SlideShare une entreprise Scribd logo
1  sur  9
Télécharger pour lire hors ligne
紹介する評価手法
•DCG
•NDCG
•MAP
•ケンドールの順位相関係数
(Kendall s Tau)
DCG(Discount Cumulative count)
p.17
•DCGとは measures the goodness
of the ranking list with the labels
4 EVALUATION
luation ontheperformance ofaranking model iscarried outbycomparison between the
output by the model and the ranking Usts given as ground truth. Several evaluation m
widelyused in IR and other fields. These include NDCG (Normalized Discounted Cum
n), DCG (Discounted Cumulative Gain) [53], MAP (Mean Average Precision) [101
nnersTakeAll),MRR (Mean Reciprocal Rank), and Kendall's Tau.
Given queryqi and associated documents D,,suppose thatttj istheranking list (perm
Diandy/ isthesetoflabels (grades) ofD,-. DCGmeasures thegoodness oftheranking
abels. Specifically, DCG at position k forqi isdefined:
DCG(k)= J2 GU)D(7n(j)),
j:ni(j)<k
reG(-) is a gain function and £>(•) isa position discount function. Note that 7T; (y) den
tion ofdij in717.Therefore, the summation is taken over the top kpositions in ranking
G represents the cumulative gain ofaccessing the information from position one topo
e, the definition ofNDCG (or DCG) arc formulated based on the indices ofdocuments. Itis also possible to defin
DCG) basedon the indices of positions.
DCGは「トップからk番目までを評価する」
G(j)の内容 p.18
• G(j)とは「関係あるdocumentがどれだけπ_iに存在する
か?」を示す指標
In a perfect ranking, the documents with higher g
multiple perfectrankings for a query and associated
n is normally defined as an exponential function
informationexponentially increases when grade o
GO) = 2^-1,
rade) ofdjj inranking list 7r, .Thediscount functio
position. That is to say, satisfaction of accessing inf
ここで,y_(i,j)はdocument, d_(i,j)に与えられたラベル
ラベルの値が高いほど,G(j)の値は高くなる.
documents, d_i=(d_(i,1), d_(i,2), d_(i,3)) に対して,
ラベル集合はY_i=(3, 3, 2)のように与えられる.
D(π_i(j))の内容 p.19
•D(π_i(j))はd_(i,j)の順位が低いほど,小さ
くなる値
π_i(j)はdocument, d_(i,j)の順位を示す.
2.2. LEARNING TASK
eases when positionof information access increases.
1
D(TTiU)) =
log20+*,(./))'
re7r,0) is the positionof djj in rankinglist7T,-.
Hence, DCG and NDCG at positionk for q-t become
V^ V'-i - 1
DCG(k) = ) -—,
NDCG(k)=DCG-uk) e j;;;;a))
DCG and NDCG of the whole ranking list for qi become
D(π_i(j))はπ_i(j)が1の時,つまりdocument,d_i(j)が一位の
時,最大.
document,d_i(j)の順位が下がるほど,分母の値は大きくなる
ので,D(π_i(j))の値は小さくなる
もどって,DCGの説明(p.17)
• G(j)は関係あるdocumentほど高くなる値
• D(π_i(j))はd_(i,j)の順位が低いほど,低くなる値
EVALUATION
uation ontheperformance ofaranking model iscarried outbycomparison between ther
output by the model and the ranking Usts given as ground truth. Several evaluation m
idelyused in IR and other fields. These include NDCG (Normalized Discounted Cum
), DCG (Discounted Cumulative Gain) [53], MAP (Mean Average Precision) [101],
nersTakeAll),MRR (Mean Reciprocal Rank), and Kendall's Tau.
Given queryqi and associated documents D,,suppose thatttj istheranking list (permu
iandy/ isthesetoflabels (grades) ofD,-. DCGmeasures thegoodness oftheranking li
bels. Specifically, DCG at position k forqi isdefined:
DCG(k)= J2 GU)D(7n(j)),
j:ni(j)<k
eG(-) is a gain function and £>(•) isa position discount function. Note that 7T; (y) deno
on ofdij in717.Therefore, the summation is taken over the top kpositions in ranking l
represents the cumulative gain ofaccessing the information from position one topos
the definition ofNDCG (or DCG) arc formulated based on the indices ofdocuments. Itis also possible to define
CG) basedon the indices of positions.
DCGはG(j)が高い値ばかりで,D(π_i(j))も高い値ばかりの時に
大きくなる.
つまり,「k番目までのdocumentが高い値のラベルをも
ち」,「k番目までのdocumentがリストπの中で高い順位に
ある」時にDCGは大きくなる.
NDCGの説明(p.19)
D(TTiU)) =
log20+*,(./))'
7r,0) is the positionof djj in rankinglist7T,-.
Hence, DCG and NDCG at positionk for q-t become
V^ V'-i - 1
DCG(k) = ) -—,
NDCG(k)=DCG-uk) e j;;;;a))
