More Interfaces for Retrieval
Information Retrieval Activities Selecting a collection –Lists, overviews, wizards, automatic selection Submitting a request –Balancing expressiveness and usability –Command line, graphical, and NL interfaces Examining the response –Contextual displays for comprehension More Interfaces –Interfaces that integrate retrieval process
User Relevance Feedback Interfaces Standard Interface –List of documents with check boxes to indicate relevance –Non-marked documents are either not relevant or the user has no opinion Other options –Relevant and non-relevant checkboxes –Scale of relevance –“more like this” buttons System then may provide an updated query to the user for editing How does the system show original query results alongside new results? –Can provide a separate window/view for old relevant results
Relevance Feedback Interface
Evaluation of Relevance Feedback Compared four conditions: –Control – no relevance feedback –Opaque – selection of relevant documents –Transparent – got to see added terms –Penetrable – could select among terms suggested for addition All conditions with relevance feedback did better in terms of precision (penetrable significantly better than control) Time difference was not significant Penetrable condition required fewer feedback cycles
Information Assistants Watch “over the shoulder” of the user to infer their interests –Use machine learning techniques Syskill and Webert –Uses explicit feedback across sessions Letizia –Implicit feedback to suggest navigation choices –Data retained across sessions Need larger amounts of data and thus are most appropriate for longer-term tasks/interests
Social Recommender Systems In some retrieval tasks, preferences tend to form patterns across users –get implicit/explicit feedback from large number of users –cluster users according to similarity of feedback –when user wants recommendations, suggest other items liked by other cluster members Examples: movie and music recommender systems Amazon suggestions “people that bought X also bought Y” are based on item data rather than clustering users.
Interfaces for the Search Process Information tasks involve more than searching for keywords and retrieving lists of results More interfaces –String matching –Window management –All-in-one systems –Search history –Connection to collection
Information Work Variety of information tasks –Short-term: Facts and references What is the escape velocity? –Long-term: Analysis and synthesis How to design a space craft? For longer-term information activities the work really begins after potentially relevant materials are located.
Information Life-Cycle Location: Searching and Browsing Comprehension: Skimming and Reading Located resources must be understood to be evaluated Understanding one document may require other documents or result in further information requests Modification: Annotation & Authoring Reading results in annotation, note taking, and writing Added content influences further access Modified from work on software libraries: [Fischer, Henninger, Redmiles 1991]
Library Table as Success Model How people make use of library resources can give design goals. Characteristics of the library table: –Integration and easy differentiation of source materials and personal interpretation –Implicit and explicit expression via spatial layout and attached annotation –Patrons can collaborate using the materials on the table as a prop for their conversation Limitations: The library table and resources are shared/limited resources, so must be cleaned up after each work session.
String Matching Interfaces Generally provide access to an unranked list of results –List may or may not be shown Having prior matches can be valuable –For reformulation –For synthesis A tabbed interface could provide access to the results of the most recent string matching requests.
String Matching Interfaces
Window Management IR systems can use multiple windows for different features/views or be monolithic. Monolithic designs have advantages –Do not accidentally hide information –User does not need to manage windows Monolithic designs have disadvantages –Designer must anticipate what features will be used simultaneously –IR system is unlikely the only application the user will be accessing and monolithic designs tend to be larger and more problematic in this more general context.
Diagrams of Monolithic Interfaces
InfoGrid
Window Management Workspaces or rooms – activity spaces that include a configuration of independent windows. Problem of screen space –Evaluating documents involves skimming/partial reading –Moving between view of results and reading application and organizing application –In our studies, transitions doubled when we provided multiple displays (compared to a single display)
Switching Between Applications Configuration Prior StudyCurrent Study Desktop PC Laptop & LCD Screen Laptop & Projected Display Tablet PC & Projected Display # of Displays1222 Avg. Total Time3,3093,5543,6424,234 Avg. Time (VKB)2,3592,4532,6273,005 Avg. Time (IE)9501,1021,0151,229 Avg. # of Transitions (shifts of focus)
SuperBook
DLITE Workspace for information access –Two dimensional space with graphical objects representing queries, libraries, results –Graphical direct manipulation interface to most actions –Animation of results –Included the notion of workspaces for different kinds of tasks
DLITE
SketchTrieve Information access as an informal activity –Same notion as spatial hypertext Part of a larger system including document viewer and control panel Problem of integrating new results with older space. –Gets at the problem of collection, not just location
SketchTrieve
IR as a Temporal Process Valuable to return to prior searches Need ways to represent prior activity –History bars –Lists of prior searches –Ability to combine and modify prior searches
VISAGE History Visualization
Query History in Melvyl
Cat-a-Cone Integrates querying and browsing a large category hierarchy Shows relevant elements at all levels of hierarchy Users can choose the level of the structure with which to interact
Cat-a-Cone