A New Methodology for Chinese Term Extraction from Scientific Publications

Abstract

To identify Chinese technical terms, this study focuses on extracting terms from a corpus of scientific publications. The process begins with the identification of term boundaries, followed by the application of Chinese part-of-speech (POS) patterns to extract candidate terms. Features of words or characters that signal term boundaries are defined, enabling the segmentation of sentences into smaller units and facilitating the removal of irrelevant terms that may not be filtered by other approaches. POS patterns are specifically designed for the extraction of Chinese technical terms. A comparison between candidate terms extracted using these POS patterns and those obtained via n-gram models shows that the proposed POS-based method effectively eliminates a significant portion of non-relevant terms while retaining most useful ones. In the term scoring phase, a novel method based on contextual information—referred to as the Hellinger distance for context information acquisition—is introduced. This approach proves more effective than existing context-based methods. Subsequently, the Hellinger distance method is integrated with Kullback–Leibler divergence to evaluate terms along the dimensions of informativeness and phraseness. The proposed term scoring method is compared with eight alternative approaches. Results demonstrate that it outperforms others in scoring Chinese terms, particularly in the extraction of multi-word terms.