A Structured Approach to Query Recommendation With Social Annotation Data 童薇 2010/12/3.

A Structured Approach to Query Recommendation With Social Annotation Data

童薇

2010/12/3

Outline

Motivation Challenges Approach Experimental Results Conclusions

2010/12/3

Motivation

Query Recommendation Help users search Improve the usability of search engines

2010/12/3

Recommend what?

Existing Work Search interests: stick to user’s search intent

Anything Missing? Exploratory Interests: some vague or delitescent interests Unaware of until users are faced with one May be provoked within a search session

2010/12/3

smartphones

apple productsnexus one

mobilemeipod touch

equivalent or highly related queries

apple iphone

Is the existence of exploratory interest common and significant?

Identified from search user behavior analysis Make use of one-week log search data

Verified by Statistical Tests(Log-likehood Ratio Test) Analyze the causality between initial queries and

consequ-ent queries Results In 80.9% of cases: Clicks on search results indeed affect

the formulation of the next queries In 43.1% of cases: Users would issue different next

queries if they clicked on different results

2010/12/3

Two different heading directions of Query Recommendation Emphasize search interests:

Help users easily refine their queries and find what they

need more quickly Enhance the “search-click-leave” behavior

Focus on exploratory interests: Attract more user clicks and make search and browse

more closely integrated Increase the staying time and advertisement revenue

Recommend queries to satisfy both search and exploratory interests of users simultaneously

2010/12/3

equivalent or highly related queries

apple iphonemobilemeipod touch

nexus one

Outline

Motivation Challenges Our Approach Experimental Results Conclusions

2010/12/3

Challenges To leverage what kind of data resource?

Search logs: Interactions between search users and search engines

Social annotation data: Keywords according to the content of the pages

“wisdom of crowds”

2010/12/3

Challenges To leverage what kind of data resource? How to present such recommendations to users?

Refine queries

Stimulate exploratory interests

2010/12/3

A Structured Approach to Query Recommendation

With Social Annotation Data

Outline


2010/12/3

Approach

Query Relation Graph A one-mode graph with the nodes representing all

the unique queries and the edges capturing relationships between queries

Structured Query Recommendation Ranking using Expected Hitting Time Clustering with Modularity Labeling each cluster with social tags

2010/12/3

Query RelationGraph

Query Formulation Model

2010/12/3

Query RelationGraph

Query Formulation Model

2010/12/3

4535

)2|2(|

quVVP

354

2

312

322

)1|1(|

uqVVP

uk

qu

Vu qu

quVV kiw

jiwijP

),(

),()|(|

Query RelationGraph

Query Formulation Model Construction of Query Relation Graph

2010/12/3

321

3121

235

4312

4535

)2|2(|

quVVP

322

)1|1(|

uqVVP

3122

)1|2(|

ut VVP

21

)2|1(| tuVVP

322

*21

*62

*125

)1|2(|

qqVVP

Ranking with Hitting Time

Apply a Markov random walk on the graph Employ hitting time as a measure to rank queries The expected number of steps before node j is visited

starting from node i The hiting time T is the first time that the random walk is at

node j from the start node i:

The mean hitting time h(j|i) is the expectation of T under the condition

2010/12/3

]1[][ )|(|1

)|(|)|(|

mTPPmTP kjVV

n

kikVVijVV qqqqqq

iX 0

]|[ 0)|(|1

)|(| iXmTPmh ijVVm

ijVV qqqq

Ranking with Hitting Time

Apply a Markov random walk on the graph Employ hitting time as a measure to rank queries The expected number of steps before node j is visited

starting from node i Satisfies the following linear system

2010/12/3

Clustering with Modularity

Group the top k recommendations into clusters It is natural to apply a graph clustering approach Modularity function

Note: In a network in which edges fall between vertices without regard for the communities they belong to ,we would have

2010/12/3

kkkk

k

k

eee

eee

eee

...

............

...

...