DCG and NDCG of the whole ranking list for qi become
DCG= £
. log2(I +^(7))'J.ni(j)<rti
NDCG = DCG~mx £. log2(l+*,(./))
J-*i{j)<ni
DCG and NDCG values are further averaged overqueries(/ = 1, ••• , m).
Table2.4 gives examples of calculating NDCG values of two ranking Usts. NDCG (
eeffect ofgiving highscores to the ranking lists inwhich relevant documents areranked
DCGを逆数として掛け合わせて,
正規化することになる.
MAP(Mean Average Precision)の説明(p.20)
. log2(l+*,(./))
J-*i{j)<ni
DCG and NDCG values are further averaged overqueries(/ = 1, ••• , m).
Table2.4 gives examples of calculating NDCG values of two ranking Usts. NDCG (DC
theeffect ofgiving highscores to the ranking lists inwhich relevant documents areranked h
the examples inTable2.4.Forthe perfect rankings, the NDCG value at each positionis alw
,whilefor imperfect rankings, the NDCG values areless than one.
MAP isanother measure widely usedin IR.In MAP,it isassumed that the gradesofreleva
at two levels: 1 and 0. Given queryq;,associated documents D,, rankingHst 7T, on D;, and la
f Di, Average Precision forqt isdefined:
£/=i yij
re ytj is the label (grade) of dij and takes on 1 or 0 as value, representing being relevan
evant. P(j) for query qt is defined:
p, .x = T,k:Tri(k)<niU) y'<k
*iU)
re JTj(j) is the position of dij in jtj. P(j) represents the precision until the position ofdij
Note that labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbedefined. Ave
cision represents averaged precision over allthepositions ofdocuments with label 1forquer
Average Precisionは
MAPの最大の特徴はラベルが「0と1」だけ
ランキングリストの平均Precisionを返す
y_(i,j)は0と1のみ
j=1からn_iまでなので,
すべてのdocumentのランキングを評価する
t ofgiving highscores to the ranking lists inwhich relevant documents ar
mples inTable2.4.Forthe perfect rankings, the NDCG value at eachpos
or imperfect rankings, the NDCG values areless than one.
isanother measure widely usedin IR.In MAP,it isassumed that the grad
vels: 1 and 0. Given queryq;,associated documents D,, rankingHst 7T, on
erage Precision forqt isdefined:
£/=i yij
s the label (grade) of dij and takes on 1 or 0 as value, representing bei
(j) for query qt is defined:
p, .x = T,k:Tri(k)<niU) y'<k
*iU)
is the position ofdij in jtj. P(j) represents the precision until the posit
labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbedef
Precisionは
MAP(Mean Average Precision)の説明(p.20)
£/=i yij
is the label (grade) of dij and takes on 1 or 0 as value, representing be
P(j) for query qt is defined:
p, .x = T,k:Tri(k)<niU) y'<k
*iU)
) is the position ofdij in jtj. P(j) represents the precision until the pos
at labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbed
epresents averaged precision over allthepositions ofdocuments with labe
Precisionは
π_i
document,jの順位
この範囲の{0,1}の
合計値
P(j)=document,jまでに関係あるdocumentがどれだけあるか?/
document,jの順位
この範囲に1(関係ある)のdocument
が多いほど,P(j)は高い値
Kendall s Tau(ケンドールの順位相関係数) p.20
ケンドールの相関係数は
「2つのリストの中でアイテムペアの順序関係がどれだけ一致しているか?」を評価
(ここではGoldのリストとシステムによるリスト)
数値の範囲は-1 +1.+1に近いほど「関係性あり」,-1に近いほど「関係なし」
arefurtheraveraged overqueries to become MeanAverage Precision
mple ofcalculating the AP value ofone ranking Ust.
re proposed in statistics. It isdefined on two rankingUsts: one is the
del, andthe other isbythe groundtruth. Kendall's Tau of rankingUst
h tt* isdefined:
2c,
Ti = -1,
2n/0»i - l)
of concordant pairs between the two Usts, and /!/ denotes the length
KendaUs Tau between two ranking Usts: (A,B,C) and (C,A,B) is as
2x1 1
een —1 and +1. If the two ranking Usts are exactlythe same, then it
reverse orderof the other, then it is —1.It is easyto verify KendaUs
n_i:アイテムの数
c_i:順序が一致したアイテムペア数
実は(n_i)C2を展開した式
例えば(A,B,C)と(C,A,B)のとき
n_i:アイテムの数=3
順序が一致したアイテムペアは(A,B)のみだから,c_i=1
分母,つまり考えられるアイテムペア数は3C2=3
結果,T_iは-0.3333...で,「あまり関係性がない」と言える
参考: http://d.hatena.ne.jp/sleepy_yoshi/20110326/p1

