Information Retrieval > Query Us User er Query Words Query Words Search Personalization Cont ntex ext Ranked List Ranked List Domain Dom Jaime Teevan Cont ntex ext Microsoft Research Ta Task/ sk/Use Use Cont ntex ext Personalization and Search Personalization and Search • Measuring the value of personalization • Measuring the value of personalization – Do people’s notions of relevance vary? – An example • Understanding the individual – Lots of relevant results ranked low – Best group ranking v. individual ranking – How can we model a person’s interests? • Understanding the individual • Calculating personal relevance • Calculating personal relevance – How can we use the model to measure relevance? • Other ways to personalize search • Other ways to personalize search – What other aspects can we personalize? Relevant Content Ranked Low Highly Relevant Relevant Irrelevant
Potential for Personalization Best Rankings … … … Potential for Personalization Potential for Personalization Potential for Potential for personalization personalization Overview Learning More Explicitly v. Implicitly • Measuring the value of personalization • Explicit • Understanding the individual – User shares more about query intent – User shares more about interests – Explicit v. implicit – Hard to express interests explicitly – Client ‐ side v. server ‐ side – Individual v. group Query Words • Calculating personal relevance uw admissions • Other ways to personalize search Washington or Wisconsin? Undergrad or grad?
Learning More Explicitly v. Implicitly Learning More Explicitly v. Implicitly • Explicit • Explicit – User shares more about query intent – User shares more about query intent – User shares more about interests – User shares more about interests Intellectual property? Rock climbing? – Hard to express interests explicitly – Hard to express interests explicitly Tobacco and guns • Implicit Arts Business Computers – Query context inferred Games Health Home – Profile inferred about the user Kids and Teens News Recreation Reference Regional Science – Less accurate, needs lots of data Shopping Society Sports Profile Information Profile Information Server information • Behavior ‐ based • Behavior ‐ based • Web page index – Click ‐ through – Click ‐ through • Link graph – Personal PageRank – Personal PageRank • Group behavior • Content ‐ based • Content ‐ based – Categories – Categories – Term vector – Term vector [topic: computers] � [computers: 2, microsoft: 1, click: 4, what: 3, tablet: 1] Server ‐ Side v. Client ‐ Side Profile Match Individual to Group • Server ‐ side • Can use groups of people to get more data – Pros: Access to rich Web/group information – Cons: Personal data stored by someone else • Client ‐ side – Pros: Privacy – Cons: Need to approximate Web statistics • Hybrid solutions – Server sends necessary Web statistics – Client sends some profile information to server
Match Individual to Group Overview • Can use groups of people to get more data • Measuring the value of personalization • Back off from individual � group � all • Understanding the individual • Collaborative filtering • Calculating personal relevance – Behavior ‐ based example – Content ‐ based example • Other ways to personalize search Behavior ‐ Based Relevance Behavior ‐ Based Relevance • People often want to re ‐ find • People often want to re ‐ find • People have trusted sites • People have trusted sites • Boost previously viewed URLs or domains • Boost previously viewed URLs or domains 43% 43% Behavior ‐ Based Relevance Content ‐ Based Relevance • People often want to re ‐ find • Explicit relevance feedback • People have trusted sites – Mark documents relevant – Used to re ‐ weight term frequencies • Boost previously viewed URLs or domains
Content ‐ Based Relevance Content ‐ Based Relevance • Explicit relevance feedback Score = Σ tf i * w i World N – Mark documents relevant (N) w i = log – Used to re ‐ weight term frequencies n i (n i ) • Lots of information about the user (r i +0.5)(N ‐ n i ‐ R+r i +0.5) r i w i = log – Consider read documents relevant R (n i ‐ r i +0.5)(R ‐ r i +0.5) – Use to re ‐ weight term frequencies Content ‐ Based Relevance Personalization Performance • Personalized search hard to evaluate World Score = Σ tf i * w i N • Mostly small improvements despite big gap (N) w i = log • Identify ambiguous queries n i (n i ) – Personalize: “uw” (r i +0.5)(N ‐ n i ‐ R+r i +0.5) r i w i = log R – Don’t personalize: “uw seattle library homepage” (n i ‐ r i +0.5)(R ‐ r i +0.5) • Identify easily personalized queries (r i +0.5)(N’ ‐ n’ i ‐ R+r i +0.5) – Re ‐ finding queries w i = log Client r i (n’ i ‐ r i +0.5)(R ‐ r i +0.5) R Where: N’ = N+R, n i ’ = n i +r i Other Ways to Personalize Ranking Results for Re ‐ Finding • Measuring the value of personalization • Understanding the individual • Calculating personal relevance • Other ways to personalize search – Match expectation for re ‐ finding queries – Personalized snippets
Ranking Results for Re ‐ Finding People Don’t Notice Change People Don’t Notice Change People Don’t Notice Change Snippets to Support Re ‐ Finding Snippets to Support Re ‐ Finding Query: “winery” Query: “winery” Winery ‐ Wikipedia, the free encyclopedia Winery ‐ Wikipedia, the free encyclopedia A winery is a building or property that produces wine, or a business involved in A winery is a building or property that produces wine, or a business involved in the production of wine, such as a wine company. Some wine companies own the production of wine, such as a wine company. Some wine companies own many wineries. Besides wine making equipment ... many wineries. Besides wine making equipment ... en.wikipedia.org/wiki/ Winery en.wikipedia.org/wiki/ Winery If the person has visited the page before: If the person has visited the page before: Winery ‐ Wikipedia, the free encyclopedia Winery ‐ Wikipedia, the free encyclopedia Last visit: November 14, 2007 Last visit: November 14, 2007 A winery is a building or property that produces wine, or a business involved in New content: It has been suggested that Winery wastewater be merged into the production of wine, such as a wine company. Some wine companies own this article or section. many wineries. Besides wine making equipment ... en.wikipedia.org/wiki/ Winery en.wikipedia.org/wiki/ Winery
Interest ‐ Based Snippets Interest ‐ Based Snippets Query: “winery” Query: “winery” Winery ‐ Wikipedia, the free encyclopedia Winery ‐ Wikipedia, the free encyclopedia A winery is a building or property that produces wine, or a business involved in A winery is a building or property that produces wine, or a business involved in the production of wine, such as a wine company. Some wine companies own the production of wine, such as a wine company. Some wine companies own many wineries. Besides wine making equipment ... many wineries. Besides wine making equipment ... en.wikipedia.org/wiki/ Winery en.wikipedia.org/wiki/ Winery If the person is interested in Maui: If the person is interested in Maui: Winery ‐ Wikipedia, the free encyclopedia Winery ‐ Wikipedia, the free encyclopedia A winery is a building or property that produces wine, or a business involved in A winery is a building or property that produces wine, or a business involved in the production of wine, such as a wine company… For example, in Maui there is the production of wine, such as a wine company… For example, in Maui there is a pineapple winery . … a pineapple winery . … en.wikipedia.org/wiki/ Winery en.wikipedia.org/wiki/ Winery Summary • Measuring the value of personalization – There’s a big gap between group and individual • Understanding the individual – Building a profile, explicit v. implicit • Calculating personal relevance – Relevance feedback, boost click through • Other ways to personalize search – Rank based on expectation, personalized snippets
Recommend
More recommend
