| The growth of digital libraries (DL) is highlighting several challenges facing the current generation of digital library interfaces. DLs are experiencing growth in three dimensions: quantity, size, and diversity of content. This makes efficient information access increasingly challenging as a result of combined issues including information overload, the vocabulary problem and user fatigue.; The next-generation digital libraries must address these challenges. This thesis presents a new approach to scientific digital libraries called vertical digital libraries (VDL). Vertical digital libraries leverage the characteristics of and benefits enabled by domain specific document collections to aid users in overcoming the problems described above. VDLs can take advantage of three subspaces of information contained in domain specific document collections that have not been fully exploited before: the document subspace, the knowledge subspace and the societal subspace.; The contributions of this thesis include the conception of vertical digital libraries, a prototype experimental VDL covering the domain of MEMS (micro electro mechanical systems), and several other techniques that become possible and practical in the context of VDLs, from each information subspace. These techniques include enhanced relevance ranking with chronological term rank (CTR), ontology learning with formal concept analysis, and analysis of collaboration community evolution. Experimental studies indicated that CTR combined with BM25 significantly out performed benchmark retrieval techniques on a variety of TREC data. Ontology learning provided a query expansion method which increased mean average precision by 6% and recall by 4% on the OHSUMED data collection. Novel network evolution tools developed include the sliding window analysis technique and normalized disconnected diameter statistic along with the identification of four stages of community evolution. |
