A Knowledge-Based Approach to Indexing Scientific Text

We are developing a system which scans articles from scientific literature for the purpose of indexing the text. That is, the system should assist in the rapid determination of the key topics and content of articles in a particular domain and in the production of brief phrases describing the content. The number of correct concepts generated should be 80% of the concepts present (a recall rate of 80%), as compared to the output of human document analysts processing the same material.