Towards Cognitive Optimisation of a Search Engine Interface

Proceedings of the 2006 Australasian Language Technology Workshop (ALTW2006), pages 163–164. Towards Cognitive Optimisation of a Search Engine Interface Kenneth Treharne School of Informatics and Engineering Flinders University of South Australia PO Box 2100 Adelaide, 5001 treh0003@flinders.edu.au Darius Pfitzner School of Informatics and Engineering Flinders University of South Australia PO Box 2100 Adelaide, 5001 pfit0022@flinders.edu.au David M W Powers School of Informatics and Engineering Flinders University of South Australia PO Box 2100 Adelaide, 5001 David.Powers@flinders.edu.au Abstract Search engine interfaces come in a range of variations from the familiar text-based approach to the more experimental graphical systems. It is rare however that psychological or human factors research is undertaken to properly evaluate or optimize the systems, and to the extent this has been done the results have tended to contradict some of the assumptions that have driven search engine design. Our research is focussed on a model in which at least 100 hits are selected from a corpus of documents based on a set of query words and displayed graphically. Matrix manipulation techniques in the SVD/LSA family are used to identify significant dimensions and display documents according to a subset of these dimensions. The research questions we are investigating in this context relate to the computational methods (how to rescale the data), the linguistic information (how to characterize a document), and the visual attributes (which linguistic dimensions to display using which attributes). 1 Introduction Any search engine must make two kinds of fundamental decisions: how to use keywords/query words and what documents to show and how. Similarly every search engine user must decide what query words to use and then be able to interact with and vet the displayed documents. Again every web page author or designer makes decisions about what keywords, headings and link descriptions to use to describe documents or sections of documents. This paper presents one set of experiments targeted at the choice of keywords as descriptive query words (nWords1) and a second set of experiments targeted at explaining the effectiveness of the graphical options available to display and interact with the search results. 2 Term relevance and Cognitive Load There is considerable literature on visualisation techniques and a number of experimental and deployed search engines that offer a visualisation interface e.g. kartoo.com and clusty.com. However, there is little research to establish the effectiveness of such techniques or to evaluate or optimise the interface. This is surprising given the many theories and studies that target memory and processing limitations and information channel capacity, including many that build on and extend the empirical results summarized in George Miller’s famous Magical Number Seven paper (1956). It seems likely that for any search task visualization to realize optimal use of channel capacity, it should permit users to draw on their powerful and innate ability of pattern recognition. Another important aspect that has never been properly evaluated relates to the question of “which words do people use to describe a document?” Techniques like TFIDF are used in an attempt to automatically weight …

Content truncated to 500 words