Query Specific Semantic Matcher and Summarization

Abstract

Creating trained models that semantically represent the corpus and summarizing relevant portions according to the user query remains a challenging task. We propose a semantic matcher based approach for identifying the relevant sentences in the corpus pertaining to the prominent entities in the corpus and their relationships in the user query. Trained models are created from the corpus by leveraging word-embeddings and are semantically searched to retrieve top results of the user query. Top matched sentences are analyzed for coherence based on semantic chains leveraging Semantic Role Labeler and are summarized, where entity relationships are exploited. The approach is applied and tested on user query based Ticketing System where policy documents in IT industry are used as corpus and specific summaries in the form of steps to be followed are created according to the user query.