interactive information seeking via selective application of contextual knowledge
DESCRIPTION
Exploratory search is a difficult activity that requires iterative interaction. This iterative process helps the searcher to understand and to refine the information need. It also generates a rich set of data that can be used effectively to reflect on what has been found (and found useful). While traditional information retrieval systems have focused on organizing the data that was retrieved, in this paper, we describe a systematic approach to organizing the metadata generated during the search process. We describe a framework for unifying transitions among various stages of exploratory search, and show how context from one stage can be applied to the next. The framework can be used both to describe existing information seeking interactions, and as a means of generating novel ones. We illustrate the framework with examples from a session-based exploratory search system prototype.TRANSCRIPT
Interactive Information Seeking via
Selective Application of Contextual Knowledge
Gene Golovchinsky and Jeremy PickensFX Palo Alto Laboratory, Inc.
First, some context …
We are concerned with how people can express their information needs to support their exploratory searching activities
In the grand scheme of things, we are interested in the microscopic level
On the other hand, the devil is in the details
Exploratory Search
Uncertain ASK (Belkin, 1980)
Recall-oriented
Cognitively challenging
Iterative “Queries consisting of facets with several … specific terms have more
discriminatory power than queries consisting of an intersection of facets…” (Vakkari, 2001)
Longitudinal Information needs can persist over long periods of time
Opportunistic Bates (1989)
Evolution of Search InteractionStart
Formulate query
Run query
Examine results
Save some docs
Stop
Examples:
Early batch retrieval systemsModern web search
Evolution of Search InteractionStart
Formulate query
Run query
Examine results
Save some docs
Done?
Stop
N
Y
Evolution of Search InteractionStart
Formulate query
Run query
Examine results
Save some docs
Done?
Stop
Revisitquery?
N
Y
N
Y
Evolution of Search InteractionStart
Formulate query
Run query
Examine results
Done?
Stop
Revisitquery?
Save some docsRelevancefeedback
N
Y
N
Y
Evolution of Search InteractionStart
Formulate query
Run query
Examine results
Done?
Stop
Revisitquery?
Save some docsRelevancefeedback
Select, fuse, filter, pivot
N
Y
N
Y
N
Trajectories
Stateless interaction Single loop Input: terms
Single user
Stateful interaction Multiple loops Input: terms,
queries, documents Multiple cooperating
or collaborating users
As we add more interaction possibilities, how do we manage interaction complexity?
Transitions
Interactive IR can be represented as transitions among information objects
Operations Keyword search Relevance feedback Query composition Recommendation Etc.
Information Objects Terms Queries Documents, Etc.
NN
Start
Formulate query
Run query
Examine results
Done?
Stop
Revisitquery?
Save some docsRelevancefeedback
Select, fuse, filter, pivot
Y
N
Y
From/to Document Document set Query
Document Link following“More like this” relevance feedback
Query expansion
Document set
Personalized recommendation
Relevance feedback Query expansion
Query“I am feeling lucky” Traditional IR, pseudo-
relevance feedbackTerm suggestion Query substitution
Information Retrieval as Transitions
More examples in the paper
So what did the title mean?
Contextual Knowledge Context can be propagated across
transitions
Selective Application Information objects can be combined and
re-combined to explore info space
Context
Iterative use of search tools accrues contextual information Existing systems may lose context
How can we use context to improve interaction? Existing systems may hide context
How do we preserve transparency of context?
No Grand Unified Theory of Context Collection of best-practices examples Some possible generalizations
It’s our GUT feeling
How to support exploration in exploratory search?
Traditional queries Relevance feedback
Exploratory combination of… Documents Queries
Reflection on the retrieval process
System Walkthrough
We’ve created an interactive search system Implements several transitions Preserves context when possible Collaborative use is baked in SACK is now called Querium UI improvements over what’s in the paper Still work in progress
Retrieval history, up close
Each bar represents a ranked result from a query Taller means more relevant
Query history progresses to the right Color represents person who found it Gaps mean that document was
not retrieved Clicking on a bar pivots to that
document in its retrieval context
Run a query…
Judge top 10 documents…
Run a second query & judge docs
Run a third query & judge docs
Run relevance feedback query…
Deselect two queries…
Examine history, find more relevant docs
Filter to see unread documents
Some generalizations
Don’t require users to remember, retype, reselect Map terms and metadata across transitions Keep search displays and documents in view
Finding aids and found objects
Provide visualizations of process, not just of data Histories Overviews Paths Collaborative traces
Facilitate exploration through Pivoting among information objects Recombination Arbitrary transitions Reversible operations
What we’ve learned
Process metadata can enhance interaction A record of users’ actions can be used to
improve the interface and the algorithms
Understanding the past can help explain the present and predict the future Was the current query effective? Am I seeing any new documents?
More to come
User-mediated query expansion Allow searchers to select expansion terms
Highlighting terms in the results Combine terms from multiple queries that retrieved a given
document
Visualize query history Provide overview of overlap among queries
Chronological history Allow searchers to explore history of actions
Integrating collaboration Awareness, communication
Questions?
An Exploration of Interaction
Start with traditional querying
Show results of subsequent queries in context of earlier ones Interactive document history visualization
Identify “saved” documents as they are re-retrieved
Perform relevance feedback on “saved” documents
Show fused results from all queries Pivot by sorting and/or filtering
De-select two queries and re-visit fused results
Transitions
Information seeking interaction can be characterized by transitions between states States can be different objects States can be different views
While transitions connect states, they also impose contextual barriers One challenge to interaction is how to
preserve context across these transitions
Some examples of transitions
From a document To document(s): highlight terms or related passages To query: query expansion To terms/metadata: “back of the book” index
From a query To document(s): Query-biased summaries To terms/metadata: faceted filtering/browsing To queries: related queries, query clustering
From a document set To document(s): aggregated highlighting, recommendation explanation To a query: relevance feedback To terms/metadata: “back of the book” index
More examples in the paper