Web Information Seeking
CS6210 survey presentation

	Guo Shuqiao
	Yang Hui
	15 Oct 2003

Outline

	Introduction
	Review on Mental Model Study
	Current Research Trends
	Hyperlink Pattern Modeling
	Web Information Seeking and Other Research Areas
	Conclusion and Perspectives

Introduction

	Information seeking
					Berry Picking
					The process engaged in by humans to change their state of knowledge
	Web Information Seeking
		Web:
			The biggest digital library available ( > 2 billion pages)
			heterogeneous collection of information resources with minimal selection, organization, and retrieval standards
		Differ from Traditional Digital Library
			No real organization
			No control of Input
			No control of customer set

Outline

	Introduction
	Review on Mental Model Study
	Current Research Trends
	Hyperlink pattern modeling
	Web Information Seeking and Other Research Areas
	Conclusion and Perspectives

Study on Mental Models

	[Slone 02] examined the influences of user’s mental model and the impact  on their searching behavior

Influencing Factors and Results

	User’s understanding of the Web
		User’s experience
		Mental model
		User’s expectation
	User’s Goal
		Situational goals
		Specific search goals
		Format goals

Outline

	Introduction
	Review on Mental Model Study
	Current Research Trends
	Hyperlink pattern modeling
	Web Information Seeking and Other Research Areas
	Conclusion and Perspectives

Current Research Trends

	Content-based approaches [Michael03]
		HTML body text
		Title and headings
		Anchor text, etc
	Link-based approaches
		Link structure infers information about pages
		Surfing behavior of users can be abstract into patterns

Link Pattern Modeling - I

	Markov chains model [Sarukkai 00]
		Create probability distribution about which of the previous links is ‘good predictors’ of the next link.

Link Pattern Modeling - I

	Markov chains model [Sarukkai 00]
		Markov chains and eigen-vector decomposition techniques
		A : matrix representing transition probabilities
		s(t): probability vector for all the states at time t

Link Pattern Modeling - II

	Longest Repeating Subsequence Model [Pitkow 99]
		Surfing paths can be represented as n-grams <X1, X2,…Xn> to indicate sequences of page clicks by a population of users visiting a web site
		Find Longest repeating subsequence
			E.g., 135 in 15135213544
		Match the performance accuracy of the one-hop Markov model while reducing the complexity by nearly 33%

Link Pattern Modeling - III

	Application of Link Pattern Modeling
		Link prediction and Prefetching
		Agent Assisted Navigation
		Web Community
		Website Organization and Optimization
		Personalization
	Limitations
		Goodness of the models depends on the amount of training data available
		Dimensionality (Markov chain matrix is typically very large)

Outline

	Introduction
	Review on Mental Model Study
	Current Research Trends
	Hyperlink pattern modeling
	Web Information Seeking and Other Research Areas
	Conclusion and Perspectives

         Related Research Areas

	Information Retrieval
		Search engines
		PageRank
		HITS
	Information Extraction
		Web Topic detection
	Data Mining
		Data warehousing
		Web log analysis

Outline

	Introduction
	Review on Mental Model Study
	Current Research Trends
	Hyperlink pattern modeling
	Web Information Seeking and Other Research Areas
	Conclusion and Perspectives

Conclusion and Perspectives

	Conclusion
		Quick Review of Web Information Seeking
		Mental Model Study
		Current Research Trends
		Relationship to Other Research Areas
	Perspectives
		Natural Language Processing
		Multimedia Interaction
		Digitalized Library Interview
		Life-long Assisted Education
		Persistence and Web Security

References

	Slone (02) The Influence of Mental Models and Goals on Search Patterns During Web Interaction, JASIST 53(13):1152-1169 (2002). CL: Z671 JASIT
	Sarukkai (00) Link prediction and path analysis using Markov chains, WWW 8.
	James Pitkow and Peter Pirolli (99) Mining Longest Repeating Subsequences to Predict WWW Surfing, USITS' 99.

Slide 18