Contenu connexe

Tendances

Automated building of taxonomies for search engines
Automated building of taxonomies for search enginesAutomated building of taxonomies for search engines
Automated building of taxonomies for search enginesBoris Galitsky
 
Encoding Linguistic Structures with Graph Convolutional Networks
Encoding Linguistic Structures with Graph Convolutional NetworksEncoding Linguistic Structures with Graph Convolutional Networks
Encoding Linguistic Structures with Graph Convolutional NetworksAleksandar Savkov
 
Deep Packet Inspection with Regular Expression Matching
Deep Packet Inspection with Regular Expression MatchingDeep Packet Inspection with Regular Expression Matching
Deep Packet Inspection with Regular Expression MatchingEditor IJCATR
 
Data warehousing and data mining
Data warehousing and data miningData warehousing and data mining
Data warehousing and data miningvamsi krishna
 
Introduction to database-Formal Query language and Relational calculus
Introduction to database-Formal Query language and Relational calculusIntroduction to database-Formal Query language and Relational calculus
Introduction to database-Formal Query language and Relational calculusAjit Nayak
 
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...Jinho Choi
 
RDataMining slides-text-mining-with-r
RDataMining slides-text-mining-with-rRDataMining slides-text-mining-with-r
RDataMining slides-text-mining-with-rYanchang Zhao
 
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For Nagios
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For NagiosNagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For Nagios
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For NagiosNagios
 
Competence-Level Prediction and Resume & Job Description Matching Using Conte...
Competence-Level Prediction and Resume & Job Description Matching Using Conte...Competence-Level Prediction and Resume & Job Description Matching Using Conte...
Competence-Level Prediction and Resume & Job Description Matching Using Conte...Jinho Choi
 
Graph-to-Text Generation and its Applications to Dialogue
Graph-to-Text Generation and its Applications to DialogueGraph-to-Text Generation and its Applications to Dialogue
Graph-to-Text Generation and its Applications to DialogueJinho Choi
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentIJERD Editor
 
Text Mining Using R
Text Mining Using RText Mining Using R
Text Mining Using RKnoldus Inc.
 
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCES
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCESFINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCES
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCESkevig
 
[系列活動] 手把手的深度學習實務
[系列活動] 手把手的深度學習實務[系列活動] 手把手的深度學習實務
[系列活動] 手把手的深度學習實務台灣資料科學年會
 
Sample Question Paper IP Class xii
Sample Question Paper IP Class xii Sample Question Paper IP Class xii
Sample Question Paper IP Class xii kvs
 
An Interactive Introduction To R (Programming Language For Statistics)
An Interactive Introduction To R (Programming Language For Statistics)An Interactive Introduction To R (Programming Language For Statistics)
An Interactive Introduction To R (Programming Language For Statistics)Dataspora
 
Context-based movie search using doc2vec, word2vec
Context-based movie search using doc2vec, word2vecContext-based movie search using doc2vec, word2vec
Context-based movie search using doc2vec, word2vecJIN KYU CHANG
 
Text mining and social network analysis of twitter data part 1
Text mining and social network analysis of twitter data part 1Text mining and social network analysis of twitter data part 1
Text mining and social network analysis of twitter data part 1Johan Blomme
 
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GANRecent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GANHao-Wen (Herman) Dong
 

Tendances (19)

Automated building of taxonomies for search engines
Automated building of taxonomies for search enginesAutomated building of taxonomies for search engines
Automated building of taxonomies for search engines
 
Encoding Linguistic Structures with Graph Convolutional Networks
Encoding Linguistic Structures with Graph Convolutional NetworksEncoding Linguistic Structures with Graph Convolutional Networks
Encoding Linguistic Structures with Graph Convolutional Networks
 
Deep Packet Inspection with Regular Expression Matching
Deep Packet Inspection with Regular Expression MatchingDeep Packet Inspection with Regular Expression Matching
Deep Packet Inspection with Regular Expression Matching
 
Data warehousing and data mining
Data warehousing and data miningData warehousing and data mining
Data warehousing and data mining
 
Introduction to database-Formal Query language and Relational calculus
Introduction to database-Formal Query language and Relational calculusIntroduction to database-Formal Query language and Relational calculus
Introduction to database-Formal Query language and Relational calculus
 
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...
Adaptation of Multilingual Transformer Encoder for Robust Enhanced Universal ...
 
RDataMining slides-text-mining-with-r
RDataMining slides-text-mining-with-rRDataMining slides-text-mining-with-r
RDataMining slides-text-mining-with-r
 
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For Nagios
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For NagiosNagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For Nagios
Nagios Conference 2012 - Nicholas Scott - Advanced Data Analytics For Nagios
 
Competence-Level Prediction and Resume & Job Description Matching Using Conte...
Competence-Level Prediction and Resume & Job Description Matching Using Conte...Competence-Level Prediction and Resume & Job Description Matching Using Conte...
Competence-Level Prediction and Resume & Job Description Matching Using Conte...
 
Graph-to-Text Generation and its Applications to Dialogue
Graph-to-Text Generation and its Applications to DialogueGraph-to-Text Generation and its Applications to Dialogue
Graph-to-Text Generation and its Applications to Dialogue
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
 
Text Mining Using R
Text Mining Using RText Mining Using R
Text Mining Using R
 
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCES
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCESFINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCES
FINDING OUT NOISY PATTERNS FOR RELATION EXTRACTION OF BANGLA SENTENCES
 
[系列活動] 手把手的深度學習實務
[系列活動] 手把手的深度學習實務[系列活動] 手把手的深度學習實務
[系列活動] 手把手的深度學習實務
 
Sample Question Paper IP Class xii
Sample Question Paper IP Class xii Sample Question Paper IP Class xii
Sample Question Paper IP Class xii
 
