Project page zero, Smart Search, Learning to Personalize suggestions

41
Project PageZero Smart Search Antonio Gulli

description

Suggestions, Search, Learning to Rank

Transcript of Project page zero, Smart Search, Learning to Personalize suggestions

Page 1: Project page zero, Smart Search, Learning to Personalize suggestions

Project PageZeroSmart Search

Antonio Gulli

Page 2: Project page zero, Smart Search, Learning to Personalize suggestions

Search transformed the way we look at the world

Page 3: Project page zero, Smart Search, Learning to Personalize suggestions
Page 4: Project page zero, Smart Search, Learning to Personalize suggestions

Search box:Looking at the world through a window

Page 5: Project page zero, Smart Search, Learning to Personalize suggestions
Page 6: Project page zero, Smart Search, Learning to Personalize suggestions
Page 7: Project page zero, Smart Search, Learning to Personalize suggestions
Page 8: Project page zero, Smart Search, Learning to Personalize suggestions
Page 9: Project page zero, Smart Search, Learning to Personalize suggestions

Satori: Entering the World of Entities

Page 10: Project page zero, Smart Search, Learning to Personalize suggestions
Page 11: Project page zero, Smart Search, Learning to Personalize suggestions

Words are ambiguous

Page 12: Project page zero, Smart Search, Learning to Personalize suggestions
Page 13: Project page zero, Smart Search, Learning to Personalize suggestions
Page 14: Project page zero, Smart Search, Learning to Personalize suggestions

Sites

Page 15: Project page zero, Smart Search, Learning to Personalize suggestions
Page 16: Project page zero, Smart Search, Learning to Personalize suggestions
Page 17: Project page zero, Smart Search, Learning to Personalize suggestions

Bing uses the world of entities as soon as you type. Not only for refining search results

Page 18: Project page zero, Smart Search, Learning to Personalize suggestions

Smart Search

Windows 8.1World Wide

Page 19: Project page zero, Smart Search, Learning to Personalize suggestions
Page 20: Project page zero, Smart Search, Learning to Personalize suggestions
Page 21: Project page zero, Smart Search, Learning to Personalize suggestions
Page 22: Project page zero, Smart Search, Learning to Personalize suggestions
Page 23: Project page zero, Smart Search, Learning to Personalize suggestions
Page 24: Project page zero, Smart Search, Learning to Personalize suggestions
Page 25: Project page zero, Smart Search, Learning to Personalize suggestions
Page 26: Project page zero, Smart Search, Learning to Personalize suggestions

SkyDrive Xbox Web

Page 27: Project page zero, Smart Search, Learning to Personalize suggestions
Page 28: Project page zero, Smart Search, Learning to Personalize suggestions

Personalized Results Email Skype

Local Files

Web

Page 29: Project page zero, Smart Search, Learning to Personalize suggestions
Page 30: Project page zero, Smart Search, Learning to Personalize suggestions
Page 31: Project page zero, Smart Search, Learning to Personalize suggestions
Page 32: Project page zero, Smart Search, Learning to Personalize suggestions

Learning to Personalize Query Auto-Completion

Milad ShokouhiMicrosoft

Page 33: Project page zero, Smart Search, Learning to Personalize suggestions

Relevance Labelling for Contextual Search

•For learning we need labels.•Relevance labelling for contextual (personalized) search

(auto-completion) is not trivial.•Previous work on personalized search [Fox et al., 2005]• Samples search impressions from the logs•Documents with SAT clicks are annotated with relevant labels.• The goal is to learn a re-ranking model that improves the

ranking of those relevant documents given the context.

Page 34: Project page zero, Smart Search, Learning to Personalize suggestions

Analogy: Auto-Completion Labels

34

Page 35: Project page zero, Smart Search, Learning to Personalize suggestions

Experimental Settings

• Ranker: Lambda-Mart [Burges et al., 2011] • AOL testbed• 657K users (Mar-May 2006)• 128,620 queries in the prefix-tree• Userid, query, timestamp

• Bing testbed• 196K logged in users with Microsoft LiveID (Jan-2013)• 699,862 queries in the prefix-tree• Userid, query, timestamp, age, gender, zip code

• Training & testing on different sets of users

Page 36: Project page zero, Smart Search, Learning to Personalize suggestions

Personalized Ranking Features

• Demographics• Age (5 groups)• Gender (2 groups)• Zip-code (10 groups)

• Search history• Short (session)• Long (all past queries)

Page 37: Project page zero, Smart Search, Learning to Personalize suggestions

Personalization by Age

Testbed Baseline Personalized MRR (Gain/Loss)

Bing (age) - - +3.80%

The effectiveness of auto-completion personalization according to the user’s age in terms of MRR. All differences are statistically significant (P < 0.01)

Below 20 21-30 31-40 41-50 Above 50

Frequently promoted suggestions for different age groups

Page 38: Project page zero, Smart Search, Learning to Personalize suggestions

Results Summary

Features AOL BingShort history +1.95% 0.91%Long history +4.45% 5.57%Age - 3.80%Gender - +3.59%Location - +4.58%All +6.45% +9.42%

Page 39: Project page zero, Smart Search, Learning to Personalize suggestions

London

Twitter: @gulliantonio

Page 40: Project page zero, Smart Search, Learning to Personalize suggestions
Page 41: Project page zero, Smart Search, Learning to Personalize suggestions