sigir05 if

Context-Sensitive IR using Implicit Feedback: 

Context-Sensitive IR using Implicit Feedback Xuehua Shen, Bin Tan, ChengXiang Zhai Department of Computer Science University of Illinois, Urbana-Champaign

Problem of Context-Independent Search: 

Problem of Context-Independent Search Jaguar

Put Search in Context: 

Other Context Info: Dwelling time Mouse movement Put Search in Context Apple software Hobby …

Problem Definition: 

Problem Definition … How to model and use all the information? e.g., Apple software e.g., Apple - Mac OS X The Apple Mac OS X product page. Describes features in the current version of Mac OS X, a screenshot gallery, latest software downloads, and a directory of ...

Outline: 

Outline Four contextual statistical language models Experiment design and results Summary and future work

Retrieval Model: 

Retrieval Model Qk D θQk θD Similarity Measure Results Basis: Unigram language model + KL divergence

Fixed Coefficient Interpolation (FixInt): 

Fixed Coefficient Interpolation (FixInt) Qk

Bayesian Interpolation (BayesInt): 

Bayesian Interpolation (BayesInt) Intuition: if the current query Qk is longer, we should trust Qk more

Online Bayesian Update (OnlineUp): 

Online Bayesian Update (OnlineUp) Intuition: continuous belief update about user information need

Batch Bayesian Update (BatchUp): 

Batch Bayesian Update (BatchUp) C1 C2 Intuition: clickthrough data may not decay

Data Set of Evaluation: 

Data Set of Evaluation Data collection: TREC AP88-90 Topics: 30 hard topics of TREC topics 1-150 System: search engine + RDBMS Context: Query and clickthrough history of 3 participants.

Experiment Design: 

Experiment Design Models: FixInt, BayesInt, OnlineUp and BatchUp Performance Comparison: Qk vs. Qk+HQ+HC Evaluation Metrics: MAP and Pr@20 docs

Overall Effect of Search Context: 

Overall Effect of Search Context Interaction history helps system improve retrieval accuracy BayesInt better than FixInt; BatchUp better than OnlineUp

Using Clickthrough Data Only: 

Using Clickthrough Data Only BayesInt (=0.0,=5.0)

Sensitivity of BatchUp Parameters: 

Sensitivity of BatchUp Parameters BatchUp is stable with different parameter settings Best performance is achieved when =2.0; =15.0

Summary: 

Summary Propose four contextual language models to exploit user interaction history for contextual search Construct an evaluation dataset based on TREC data (http://sifaka.cs.uiuc.edu/ir/ucair/QCHistory.zip) Experiment results show that user interaction history, especially clickthrough data, can improve the retrieval accuracy

Future Work: 

Future Work Study a general framework for interactive information retrieval Study more sophisticated models to incorporate context information Build a system on the client side to capture and exploit user context information

Slide18: 

Thank you ! The End