Abstract: With the rapid development of internet technology, a networking machine translation（NMT） system came into being. However, the formats of information streams on internet are characterized by variety and changeability, confronting the technology of NMT with new difficulties and challenges. This paper is set in the networking real time machine translation system, and focuses on the research on processing strategies of format information streams in Web site homepages. A practical processing method is proposed, which is based on the dynamic knowledge acquisition and context sensitive analysis. The above strategies are used to process the HTML documents in a commercial system and the result is satisfying.