Related scientific information: a study on user-defined relevance.

This dissertation presents an investigation into the manifestations of relevance observed in the context of related scientific information. The main motivation is to observe if researchers, in the context of knowledge discovery, use different criteria to judge the relevance of the information presented. Additionally, the effects that discipline and research experience background may have on these manifestations are investigated. The scenario selected to carry out the observation is that of Literature Based Discovery (LBD). LBD is a trial-error interactive search strategy, developed by Swanson (1986a), which supports the finding and retrieving of complementary bodies of literature “ sets of articles that are bibliographically non-interactive yet logically connected. Research scientists from three different disciplines and research experience backgrounds are observed while they interact with an LBD system built for the purposes of this study. Their cognitive processes and interactions are recorded and analysed. To aid in the analysis of the data, the concept of relevance criteria profiles is developed. Relevance criteria profiles are a technique to count and group the expressions of relevance criteria as observed during the search sessions. These offer the possibility of aggregating the observations into group profiles as well as the ability to measure the (dis)similarities that may arise in between profiles. As relevance criteria profiles provide a global view of the criteria used to judge relevance, a complementary visualisation technique is also developed. This technique displays the relevance judgement processes, as well as the interactions, in a sequential fashion allowing the researcher to perform temporal analyses on the session data. The results show that researchers do use a variety of criteria when judging the relevance of information in the context of LBD. Moreover, individuals use these criteria in different frequencies; both discipline and research experience background seem to influence these frequencies however they may not be the only intervening factors. The observed interaction patterns suggest that researchers approach the problem in two stages: i) an initial more exploratory stage followed by ii) a more focused and engaged stage. The main contribution of this thesis is the observation of these manifestations of relevance together with the interaction patterns. The final recommendation offered is that the multi-dimensional nature of relevance in this context should be addressed when evaluating LBD systems. Additionally, it is acknowledged that certain interaction behaviours may also be used during the design and testing of such systems