Hybrid framework of context based relation extraction for relation completion

Abstract

Nowadays there is increase in the size of data from the different sources which leads the problem of extracting the relation among related from the similar sources in many applications. The available data is already in unstructured form and having a lack of binding knowledge. There are some conventional methods of data integration, however linking the current data is beyond capabilities of such methods. To address such problems, different information extraction methods have been presented such as relation extraction (RE), named entity recognition (NER) etc. However, the relation completion (RC) is a challenging research problem while working with a framework like RE and NER. Therefore, designing an efficient RC algorithm in such a framework is the main research problem. The question answering systems are mostly depending on the Relation Extraction technique to develop the offline system for providing answers to specific questions. The Relation Extraction techniques basically used for the question answering system, it is used to find the arbitrary entity pairs which satisfy the semantic relation R. Meanwhile, RC can be perceived as a more specialized and constrained version of the RE task with the objective of matching’s two sets of given entities under a relation R. There are two recent methods for RC in RE such as PaRE and CoRE. The CoRE framework is produced efficient results for RE in context of RC, however, still, some challenges for CoRE to address such as the existing methods not supporting the many-to-many mapping as there is a poor performance for such mappings. In this research work, we attempt to design N_CoRE (Noise_Context Aware Relation Extraction) in order to support both many-to-one and many-to-many mapping.