An Interactive Introduction To R (Programming Language For Statistics)
An Interactive Introduction To R (Programming Language For Statistics)An Interactive Introduction To R (Programming Language For Statistics)
An Interactive Introduction To R (Programming Language For Statistics)
 
Context-based movie search using doc2vec, word2vec
Context-based movie search using doc2vec, word2vecContext-based movie search using doc2vec, word2vec
Context-based movie search using doc2vec, word2vec
 
Text mining and social network analysis of twitter data part 1
Text mining and social network analysis of twitter data part 1Text mining and social network analysis of twitter data part 1
Text mining and social network analysis of twitter data part 1
 
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GANRecent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
 

En vedette

クラシックな機械学習の入門  11.評価方法
クラシックな機械学習の入門  11.評価方法クラシックな機械学習の入門  11.評価方法
クラシックな機械学習の入門  11.評価方法Hiroshi Nakagawa
 
情報検索における評価指標の最新動向と新たな提案
情報検索における評価指標の最新動向と新たな提案情報検索における評価指標の最新動向と新たな提案
情報検索における評価指標の最新動向と新たな提案Mitsuo Yamamoto
 
アダルトデータマイニングの勧め
アダルトデータマイニングの勧めアダルトデータマイニングの勧め
アダルトデータマイニングの勧めKensuke Mitsuzawa
 
slides for "Supervised Model Learning with Feature Grouping based on a Discre...
slides for "Supervised Model Learning with Feature Grouping based on a Discre...slides for "Supervised Model Learning with Feature Grouping based on a Discre...
slides for "Supervised Model Learning with Feature Grouping based on a Discre...Kensuke Mitsuzawa
 
形態素解析器の比較できるPythonパッケージつくった話
形態素解析器の比較できるPythonパッケージつくった話形態素解析器の比較できるPythonパッケージつくった話
形態素解析器の比較できるPythonパッケージつくった話Kensuke Mitsuzawa
 
Tweet Recommendation with Graph Co-Ranking
Tweet Recommendation with Graph Co-RankingTweet Recommendation with Graph Co-Ranking
Tweet Recommendation with Graph Co-RankingYoshinari Fujinuma
 
ラベル付けのいろは
ラベル付けのいろはラベル付けのいろは
ラベル付けのいろはKensuke Mitsuzawa
 
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJPYahoo!デベロッパーネットワーク
 
サポーターズ勉強会スライド
サポーターズ勉強会スライドサポーターズ勉強会スライド
サポーターズ勉強会スライドKensuke Mitsuzawa
 
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...Md. Main Uddin Rony
 
INTRODUCTION INFORMATION RETRIEVAL EVALUVATION
 INTRODUCTION INFORMATION RETRIEVAL EVALUVATION INTRODUCTION INFORMATION RETRIEVAL EVALUVATION
INTRODUCTION INFORMATION RETRIEVAL EVALUVATIONPremsankar Chakkingal
 
今日から使える! みんなのクラスタリング超入門
今日から使える! みんなのクラスタリング超入門今日から使える! みんなのクラスタリング超入門
今日から使える! みんなのクラスタリング超入門toilet_lunch
 
機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレスト機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレストTeppei Baba
 
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京Koichi Hamada
 

En vedette (14)

クラシックな機械学習の入門  11.評価方法
クラシックな機械学習の入門  11.評価方法クラシックな機械学習の入門  11.評価方法
クラシックな機械学習の入門  11.評価方法
 
情報検索における評価指標の最新動向と新たな提案
情報検索における評価指標の最新動向と新たな提案情報検索における評価指標の最新動向と新たな提案
情報検索における評価指標の最新動向と新たな提案
 
アダルトデータマイニングの勧め
アダルトデータマイニングの勧めアダルトデータマイニングの勧め
アダルトデータマイニングの勧め
 
slides for "Supervised Model Learning with Feature Grouping based on a Discre...
slides for "Supervised Model Learning with Feature Grouping based on a Discre...slides for "Supervised Model Learning with Feature Grouping based on a Discre...
slides for "Supervised Model Learning with Feature Grouping based on a Discre...
 
形態素解析器の比較できるPythonパッケージつくった話
形態素解析器の比較できるPythonパッケージつくった話形態素解析器の比較できるPythonパッケージつくった話
形態素解析器の比較できるPythonパッケージつくった話
 
Tweet Recommendation with Graph Co-Ranking
Tweet Recommendation with Graph Co-RankingTweet Recommendation with Graph Co-Ranking
Tweet Recommendation with Graph Co-Ranking
 
ラベル付けのいろは
ラベル付けのいろはラベル付けのいろは
ラベル付けのいろは
 
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP
第16回Lucene/Solr勉強会 – ランキングチューニングと定量評価 #SolrJP
 
サポーターズ勉強会スライド
サポーターズ勉強会スライドサポーターズ勉強会スライド
サポーターズ勉強会スライド
 
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...
Data Analysis: Evaluation Metrics for Supervised Learning Models of Machine L...
 
