Assessing the Quality of Unstructured Data: An Initial Overview

In contrast to structured data, unstructured data such as texts, speech, videos and pictures do not come with a data model that enables a computer to use them directly. Nowadays, computers can interpret the knowledge encoded in unstructured data using methods from text analytics, image recognition and speech recognition. Therefore, unstructured data are used increasingly in decision-making processes. But although decisions are commonly based on unstructured data, data quality assessment methods for unstructured data are lacking. We consider data analysis pipelines built upon two types of data consumers, human consumers that usually come at the end of the pipeline and non-human / machine consumers (e.g., natural language processing modules such as part of speech tagger and named entity recognizer) that mainly work intermediate. We define data quality of unstructured data via (1) the similarity of the input data to the data expected by these consumers of unstructured data and via (2) the similarity of the input data to the data representing the real world. We deduce data quality dimensions from the elements in analytic pipelines for unstructured data and characterize them. Finally, we propose automatically measurable indicators for assessing the quality of unstructured text data and give hints towards an implementation.