Application of a hybrid relation extraction framework for intelligent natural language processing

Abstract

When an intelligent system needs to carry out a task, it needs to understand the instructions given by the user. But natural language instructions are unstructured and cannot be resolved by a machine without processing. Hence Natural Language Processing (NLP) needs to be done by extracting relations between the words in the input sentences. As a result of this, the input gets structured in the form of relations which are then stored in the system’s knowledge base. In this domain, majorly two kinds of extraction techniques have been discovered and exploited - rule based and machine learning based. These approaches have been separately used for text classification, data mining, etc. However progress still needs to be made in the field of information extraction from human instructions. The work done here, takes both the approaches, combines them to form a hybrid algorithm and applies this to the domain of human robot interactions. The approach first uses rules and patterns to extract candidate relations. It then uses a machine learning classifier called Support Vector Machine (SVM) to learn and identify the correct relations. The algorithm is then validated against a standard text corpus taken from the RoCKIn transcriptions and the accuracy achieved is shown to be around 91%.