Comparing Web N-grams and Other Means of Identifying Named Entities in Corporate Blogs

Abstract

Identifying named entities is crucial to many text-analysis tasks. Several means of entity identification exist, but most means need to be improvised for use over informal text such as those found in corporate (internal) blogs. In this paper, we report on a comparison of four means – part of speech, capitalized phrases with local n-grams and web n-grams, and a black box service – to identifying entities in a real-life corporate blog system. Our experiments show that capitalized phrases provide the best means, and in particular, using local n-grams provide marginally better performance than using web n-grams.