Machine learning based text analysis for intelligence collation

This paper presents the initial findings of an investigation into automated support for fusing textual information and non-text data from different sources within the military domain. Our aim is to develop a prototype fusion capability that can categorize, fuse and present intelligence information to military officers. To date, our focus has been on text information categorization using both existing text classification techniques (e.g. Weighted Feature Vector (WFV) classification) and machine learning algorithms based on inductive logic programming (ILP) and natural language processing techniques. The algorithms have been used to automatically assign documents to a pre-existing set of categories by correlating text within the documents to text relating to classifications/categories. Once text documents are categorized according to their content, then fusion can begin. The results of initial experiments indicate that the ILP approach performs at least as well as the WFV technique, and outperforms it in one set of experiments. Both techniques have the scope for further improvements which we outline in this paper.