Classification of financial accounting concepts through the use of latent semantic indexing and clustering techniques

Information and standards overload are part of the current business environment. In accounting this is exacerbated due to the variety of users and the evolving nature of accounting language. This dissertation describes a research project that determines the feasibility of using statistical methods to automatically group related accounting concepts together. Starting with the frequencies of words in documents and modifying them for local and global weighting, Latent Semantic Indexing (LSI) and agglomerative clustering were used to derive clusters of related accounting concepts. Resultant clusters were compared to terms generated randomly and terms identified by individuals to determine if related terms are identified. A recognition test was used to determine if providing individuals with lists of terms generated automatically allowed them to identify additional relevant terms.; Results found that both clusters obtained from the weighted term-document matrix and clusters from a LSI matrix based on 50 dimensions contained significant numbers of related terms. There was no statistical difference in the number of related terms found by the methods. However, the LSI clusters contained terms that were of a lower frequency in the corpus. This finding may have significance in using cluster terms to assist in retrieval. When given a specific term and asked for related terms, providing individuals with a list of potential terms significantly increased the number of related terms they were able to identify when compared to their free recall.