INTRODUCTION INFORMATION RETRIEVAL EVALUVATION
 INTRODUCTION INFORMATION RETRIEVAL EVALUVATION INTRODUCTION INFORMATION RETRIEVAL EVALUVATION
INTRODUCTION INFORMATION RETRIEVAL EVALUVATION
 
今日から使える! みんなのクラスタリング超入門
今日から使える! みんなのクラスタリング超入門今日から使える! みんなのクラスタリング超入門
今日から使える! みんなのクラスタリング超入門
 
機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレスト機会学習ハッカソン:ランダムフォレスト
機会学習ハッカソン:ランダムフォレスト
 
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京
「はじめてでもわかる RandomForest 入門-集団学習による分類・予測 -」 -第7回データマイニング+WEB勉強会@東京
 

Similaire à Learning to rankの評価手法

learned optimizer.pptx
learned optimizer.pptxlearned optimizer.pptx
learned optimizer.pptxQingsong Guo
 
Parallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-JoinsParallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-JoinsJonny Daenen
 
PosterPresentations.com-3 6x48-Template-V5 - 副本
PosterPresentations.com-3 6x48-Template-V5 - 副本PosterPresentations.com-3 6x48-Template-V5 - 副本
PosterPresentations.com-3 6x48-Template-V5 - 副本Yijun Zhou
 
A modeling approach for integrating durability engineering and robustness in ...
A modeling approach for integrating durability engineering and robustness in ...A modeling approach for integrating durability engineering and robustness in ...
A modeling approach for integrating durability engineering and robustness in ...Phuong Dx
 
Planning Under Uncertainty With Markov Decision Processes
Planning Under Uncertainty With Markov Decision ProcessesPlanning Under Uncertainty With Markov Decision Processes
Planning Under Uncertainty With Markov Decision Processesahmad bassiouny
 
論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations
論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations
論文紹介:Towards Robust Adaptive Object Detection Under Noisy AnnotationsToru Tamaki
 
Discrete form of the riccati equation
Discrete form of the riccati equationDiscrete form of the riccati equation
Discrete form of the riccati equationAlberth Carantón
 
learning boolean weight learning real valued weights rank learning as ordina...
learning boolean weight learning real valued weights  rank learning as ordina...learning boolean weight learning real valued weights  rank learning as ordina...
learning boolean weight learning real valued weights rank learning as ordina...jaishriramm0
 
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...Spark Summit
 
Determining the k in k-means with MapReduce
Determining the k in k-means with MapReduceDetermining the k in k-means with MapReduce
Determining the k in k-means with MapReduceThibault Debatty
 
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1A Stochastic Model by the Fourier Transform of Pde for the Glp - 1
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1IJERA Editor
 
Optimization Techniques
Optimization TechniquesOptimization Techniques
Optimization TechniquesAjay Bidyarthy
 
Lecture 06 relational algebra and calculus
Lecture 06 relational algebra and calculusLecture 06 relational algebra and calculus
Lecture 06 relational algebra and calculusemailharmeet
 
Local Model Checking Algorithm Based on Mu-calculus with Partial Orders
Local Model Checking Algorithm Based on Mu-calculus with Partial OrdersLocal Model Checking Algorithm Based on Mu-calculus with Partial Orders
Local Model Checking Algorithm Based on Mu-calculus with Partial OrdersTELKOMNIKA JOURNAL
 
Cmt458 chapter 1 chemical thermodynamic
Cmt458 chapter 1 chemical thermodynamicCmt458 chapter 1 chemical thermodynamic
Cmt458 chapter 1 chemical thermodynamicMiza Kamaruzzaman
 

Similaire à Learning to rankの評価手法 (20)

learned optimizer.pptx
learned optimizer.pptxlearned optimizer.pptx
learned optimizer.pptx
 
Parallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-JoinsParallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-Joins
 
PosterPresentations.com-3 6x48-Template-V5 - 副本
PosterPresentations.com-3 6x48-Template-V5 - 副本PosterPresentations.com-3 6x48-Template-V5 - 副本
PosterPresentations.com-3 6x48-Template-V5 - 副本
 
A modeling approach for integrating durability engineering and robustness in ...
A modeling approach for integrating durability engineering and robustness in ...A modeling approach for integrating durability engineering and robustness in ...
A modeling approach for integrating durability engineering and robustness in ...
 
Planning Under Uncertainty With Markov Decision Processes
Planning Under Uncertainty With Markov Decision ProcessesPlanning Under Uncertainty With Markov Decision Processes
Planning Under Uncertainty With Markov Decision Processes
 
論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations
論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations
論文紹介:Towards Robust Adaptive Object Detection Under Noisy Annotations
 
Discrete form of the riccati equation
Discrete form of the riccati equationDiscrete form of the riccati equation
Discrete form of the riccati equation
 
learning boolean weight learning real valued weights rank learning as ordina...
learning boolean weight learning real valued weights  rank learning as ordina...learning boolean weight learning real valued weights  rank learning as ordina...
learning boolean weight learning real valued weights rank learning as ordina...
 
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...
A Scalable Hierarchical Clustering Algorithm Using Spark: Spark Summit East t...
 
