Entity relationship ranking using differential keyword-role affinity

Identifying relationship between named entities from a corpus of text is a well studied NLP problem. In this paper, we consider a tractable version of this wherein sample text snippets and corresponding roles are extracted and need to be ranked on relevance of text to the role. Our scoring approach uses a cumulative estimated relevance of all keywords observed in the text snippet. Relevance metrics are computed based on differential affinity of keywords to the roles observed in the training data.