Unsupervised Relation Extraction

Abstract: World Wide Web consists of vast information which is scattered across millions of web pages. We consider the problem of extracting relations from this huge data. Relations can be unary such as, creating just lists of various cities, movies, actors, etc. or binary such as all the (author, book) pairs. We want to propose an unsupervised algorithm to extract the required information from the corpus. Some small no. of seed examples can be used. Here, we are interested in finding out relationships which may be spanned over the entire length of the document. It is important to note that, our algorithm differs from the previous algorithms proposed in following aspects.
1. Rather than a sentence level or limited context, we would like to consider larger context, i.e. whole document. 2. Also, we would like to extend to higher order relations, i.e. extracting n-tuples, where, n > 2.

Keywords: Relation extraction, event extraction, pattern extraction, unsupervised