cikm_field_weights.dvi

Size: px

Start display at page:

Download "cikm_field_weights.dvi"

まなありの
7 years ago
Views:

1 13Simple BM25 Extension to Multiple Weighted Fields Stephen Robertson, Hugo Zaragoza and Michael Taylor Microsoft Research ﾁﾁﾁ1 7 ﾁ Cambridge, U.K ﾁﾁﾁﾁﾁﾁ ser,hugoz,mitaylor1 7@microsoft.com ﾁﾁ1 7 ﾁﾁﾁﾁﾁ02 ABSTRACT ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ Categories and Subject Descriptors ﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ0507 ﾁﾁ General Terms ﾁﾁ06 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ00 1. INTRODUCTION Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CIKM ｯ04, November 8ｨC13, 2004, Washington, DC, USA. Copyright 2004 ACM /04/ ﾁﾁﾁﾁﾁ0706 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ07 ﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁ0706 ﾁﾁﾁﾁﾁﾁ02 ﾁﾁﾁ02 ﾁﾁ1 706 ﾁﾁﾁﾁﾁ0706 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ06 ﾁﾁﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁ00 ﾁﾁﾁﾁｻﾁﾁ021 7 ﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁｻﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁ Prior Art

001 7061 71 7061 71 707000502041 701071 71 71 704041 7011 7 ﾁ0100020102001 71 71 70702051 70601 ﾁ020906011 71 7 081 7 ﾁ011 7011 71 7 ﾁ1 71 71 7060705050706010707050802011 7 ﾁ07001 7 ﾁ1 70700011 71 71

2 ﾁﾁﾁﾁﾁﾁﾁﾁ06 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ06 ﾁﾁﾁﾁﾁﾁﾁﾁｻ80800 ﾁ06 ﾁﾁﾁｻﾁﾁ071 7 ﾁﾁｻﾁﾁ06 ﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁ02 ﾁﾁ071 7 ﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ071 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ THE PROBLEM 2.1 Documents and Weighted Fields ﾁｻﾁﾁﾁﾁﾁｻﾁﾁｻﾁｻﾁﾁﾁｻｻｻｻﾁﾁ0706 ﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁｻｻﾁﾁｻｻﾁ1 7 ﾁﾁ02 ﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁ0807 ﾁﾁﾁﾁﾁﾁ02 ﾁﾁﾁ Field Score Combination

700051 706061 7000107 0800071 7041 705071 7 ﾁ0100020102001 71 70607051 71 7061 7061 71 706 ﾁ07051 7041 706011 71 7 ﾁ07001 7 ﾁ 0807041 7031 7001 70700011 71 70507 ﾁ01081 700011 71 7011 707001 701001

3 13term weight ScoreComb FreqComb Raw ﾁｻﾁﾁﾁ tf(body) + weight * tf(title) ﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁ Issues with the linear combination of scores ﾁﾁﾁﾁ001 7 ﾁﾁﾁﾁﾁ Nonlinear TF ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁ011 7 ﾁﾁﾁﾁﾁ Choosing ﾁﾁ1 7 ﾁﾁﾁﾁ Collection Statistics ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁｻ Equal field weights 0906 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁ Document length ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ Function ﾁ1 704 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ parameters ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁ021 7 ﾁﾁ02 3. PROPOSED METHOD: LINEAR COM- BINATION OF TFS

70508041 71 70106071 7011 71 7 011 7041 71 7061 7011 71 706051 705011 71 7051 7051 76ｻ80209021 7051 7011 7071 7 ﾁ051 706041 71 7000406 041 7061 71 70007051 71 7061 7011 70706071 7 ﾁ07001 7 ﾁ02 0.

