Extraction of Graph Information Based on Image Contents and the Use of Ontology.

A graph is an effective form of data representation used to summarize complex information. Explicit information such as the relationship between the Xand Y-axes can be easily extracted from a graph by applying human intelligence. However, implicit knowledge such as information obtained from other related concepts in an ontology also resides in the graph. As this is less accessible, automatic graph information extraction could prove beneficial to users. In this study, we proposed a novel method for extracting both explicit and implicit knowledge from graphs. This was based on our ontology that uses essential information pertaining to the graph and sentence dependency parsing. We focused on two graph types: bar graphs and two-dimensional (2D) charts. Different graph types require different extraction methods and have different extractable features. From the bar graph, we extracted axis labels, the global trend in the data, and the height of the bars. From the 2D charts, we additionally obtained local trends and regression types. The objective was to propose a method for acquiring the implicit and explicit information available in the graphs and entering this into our ontology. For evaluation purposes, we simulated an inquiry involving five questions. Accurate answers were retrieved and significant results were achieved by the shared concepts used in our ontology.