BNL_Research_Report
BNL_Research_ReportBNL_Research_Report
BNL_Research_Report
 
Determining the k in k-means with MapReduce
Determining the k in k-means with MapReduceDetermining the k in k-means with MapReduce
Determining the k in k-means with MapReduce
 
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1A Stochastic Model by the Fourier Transform of Pde for the Glp - 1
A Stochastic Model by the Fourier Transform of Pde for the Glp - 1
 
Optimization Techniques
Optimization TechniquesOptimization Techniques
Optimization Techniques
 
Lecture 06 relational algebra and calculus
Lecture 06 relational algebra and calculusLecture 06 relational algebra and calculus
Lecture 06 relational algebra and calculus
 
Local Model Checking Algorithm Based on Mu-calculus with Partial Orders
Local Model Checking Algorithm Based on Mu-calculus with Partial OrdersLocal Model Checking Algorithm Based on Mu-calculus with Partial Orders
Local Model Checking Algorithm Based on Mu-calculus with Partial Orders
 
Cmt458 chapter 1 chemical thermodynamic
Cmt458 chapter 1 chemical thermodynamicCmt458 chapter 1 chemical thermodynamic
Cmt458 chapter 1 chemical thermodynamic
 
05-Debug.pdf
05-Debug.pdf05-Debug.pdf
05-Debug.pdf
 
QUARTILE DEVIATION
QUARTILE DEVIATIONQUARTILE DEVIATION
QUARTILE DEVIATION
 
PCA and SVD in brief
PCA and SVD in briefPCA and SVD in brief
PCA and SVD in brief
 
Missing Data imputation
Missing Data imputationMissing Data imputation
Missing Data imputation
 

Dernier

Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural Resources
Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural ResourcesEnergy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural Resources
Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural ResourcesShubhangi Sonawane
 
Application orientated numerical on hev.ppt
Application orientated numerical on hev.pptApplication orientated numerical on hev.ppt
Application orientated numerical on hev.pptRamjanShidvankar
 
Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17Celine George
 
PROCESS RECORDING FORMAT.docx
PROCESS      RECORDING        FORMAT.docxPROCESS      RECORDING        FORMAT.docx
PROCESS RECORDING FORMAT.docxPoojaSen20
 
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17  How to Extend Models Using Mixin ClassesMixin Classes in Odoo 17  How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17 How to Extend Models Using Mixin ClassesCeline George
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxDenish Jangid
 
Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactPECB
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdfQucHHunhnh
 
Measures of Dispersion and Variability: Range, QD, AD and SD
Measures of Dispersion and Variability: Range, QD, AD and SDMeasures of Dispersion and Variability: Range, QD, AD and SD
Measures of Dispersion and Variability: Range, QD, AD and SDThiyagu K
 
Sociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning ExhibitSociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning Exhibitjbellavia9
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsTechSoup
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfAyushMahapatra5
 
Micro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdfMicro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdfPoh-Sun Goh
 
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...Shubhangi Sonawane
 
Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfagholdier
 
The basics of sentences session 2pptx copy.pptx
The basics of sentences session 2pptx copy.pptxThe basics of sentences session 2pptx copy.pptx
The basics of sentences session 2pptx copy.pptxheathfieldcps1
 
This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.christianmathematics
 
Making and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdfMaking and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdfChris Hunter
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxVishalSingh1417
 

Dernier (20)

Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural Resources
Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural ResourcesEnergy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural Resources
Energy Resources. ( B. Pharmacy, 1st Year, Sem-II) Natural Resources
 
Application orientated numerical on hev.ppt
Application orientated numerical on hev.pptApplication orientated numerical on hev.ppt
Application orientated numerical on hev.ppt
 
Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17Advanced Views - Calendar View in Odoo 17
Advanced Views - Calendar View in Odoo 17
 
PROCESS RECORDING FORMAT.docx
PROCESS      RECORDING        FORMAT.docxPROCESS      RECORDING        FORMAT.docx
PROCESS RECORDING FORMAT.docx
 
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17  How to Extend Models Using Mixin ClassesMixin Classes in Odoo 17  How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
 
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptxBasic Civil Engineering first year Notes- Chapter 4 Building.pptx
Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
 
Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global Impact
 
1029-Danh muc Sach Giao Khoa khoi 6.pdf
1029-Danh muc Sach Giao Khoa khoi  6.pdf1029-Danh muc Sach Giao Khoa khoi  6.pdf
1029-Danh muc Sach Giao Khoa khoi 6.pdf
 
Measures of Dispersion and Variability: Range, QD, AD and SD
Measures of Dispersion and Variability: Range, QD, AD and SDMeasures of Dispersion and Variability: Range, QD, AD and SD
Measures of Dispersion and Variability: Range, QD, AD and SD
 
Sociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning ExhibitSociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning Exhibit
 
Introduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The BasicsIntroduction to Nonprofit Accounting: The Basics
Introduction to Nonprofit Accounting: The Basics
 
Class 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdfClass 11th Physics NEET formula sheet pdf
Class 11th Physics NEET formula sheet pdf
 
Micro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdfMicro-Scholarship, What it is, How can it help me.pdf
Micro-Scholarship, What it is, How can it help me.pdf
 
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...
Ecological Succession. ( ECOSYSTEM, B. Pharmacy, 1st Year, Sem-II, Environmen...
 