4 ﾁﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ Theoretical basis and BM ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁ1 7 ﾁﾁ01 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ00 ﾁﾁﾁﾁ 1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁﾁﾁ Properties of the method ﾁ02 ﾁﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 701 ﾁﾁ1 7 ﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁ0706 ﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ and ﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁ01 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ

71 70408 0906011 71 7 ﾁﾁ1 7061 7001 70700011 71 7011 70005041 71 71 71 7011 7061 71 7061 7 ﾁ07001 7061 71 70206011 70706 ﾁ 1 7001 71 70808041 71 71 70706040607061 701071 71 71 71 70702051 7060102 1

5 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ Generality of the approach ﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ02 ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁ00 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ Anchor text ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ EMPIRICAL EVALUATION ﾁﾁﾁﾁﾁﾁﾁﾁ06 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ071 7 ﾁﾁﾁﾁﾁ Evaluation Method ﾁﾁｻｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 701 ﾁﾁﾁ1 7 ﾁﾁﾁﾁ06 ﾁﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻ8 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ0706 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁ

7061 71 7 ﾁ051 7011 7071 7001 709021 7001 7 ﾁﾁ021 7 ﾁ011 706011 71 7040406041 7 ﾁﾁ 0708011 7051 7 ﾁ1 7011 70706011 71 706011 71 7041 7061 71 70007051 71 7061 7011 70706071 7 ﾁ07001 7 ﾁ020907051 7 1

6 Reuters vi ﾁ Title & Body 0.25 dotgov, Title and Body fields only 0.64 Body Only Precision at rank= FreqComb ScoreComb Precision at rank= FreqComb ScoreComb Title Weight (scale is 10 6ﾓ11 for ScoreComb) 0.18 Body & Title (no weighting) Body Only Title Weight (scale is 10 6ﾓ11 for ScoreComb) Title Weight FreqComb Title Weight ﾁﾁﾁ ScoreComb Anchor Weight Anchor Weight

18 Body & Title (no weighting) Body Only 0.17 0 10 20 30 40 50 60 70 80 90 100 Title Weight (scale is 10 6ﾓ11 for ScoreComb) Title Weight 100 90 80 70 60 50 40 30 0.24493 0.24493 0.23676 FreqComb 0.

7 ﾁﾁﾁ1 7 ﾁﾁ021 7 ﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ07 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ01 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ071 7 ﾁﾁﾁﾁﾁﾁﾁ Body and Text ﾁ0100 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁ041 7 ﾁﾁﾁﾁﾁﾁ0100 ﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁ1 7 ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁ07 ﾁﾁﾁﾁﾁﾀﾁﾁﾁ1 7 ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾀﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ0800 ﾁﾁﾁ001 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ0706 ﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁ Anchor Text ﾁ1 7 ﾁﾁﾁ02 ﾁﾁﾁﾁﾁﾁﾁ0706 ﾁﾁﾁﾁ 1 7 ﾁﾁﾁﾁ071 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ07 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 702 ﾁﾁﾁ 1 7 ﾁﾁﾁﾁﾁﾁﾁﾀﾁﾁﾁﾁﾁﾀﾁﾁﾁﾁ011 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾀﾁ

7000705011 71 7041 71 71 71 701 ﾁ1 7061 7011 71 70704041 7011 70706 ﾁ011 7011 7 ﾁ011 7 ﾁ00 1 7061 7 ﾁ07011 71 7 ﾁ1 7 ﾁ1 707061 7041 7011 70702011 7060607 ﾁ0102 1 71 71 7041 7041 71 71 71 701 ﾁ031

8 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ02 ﾁﾁ CONCLUSION AND FUTURE WORK ﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ021 7 ﾁﾁﾁ071 7 ﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁ1 7 ﾁﾁﾁﾁﾁﾁﾁ REFERENCES ｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁｻﾁﾁｻﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁｻﾁｻﾁ1 7 ﾁｻﾁﾁｻﾁﾁﾁﾁﾁﾁﾁｻｻﾁｻﾁﾁ081 7 ﾁﾁﾁﾁｻﾁﾁﾁﾁﾁｻﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁｻﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁﾁｻ80807 ﾁﾁﾁﾁﾁﾁﾁ

71 70700001 701041 7041 71 71 71 701041 7061 7011 71 7 1 7061 70700041 71 71 71 70101070402071 71 7071 7061 71 71 7011 71 71 7061 7 ﾁ1 707061 7 ﾁ011 7001 706011 71 7 011 702001 70802001 71 7 ﾁ011

すべて見る

mobicom.dvi

mobicom.dvi 13Dynamic Voltage Scaling on a Low-Power Microprocessor Johan Pouwelse 5 Koen Langendoen Henk Sips Faculty of Information Technology and Systems Delft University of Technology, The Netherlands 1 78724