科技論文英文翻譯~急 - 生活

We investigated the use of bigrams and named entities to perform content filtering.
Our domain of application was the classification of call for tenders. Our findings are that filtering a window of terms around selected bigrams works well for this kind of collection: we could obtain an increase of 10.9% of micro-F1 on our collection, and 4.1% on the 4-Universities data set.
More tests are needed in the future to verify our claim that this technique works well on web pages.