A new niche search engine, eBizSearch, was described. It has
found and indexed over 20,000 documents in e-Business. Its
organization around CiteSeer, was presented, along with the
efforts to make it OAI-compliant. Finally the limitations of
CiteSeer in terms of metadata availability, reliability and
portability led us to propose a SVM machine learning approach
for metadata extraction. Initial results show that this method
accurately automatically extracts, tags untagged text and has the
potential for extending the domain of tagged metadata.