期刊名称：Romanian Conference on Human-Computer Interaction
印刷版ISSN：2344-1690
出版年度：2017
卷号：RoCHI 2017
页码：155-158
语种：English
出版社：Matrix ROM
摘要：This paper describes a statistical method to identify ontology components within natural language questions. The main purpose for this step is to improve question answering systems over linked data by reducing the ambiguity in the subsequent matching and query generation steps. To accomplish this task, we have trained a Conditional Random Field (CRF) classifier to label sentence tokens with the core data elements of the DBpedia ontology. The classifier was trained on a manually annotated corpus labelled with ontology elements for each token. Several features were investigated for the classifier and the results (F1=0.92) prove that this task can be successfully solved using the CRF tagger.