Holdier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdfHoldier Curriculum Vitae (April 2024).pdf
Holdier Curriculum Vitae (April 2024).pdf
 
Asian American Pacific Islander Month DDSD 2024.pptx
Asian American Pacific Islander Month DDSD 2024.pptxAsian American Pacific Islander Month DDSD 2024.pptx
Asian American Pacific Islander Month DDSD 2024.pptx
 
The basics of sentences session 2pptx copy.pptx
The basics of sentences session 2pptx copy.pptxThe basics of sentences session 2pptx copy.pptx
The basics of sentences session 2pptx copy.pptx
 
This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.This PowerPoint helps students to consider the concept of infinity.
This PowerPoint helps students to consider the concept of infinity.
 
Making and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdfMaking and Justifying Mathematical Decisions.pdf
Making and Justifying Mathematical Decisions.pdf
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptx
 

Learning to rankの評価手法

  • 2. DCG(Discount Cumulative count) p.17 •DCGとは measures the goodness of the ranking list with the labels 4 EVALUATION luation ontheperformance ofaranking model iscarried outbycomparison between the output by the model and the ranking Usts given as ground truth. Several evaluation m widelyused in IR and other fields. These include NDCG (Normalized Discounted Cum n), DCG (Discounted Cumulative Gain) [53], MAP (Mean Average Precision) [101 nnersTakeAll),MRR (Mean Reciprocal Rank), and Kendall's Tau. Given queryqi and associated documents D,,suppose thatttj istheranking list (perm Diandy/ isthesetoflabels (grades) ofD,-. DCGmeasures thegoodness oftheranking abels. Specifically, DCG at position k forqi isdefined: DCG(k)= J2 GU)D(7n(j)), j:ni(j)<k reG(-) is a gain function and £>(•) isa position discount function. Note that 7T; (y) den tion ofdij in717.Therefore, the summation is taken over the top kpositions in ranking G represents the cumulative gain ofaccessing the information from position one topo e, the definition ofNDCG (or DCG) arc formulated based on the indices ofdocuments. Itis also possible to defin DCG) basedon the indices of positions. DCGは「トップからk番目までを評価する」
  • 3. G(j)の内容 p.18 • G(j)とは「関係あるdocumentがどれだけπ_iに存在する か?」を示す指標 In a perfect ranking, the documents with higher g multiple perfectrankings for a query and associated n is normally defined as an exponential function informationexponentially increases when grade o GO) = 2^-1, rade) ofdjj inranking list 7r, .Thediscount functio position. That is to say, satisfaction of accessing inf ここで,y_(i,j)はdocument, d_(i,j)に与えられたラベル ラベルの値が高いほど,G(j)の値は高くなる. documents, d_i=(d_(i,1), d_(i,2), d_(i,3)) に対して, ラベル集合はY_i=(3, 3, 2)のように与えられる.
  • 4. D(π_i(j))の内容 p.19 •D(π_i(j))はd_(i,j)の順位が低いほど,小さ くなる値 π_i(j)はdocument, d_(i,j)の順位を示す. 2.2. LEARNING TASK eases when positionof information access increases. 1 D(TTiU)) = log20+*,(./))' re7r,0) is the positionof djj in rankinglist7T,-. Hence, DCG and NDCG at positionk for q-t become V^ V'-i - 1 DCG(k) = ) -—, NDCG(k)=DCG-uk) e j;;;;a)) DCG and NDCG of the whole ranking list for qi become D(π_i(j))はπ_i(j)が1の時,つまりdocument,d_i(j)が一位の 時,最大. document,d_i(j)の順位が下がるほど,分母の値は大きくなる ので,D(π_i(j))の値は小さくなる
  • 5. もどって,DCGの説明(p.17) • G(j)は関係あるdocumentほど高くなる値 • D(π_i(j))はd_(i,j)の順位が低いほど,低くなる値 EVALUATION uation ontheperformance ofaranking model iscarried outbycomparison between ther output by the model and the ranking Usts given as ground truth. Several evaluation m idelyused in IR and other fields. These include NDCG (Normalized Discounted Cum ), DCG (Discounted Cumulative Gain) [53], MAP (Mean Average Precision) [101], nersTakeAll),MRR (Mean Reciprocal Rank), and Kendall's Tau. Given queryqi and associated documents D,,suppose thatttj istheranking list (permu iandy/ isthesetoflabels (grades) ofD,-. DCGmeasures thegoodness oftheranking li bels. Specifically, DCG at position k forqi isdefined: DCG(k)= J2 GU)D(7n(j)), j:ni(j)<k eG(-) is a gain function and £>(•) isa position discount function. Note that 7T; (y) deno on ofdij in717.Therefore, the summation is taken over the top kpositions in ranking l represents the cumulative gain ofaccessing the information from position one topos the definition ofNDCG (or DCG) arc formulated based on the indices ofdocuments. Itis also possible to define CG) basedon the indices of positions. DCGはG(j)が高い値ばかりで,D(π_i(j))も高い値ばかりの時に 大きくなる. つまり,「k番目までのdocumentが高い値のラベルをも ち」,「k番目までのdocumentがリストπの中で高い順位に ある」時にDCGは大きくなる.
