期刊名称：Lecture Notes in Engineering and Computer Science
印刷版ISSN：2078-0958
电子版ISSN：2078-0966
出版年度：2018
卷号：2231&2232
页码：178-183
出版社：Newswood and International Association of Engineers
摘要：In this study, we propose the metrics to measure
textual coherence by considering the document structure such
as section, paragraph, and sentence. We suppose that when
textual coherence is captured by considering the document
structure, it is possible to fit human intuition. In this metrics,
we first make graph structures at each document structure
base on sentence similarity. Next, we recursively aggregate
the coherence values of structures at a layer in a bottomup
manner. We produce test data employing the sentence
ordering task for individual layers. We assess the proposed
metrics employing sentence ordering tasks: discrimination and
insertion. We compare the performance of our metrics with
the conventional graph-based local coherence model. The our
metrics outperforms, in particular, at sentence ordering tasks
conducted of the section layer; it follows that the proposed
metrics works well for a structured document as containing
sections, paragraphs such as a technical paper and a product
manual.
