A conception‐based approach to automatic subject term assignment for scientific journal articles

This study proposes a conception‐based approach to automatic subject term assignment when using Text Classification (TC) techniques. From the perspective of conceptual and theoretical views of subject indexing, this study identifies three conception‐based approaches, Domain‐Oriented, Document‐Oriented, and Content‐Oriented, in conjunction with eight semantic sources in typical scientific journal articles. Based on the identification of semantic sources and conception‐based approaches, the experiment explores the significance of individual semantic sources and conception‐based approaches for the effectiveness of subject term assignment. The results of the experiment demonstrate that some semantic sources and conception‐based approaches are better performers than the full text‐based approach which has been dominant in TC fields. In fact, this study indicates that subject terms are better assigned by TC techniques when the indexing conceptions are considered in conjunction with semantic sources.