The inherent limitation of the conventional approaches in automatic indexing lies in the fact that they compute the relevancy between index terms and documents rather indirectly or relatively. As an alternative the anlaysis of document texts seeks a means of establishing a direct relevancy of the terms. More rigorous linguistic analysis will ensure better chance of relevancy. Various semantic topologies among terms may suggest the sufficient quality for relevancy. The enhanced and guaranteed relevance will allow the high precision of retrieval. Along with this line the on going project in KAIST pursues the user oriented retrieval system that spawns still may other issues that are not c o m n in traditional perspective.