Presentation on theme: "Graphical User Interface (GUI) for Information Retrieval (IR) CS5604 Wen-Hung Kuo."— Presentation transcript:
Graphical User Interface (GUI) for Information Retrieval (IR) CS5604 Wen-Hung Kuo
Main Goal of GUI for IR The main goal of a GUI, as well as Human Computer Interaction (HCI) in general, is to make a system easy to learn and easy to use. The evaluation of an IR system has traditionally ignored the user interface aspect of its design. There are, however, recent advances in HCI for IR that look promising.
Articles Used for Today’s Presentation Interactive Information Retrieval Systems: From User Centered Interface Design to Software Design by P. Mulhem and L. Nigay (1996) LyberWorld - A Visualization User Interface Supporting Fulltext Retrieval by Matthias Hemmje, Clemens Kunkel, and Alexander Willett (1994)
Evaluating HCI for IR The experimental approach for designing HCI is used in the Mulhem/Nigay paper. IR does not attempt to solve problems or answer questions, it merely helps user to find information that may be useful. A user’s situation can be described by his/her cognitive space (work-task, interest domain, cognitive state, and the problem space).
Theory of Action Model Applied to IR Combining Norman’s Thoery of Action with the cognitive space model, Mulhem/Nigay developed the Thoery of Action model for IR (Figure 1):
Theory of Action Model (Fig. 1) The semantic and articulatory distances must be traversed by the user. The goal of successful IR user interface is to reduce these distances
What Do These ‘Distances’ Mean? This distance is defined by how the input user interface (LyberWorld, InfoCrystal) allow user to evaluate the internal state of the system, and understand the concepts of the system (how to specify queries on the system) This distance is defined by the articulatory directness between the meaning of expressions and its physical form (graphical aids can be formulated to reduce cognitive effort) The studies done on visual representation of huge amounts of data are all intended to reduce this distance This distance describes how easily user can comprehend the output and understand how the system operated to obtain these results
Coupling of Input and Output Even though in Figure 1 inputs and outputs are clearly distinguished, the input and output interfaces cannot be designed independently: input/output are tightly coupled. Two ergonomic principles that show the tight coupling between input and output: immediate feedback and equal opportunity principles.
Immediate Feedback Principle Measures the rate of communication between the IRS and the user. In general, immediate response times are desirable. Partial responses are allowed in order to preserve immediate response time.
Equal Opportunity Principle Distinction between inputs and outputs are blurred. The user has the choice of what is input and what is output. Outputs are re-used by user as inputs.
Query Result in a Starfield Display Format Each star corresponds to a retrieved document The weight of the query can be adjusted via the slidebars here Any adjustments made in the slidebars will immediately reflect in the starfield Clicking on a star (retrieved document from previous query) will set the weight of the terms in the slidebars to the weight that they are defined as in the star
The Pipe-Lines Model The Theory of Action Model identifies the mental and physical actions of the USER. The Pipe-Lines Model is a system’s point of view of IRS processes. Pipe-Lines expands on the system processes that are not defined in the Theory of Action. Pipe-Lines describes what goes on inside the computer
The Pipe-Lines Model (cont.) “Figure 1” User enters keywords System identifies weight for each entered term Sends the complete “query” for internal processing Converts to internal query model readable by matching mechanism Generates a list of weighted ids Translate weighted ids into a list of weighted docs Interface receives list of weighted docs System generates output formatting information Output to display device (monitor) Shortcuts that allow information feedback
LyberWorld Prototype IR user interface. Applying metaphors of spatial navigation, which allows visualization of an abstract information space, in this case fulltext retrieval. Presence of a spatial model provides natural interaction with user, reducing cognitive costs.
What Needs to Be Visualized? Before designing an IR process for spatial navigation, an underlying conceptual model should be formulated Concepts to consider: Content of the database Interest of the user Context of a retrieval session Result of the current dialog
Spatial Concepts Explained The database’s content space contains all the information stored in the database. The context space of a session is created during navigation within the content space. The interest space of the user contains all information that are relevant the user’s needs. The result space is the context space partition relevant to the user’s interest.
Thinking Spatially is Not New The information spaces described above have been represented in various data models (document network, two level abstraction, knowledge-based network, vector-based approaches, etc.). All previous data models can be considered having ‘spatial’ properties, but they are not explicitly visualized in the resulting system’s user interface. For this paper, Hemmje/Kunkel/Willett decides to use a network representation to model relations between points in an information space
What Does a Spatial Visualization Look Like? Two visualization tools discussed in article: NavigationCones and RelevanceSpheres. Both are used in the LyberWorld prototype system.
NavigationCones Developed to visualize the retrieval dialog and the extent to which the database has been searched. NavigationCones transforms a non hierarchical content network into a hierarchical Cone Tree to improve perception of complex and abstract information spaces in a text retrieval application.
NavigationCones (cont.) Comparison between original layout and after conversion to Cone Tree layout Note some leaf nodes have been duplicated (identical nodes are linked by hidden paths) in order to retain the tree structure
RelevanceSpheres A tool that allows relevance judgment by visualizing the context space and the distance between its items. It transforms the content network into a spatial visualization displaying path lengths while preserving the numerical proportions between themselves.
RelevanceSpheres (cont.) Conversion of a content space network graph into a circular graph The term nodes now reside on the circle, while the document nodes resides at the end of paths inside the circle The closer to the border of the circle, the more relevant the document is to the query context. Also, the closer a document is positioned to a term, the more relevant it is with regard to the term. Which term is more responsible for the position of document B? Can be clarified by positioning documents inside a sphere instead of a circle (2D-3D conversion)
RelevanceSpheres (cont.) Figure 7 displays the attraction vectors defined between a document and its paths. Figure 8 displays the generation of a document’s position in the sphere with respect to the whole search path’s vectors and the sphere’s center as anchor point.
Screenshots of LyberWorld NavigationCone RelevanceSphere InformationRoom
Conclusion The main goal of GUI/HCI is to make a system easier to learn and easy to use. IRS can benefit greatly from improved HCI interfaces. The advance in GUI/HCI development for IRS such as ones in the LyberWorld prototype shows that improvements in one (HCI) can increase the user performance of the other (IR)
References LyberWorld — a visualization user interface supporting fulltext retrieval Annual ACM Conference on Research and Development in Information Retrieval Proceedings of the 17th annual international ACM SIGIR conference on Research and development in information retrieval Dublin, Ireland Pages: 249 – 259 Year of Publication: 1994 ISBN:0-387-19889-X Interactive information retrieval systems: from user centered interface design to software design Annual ACM Conference on Research and Development in Information Retrieval Proceedings of the 19th annual international ACM SIGIR conference on Research and development in information retrieval Zurich, Switzerland Pages: 326 - 334 Year of Publication: 1996 ISBN:0-89791-792-8