Bayesian Extension to the Language Model for Ad Hoc Information Retrieval

H. Zaragoza; Djoerd Hiemstra; M. Tipping; S.E. Robertson

doi:10.1145/860435.860439

Bayesian Extension to the Language Model for Ad Hoc Information Retrieval

H. Zaragoza
, Djoerd Hiemstra
, M. Tipping
, S.E. Robertson

Databases (Former)

Research output: Contribution to conference › Paper › peer-review

13 Downloads (Pure)

Abstract

We propose a Bayesian extension to the ad-hoc Language Model. Many smoothed estimators used for the multinomial query model in ad-hoc Language Models (including Laplace and Bayes-smoothing) are approximations to the Bayesian predictive distribution. In this paper we derive the full predictive distribution in a form amenable to implementation by classical IR models, and then compare it to other currently used estimators. In our experiments the proposed model outperforms Bayes-smoothing, and its combination with linear interpolation smoothing outperforms all other estimators.

Original language	Undefined
Pages	4-9
Number of pages	6
DOIs	https://doi.org/10.1145/860435.860439
Publication status	Published - Aug 2003
Event	26th Annual International ACM/SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2003 - Toronto, Canada Duration: 28 Jul 2003 → 1 Aug 2003 Conference number: 26

Conference

Conference	26th Annual International ACM/SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2003
Abbreviated title	SIGIR
Country/Territory	Canada
City	Toronto
Period	28/07/03 → 1/08/03

Keywords

DB-IR: INFORMATION RETRIEVAL
EWI-7381
IR-63543

Access to Document

10.1145/860435.860439

Cite this

@conference{594456d6ba0d44eaa10db2571dc5b5f2,

title = "Bayesian Extension to the Language Model for Ad Hoc Information Retrieval",

abstract = "We propose a Bayesian extension to the ad-hoc Language Model. Many smoothed estimators used for the multinomial query model in ad-hoc Language Models (including Laplace and Bayes-smoothing) are approximations to the Bayesian predictive distribution. In this paper we derive the full predictive distribution in a form amenable to implementation by classical IR models, and then compare it to other currently used estimators. In our experiments the proposed model outperforms Bayes-smoothing, and its combination with linear interpolation smoothing outperforms all other estimators.",

keywords = "DB-IR: INFORMATION RETRIEVAL, EWI-7381, IR-63543",

author = "H. Zaragoza and Djoerd Hiemstra and M. Tipping and S.E. Robertson",

note = "Imported from EWI/DB PMS [db-utwente:inpr:0000003269]; 26th Annual International ACM/SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2003 ; Conference date: 28-07-2003 Through 01-08-2003",

year = "2003",

month = aug,

doi = "10.1145/860435.860439",

language = "Undefined",

pages = "4--9",

}

TY - CONF

T1 - Bayesian Extension to the Language Model for Ad Hoc Information Retrieval

AU - Zaragoza, H.

AU - Hiemstra, Djoerd

AU - Tipping, M.

AU - Robertson, S.E.

N1 - Conference code: 26

PY - 2003/8

Y1 - 2003/8

N2 - We propose a Bayesian extension to the ad-hoc Language Model. Many smoothed estimators used for the multinomial query model in ad-hoc Language Models (including Laplace and Bayes-smoothing) are approximations to the Bayesian predictive distribution. In this paper we derive the full predictive distribution in a form amenable to implementation by classical IR models, and then compare it to other currently used estimators. In our experiments the proposed model outperforms Bayes-smoothing, and its combination with linear interpolation smoothing outperforms all other estimators.

AB - We propose a Bayesian extension to the ad-hoc Language Model. Many smoothed estimators used for the multinomial query model in ad-hoc Language Models (including Laplace and Bayes-smoothing) are approximations to the Bayesian predictive distribution. In this paper we derive the full predictive distribution in a form amenable to implementation by classical IR models, and then compare it to other currently used estimators. In our experiments the proposed model outperforms Bayes-smoothing, and its combination with linear interpolation smoothing outperforms all other estimators.

KW - DB-IR: INFORMATION RETRIEVAL

KW - EWI-7381

KW - IR-63543

U2 - 10.1145/860435.860439

DO - 10.1145/860435.860439

M3 - Paper

SP - 4

EP - 9

T2 - 26th Annual International ACM/SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2003

Y2 - 28 July 2003 through 1 August 2003

ER -