-07-2009 Through 12-07-2009",
}