Data Harmonization for Heterogeneous Datasets in Big Data - A Conceptual Model

Data comes from machines, transactions, and social media, which is gigantic and disparate in nature. About 80 of today�s data is unstructured, while the remaining percentage is semistructured and structured. It is a big challenge for management to make efficient decisions on run time and also to store heterogeneous nature of data by existing tools. Data Harmonization can be used to solve the heterogeneity problem; the idea of data harmonization is to provide a uniform representation and remove all forms of heterogeneity from the heterogeneous datasets. In recent studies, various models have been developed for integrating, mapping, and fusion of structured and semistructured datasets, but no such model has been developed for structured, semistructured, and unstructured datasets. Information extraction is used as a vital component to extract data from different textual datasets that information formats may comprise in different file formats, i.e., Excel, JSON, and text. For developing textual data harmonization model for heterogeneous datasets, comprises of structured, semistructured, and unstructured data based on phrases similarity techniques, it needs to be first preprocessed using Natural Language Processing and its techniques like Bag of Phrases, Parts of Speech and so on. Therefore this paper focuses on the conceptual data harmonization model based on text similarity technique, which will help to blend structured, semistructured, and unstructured data. The selected phrases from heterogeneous datasets will go through training and testing using Recurrent Neural Network. © 2020, The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG.