Searching of brief message based on semantic context

There are a lot of brief semi-structure messages in the Internet. A semantic matching method based on contexts of concepts is proposed. Each word of space vectors representing brief messages is extended with synonyms in a controlled process according to the scores of information gains. Patterns of messages are extracted with naive Bayesian classifier. The similarity between a query and the message is measured by combining knowledge of their hierarchical structures with statistics on their actual usage in messages derived from a large corpus. The experiment result indicates the method has significantly improved the precision of retrieval systems.