A neural algorithm for document clustering

摘要

A difficulty in implementing document clustering using algorithms based on sequential architectures is that a computational bottleneck arises eventually in the classification of documents. Neural networks have the potential to alleviate this problem. This paper reviews the fundamentals of a framework for describing neural nets. Next, the MacLeod algorithm, a neural network algorithm designed specifically for document clustering is presented. The features of this algorithm are examined. Experimental results from two small test collections are reported. Based on these results the algorithm exhibits effectiveness comparable to hierarchic (sequential) clustering algorithms. The MacLeod algorithm also appears to require time and space complexities of O(n2) and 0(n), respectively. Experimental results show that the algorithm's performance is order independent.