LyberWorld - A 3D Graphical User Interface for Fulltext Retrieval

Keywords:

WHY VISUAL TEXT RETRIEVAL ?

Information retrieval is also executed within very large but often abstract information spaces. Users have to check a lot of cross references between data items (see e.g. [3]) to achieve insights about hidden information carried by relations between complex and often heterogeneous data structures. Therefore it is reasonable to enable IR users to acquire abstract data similarly to the way in which Scientific Visualization allows it for numerical data. The terms Information Visualization" and "Visual Information Seeking" ([3], [1]) describe a new medium for visually communicating sets of abstract information within information retrieval user interfaces. Different perspectives on data sets and correlations of different data dimensions into one display perspective (see e.g. [6]) are provided. IR research has further realized that user interfaces suffer from problems related to the user's perception and understanding of the ongoing information dialogue ([8]). The user can neither understand how the system's information offers were initiated, nor can he estimate wether his information requests can be satisfied. He is unable to judge the relevance of a result in accordance to his query with low cognitive effort. The metaphors of spatial navigation and attraction provided in the LyberWorld ([5]) prototype reduce such problems; corresponding visualization tools support the demands of a highly interactive information process. LyberWorld is a prototype developed at GMD-IPSI in the visual interaction tools (VISIT) project which aims at intuitive IR user interfaces. It concentrates on visualizations of one abstract information space: fulltext. A spatial model for visualizations and a corresponding UI design were developed for the probabilistic fulltext retrieval system INQUERY [2]. Complementary visualization tools support the exploration and relevance judgement of items in information spaces in a natural way. To achieve this, metaphors of spatial navigation and attraction are applied to a textual information space. The result strengthens the working hypotheses: Visualizing database contents with spatial properties and communicating IR by means of spatial navigation metaphors supports users in developing a mental model of db contents as a space". Their retrieval activities and the ones executed by the system's retrieval engine are perceived as navigations in this space and manipulations of its objects. We regard the presence of a spatial model in the user's mind and a system using corresponding display methods as an essential contribution towards natural interaction and reduction of cognitive costs, e.g., during query construction, orientation within the content space, relevance feedback and orientation within the retrieval context. LyberWorld's visualization tools for information retrieval tasks are introduced in the following video summary.

LYBERWORLD - AN EXAMPLE SESSION

Initiating a navigation: After initiating a search by entering a first keyword heat, associated to the user's interest, the NavigationCone is activated. It displays all documents containing the keyword heat on a red document level (document items are always color coded red). After reading the document titles, the user decides to follow the document entitled Courtyard Passive Solar Houses. He selects the document and unfolds its blue term sublevel (term items are always color coded blue). The resulting term view, displayed as a blue sublevel cone, is browsed. It contains the term solar, which is selected to continue the search with. On the next red document level, the user selects the document item Solar Architecture in Europe after browsing the titles of the displayed documents. He expands its blue term sublevel and browses the displayed terms. The user recognizes the terms thermal and house. The blue color of the house term is darker. It leads into an already explored area of the context, i.e. one of the already explored documents must be dealing with the term house. The user unfolds the document view of house and an animation directs his attention to the appropriate position where house first appeared. The document sublevel of house is then unfolded.

Experiencing search loops and boundaries of interesting areas: This level contains only a small number of documents, all colored orange instead of red. The documents are displayed in orange because they were already found earlier in the session. As all documents bear such hidden paths, it does not make sense to continue with one of them into an already searched area. The user decides to continue with the keyword thermal. The system recognizes thermal as already explored and an animation moves the user to the correct view position and rotates the appropriate keyword label to the front. The unfolded document view contains a lot of items and only a few of them are already known (orange). As the new document titles all look a bit misleading, the user stops his explorations and starts identifying the result kernel of his search context. The NavigationCones are closed and the RelevanceSphere is activated.

Experiencing Relevance and Expressing Interest: The system transfers the retrieved documents into the sphere center. They appear as a cluster of book symbols. Terms selected in the NavigationCones are displayed as blue spheres equally distributed on the surface of the RelevanceSphere. Each term attracts relevant documents from the sphere center towards its position. The user clusters the house and the solar terms on the surface of the RelevanceSphere. One can recognize a subgroup of documents following the movements of the two keyword spheres, while two other streams" of documents remain attracted by heat and thermal. To isolate the influenced group from the kernel, the size and therfore the attraction of solar and house are increased. The document cluster moves towards the two strengthened keywords on the surface. The user reduces the document density in the sphere and the cluster becomes fully isolated. The outmost document of the isolated cluster is displayed as an open book. The system regards it as the most relevant document item of the result. The user activates the InformationRoom and the document's textual content is displayed on the back wall. By opening other books of the isolated cluster the user recognizes that most of them meet the initial interest of his search. The kernel result of his information search is determined.

Complementarity of Tools: In the above-outlined example, the visualization tools have been used one after the other, but they can be changed or combined at all times. Whenever the user switches between the three tools, the context of his search is preserved and transferred into the other tool's visualization metaphor. Therefore the user is free to conduct his information search using whatever information tool he chooses.