Using Semantic Graph for Automatic Key Concept Extraction

Using Semantic Graph for Automatic Key Concept Extraction

Abstract

Human is showered with enormous data and information. Finding the required information among them is difficult and time-consuming. Extracting Key concepts or Key phrases would assist the searcher to find the wanted information as soon as possible. In this study, a new approach is proposed regarding the Key Concept Extraction (KCE) through FrameNet lexical database. This approach is based on the natural language processing methods. The FrameNet is first applied for shallow semantic parsing of the original texts and then to construct the semantic graphs. The nodes of this graph are frames with edges of frame-to-frame relations. The concepts are weighted through the semantic graph. If the weight of concept is more than that of the threshold, it is extracted as a Key concept. The two types of the Zipfian distribution based and normal distribution based thresholds are applied here. The first outperforms the last. The human-based subjective assessment is run here.

References