21

22221

11211

1C

1CkC

kC

2C

2C

qi qj

qi qj

VV VVij

CV CVij

ij P

P

e

ka

2a1a

j

iji ea

tsst aae



Employ the fast unfolding algorithm to perform clustering

2010/12/3



Employ the fast unfolding algorithm to perform clustering

Label each cluster explicitly with social tagsThe expected tag distribution given a query:

The expected tag distribution under a cluster:

2010/12/3

Outline


2010/12/3

Experimental Results

Data set Query Logs: Spring 2006 Data Asset (Microsoft Research)

15 million records (from US users) sampled over one month in May, 2006 2.7 million unique queries and 4.2 million unique URLs

Social Annotation Data: Delicious data Over 167 million taggings sampled during October and November, 2008 0.83 million unique users, 57.8 unique URLs and 5.9 million unique tags

Query Relation Graph: 538, 547 query nodes

Baseline Methods BiHit: Hitting Time approach based on query logs (Mei et al.

CIKM ’08) TriList: list-based approach to query recommendation

considering both search and exploratory interests TriStrucutre: Our approach2010/12/3

Examplesof RecommendationResults Query = espn

BiHit

espn magazine

espn go

espn news

espn sports

esonsports

baseball news espn

espn mlb

sports news

espn radio

espn 103.3

espn cell phone

espn baseball

sports

mobile espn

espn hockey

TriList

espn radio

espn news

yahoo sports

nba news

cbs sportsline

espn nba

sports

espn mlb

espn sports

sporting news

scout

sportsline

sports illustrated

bill simmons

fox sports

TriStructure

[sports espn news]

espn radio

espn news

espn nba

espn mlb

espn sports

bill simmons

[sports news scores]

yahoo sports

nba news

cbs sportsline

sports

sporting news

scout

sportsline

sports illustrated

fox sports2010/12/3

Examplesof RecommendationResults Query = 24

BiHit

24 season 5

24 series

24 on fox

24 fox

fox 24

24 tv show

tv show 24

24 hour

fox television network

fox broadcasting

fox tv

fox sports net

fox sport

ktvi 2

fox five news

TriList

fox 24

kiefer sutherland

tv guide

24 tv show

24 fox

jack bauer

grey’s anatomy

24 on fox

desperate housewives

prison break

24 spoilers

abc

tv listings

fox

one tree hill

TriStructure

[tv 24 entertainment]

fox 24

kiefer sutherland

24 tv show

24 fox

jack bauer

24 on fox

24 spoilers

[tv televisions entertainment]

tv guide

abc

tv listings

fox

[tv television series]

grey’s anatomy

desperate housewives

prison break

one tree hill2010/12/3

Manual Evaluation

Comparison based on users’click behavior A label tool to simulate the real search scenario Label how likelihood the user would like to click (6-point scale) Randomly sampled 300 queries, 9 human judges

2010/12/3

Distributions of Labeled Score over Recommendations

Experimental Results (cont.)

Overall Performance non-zero label score click➡

Clicked Recommendation Number (CRN)

Clicked Recommendation Score (CRS)

Total Recommendation Score (TRS)Click Performance Comparison

2010/12/3

How Structure Helps How the structured approach affects users’ click willingness Click Entropy


The Average Click Entropy over Queries under the TriList and TriStructure Methods.2010/12/3

How Structure Helps How the structured approach affects users’ click patterns Label Score Correlation


Correlation between the Average Label Scores on Same Recommendations for Queries.2010/12/3

Outline


2010/12/3

Conclusions

Recommend queries in a structured way for better

satisfying both search and exploratory interests of users Introduce the social annotation data as an important

resource for recommendation Better satisfy users interests and significantly enhance

user’s click behavior on recommendations Future work

Trade-off between diversity and concentration Tag propagation

2010/12/3

Thanks！

2010/12/3

A Structured Approach to Query Recommendation With Social Annotation Data 童薇 2010/12/3.

Documents

Transcript of A Structured Approach to Query Recommendation With Social Annotation Data 童薇 2010/12/3.