  • 6. NDCGの説明(p.19) D(TTiU)) = log20+*,(./))' 7r,0) is the positionof djj in rankinglist7T,-. Hence, DCG and NDCG at positionk for q-t become V^ V'-i - 1 DCG(k) = ) -—, NDCG(k)=DCG-uk) e j;;;;a)) DCG and NDCG of the whole ranking list for qi become DCG= £ . log2(I +^(7))'J.ni(j)<rti NDCG = DCG~mx £. log2(l+*,(./)) J-*i{j)<ni DCG and NDCG values are further averaged overqueries(/ = 1, ••• , m). Table2.4 gives examples of calculating NDCG values of two ranking Usts. NDCG ( eeffect ofgiving highscores to the ranking lists inwhich relevant documents areranked DCGを逆数として掛け合わせて, 正規化することになる.
  • 7. MAP(Mean Average Precision)の説明(p.20) . log2(l+*,(./)) J-*i{j)<ni DCG and NDCG values are further averaged overqueries(/ = 1, ••• , m). Table2.4 gives examples of calculating NDCG values of two ranking Usts. NDCG (DC theeffect ofgiving highscores to the ranking lists inwhich relevant documents areranked h the examples inTable2.4.Forthe perfect rankings, the NDCG value at each positionis alw ,whilefor imperfect rankings, the NDCG values areless than one. MAP isanother measure widely usedin IR.In MAP,it isassumed that the gradesofreleva at two levels: 1 and 0. Given queryq;,associated documents D,, rankingHst 7T, on D;, and la f Di, Average Precision forqt isdefined: £/=i yij re ytj is the label (grade) of dij and takes on 1 or 0 as value, representing being relevan evant. P(j) for query qt is defined: p, .x = T,k:Tri(k)<niU) y'<k *iU) re JTj(j) is the position of dij in jtj. P(j) represents the precision until the position ofdij Note that labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbedefined. Ave cision represents averaged precision over allthepositions ofdocuments with label 1forquer Average Precisionは MAPの最大の特徴はラベルが「0と1」だけ ランキングリストの平均Precisionを返す y_(i,j)は0と1のみ j=1からn_iまでなので, すべてのdocumentのランキングを評価する t ofgiving highscores to the ranking lists inwhich relevant documents ar mples inTable2.4.Forthe perfect rankings, the NDCG value at eachpos or imperfect rankings, the NDCG values areless than one. isanother measure widely usedin IR.In MAP,it isassumed that the grad vels: 1 and 0. Given queryq;,associated documents D,, rankingHst 7T, on erage Precision forqt isdefined: £/=i yij s the label (grade) of dij and takes on 1 or 0 as value, representing bei (j) for query qt is defined: p, .x = T,k:Tri(k)<niU) y'<k *iU) is the position ofdij in jtj. P(j) represents the precision until the posit labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbedef Precisionは
  • 8. MAP(Mean Average Precision)の説明(p.20) £/=i yij is the label (grade) of dij and takes on 1 or 0 as value, representing be P(j) for query qt is defined: p, .x = T,k:Tri(k)<niU) y'<k *iU) ) is the position ofdij in jtj. P(j) represents the precision until the pos at labels areeither 1or 0, and thusprecision (i.e.,ratioof label 1)canbed epresents averaged precision over allthepositions ofdocuments with labe Precisionは π_i document,jの順位 この範囲の{0,1}の 合計値 P(j)=document,jまでに関係あるdocumentがどれだけあるか?/ document,jの順位 この範囲に1(関係ある)のdocument が多いほど,P(j)は高い値
  • 9. Kendall s Tau(ケンドールの順位相関係数) p.20 ケンドールの相関係数は 「2つのリストの中でアイテムペアの順序関係がどれだけ一致しているか?」を評価 (ここではGoldのリストとシステムによるリスト) 数値の範囲は-1 +1.+1に近いほど「関係性あり」,-1に近いほど「関係なし」 arefurtheraveraged overqueries to become MeanAverage Precision mple ofcalculating the AP value ofone ranking Ust. re proposed in statistics. It isdefined on two rankingUsts: one is the del, andthe other isbythe groundtruth. Kendall's Tau of rankingUst h tt* isdefined: 2c, Ti = -1, 2n/0»i - l) of concordant pairs between the two Usts, and /!/ denotes the length KendaUs Tau between two ranking Usts: (A,B,C) and (C,A,B) is as 2x1 1 een —1 and +1. If the two ranking Usts are exactlythe same, then it reverse orderof the other, then it is —1.It is easyto verify KendaUs n_i:アイテムの数 c_i:順序が一致したアイテムペア数 実は(n_i)C2を展開した式 例えば(A,B,C)と(C,A,B)のとき n_i:アイテムの数=3 順序が一致したアイテムペアは(A,B)のみだから,c_i=1 分母,つまり考えられるアイテムペア数は3C2=3 結果,T_iは-0.3333...で,「あまり関係性がない」と言える 参考: http://d.hatena.ne.jp/sleepy_yoshi/20110326/p1