| Documents are not produced in a vacuum. They reflect events and circumstances in the world and in the minds of their authors. Of particular interest to the present work are the influence of three contextual dimensions on document production and perception. Temporal context refers to an author's choice of words at a particular point in time, and how these choices are influenced by previously written documents. Categorical context refers to an author's choice of category labels to assign to their completed document when presented with a restricted set created by an outside agency. Finally, network context refers to how a document is embedded in a network, that is, which documents the author chooses to link to, and which documents in turn link to their work. The networks most closely scrutinized in the present work are citation networks of scientific research documents.;The purposes of this dissertation are to explore the ways in which temporal, categorical, and network context---and the interaction of the three---affect the lexical content of documents, and to develop practical applications using knowledge gleaned from these explorations. To this end, tools and techniques for tasks in computational linguistics and network analysis that exploit contextual knowledge are developed.;First, knowledge of lexical dynamics---that is, how terms in a corpus change in frequency, meaning, and semantic contexts---is incorporated into applications for information retrieval and text categorization. Attention is then turned to the problem of predicting links within a network of documents, where knowledge of the statistical and social processes behind link creation is used. Finally, after demonstrating correlations between the three contextual dimensions, they are combined into novel applications that are relevant to keyword assignment, text database browsing, document clustering, and text categorization. |
