Advances in information retrieval through the use of full text databases

Information is exploding and can't be kept up with. Most will not be referenced, but, with declining hardware costs, storing all of it is cheaper than trying to predict what won't be needed. The most efficient use of human facilities is at the search and output stage. Let the user do the editing, based on a full text database.

The first and second of the four sections of this thesis form a survey-conclusions pair. The first section surveys several factors which have contributed to current possibilities in electronic publishing. But current access methods are inadequate. The second section draws some original distinctions which separate full text retrieval from other types of IR, namely bibliographic and question-answering systems.

The third and fourth sections form a similar pair. The third surveys the existing methods of processing text, while the fourth offers some techniques for full text retrieval which are a step beyond current practice. By processing relevancy feedback elicited from users, a future full text system will automatically expand and update its search vocabulary as it gets more use. The emphasis should be on making the collection more visible to the user.