Toward the automatic extraction of knowledge of usable goods

摘要

Knowledge of usable goods (e.g., toothbrush is used to clean the teeth and treadmill is used for exercise) is ubiquitous and in constant demand. This study proposes semantic labels to capture aspects of knowledge of usable goods and builds a benchmark corpus, Usable Goods Corpus, to explore this new semantic labeling task. Our human annotation experiment shows that human annotators can generally identify pieces of information of usable goods in text. Our first attempt toward the automatic identification of such knowledge shows that a model using conditional random fields approaches the human annotation (F score 73.2%). These results together suggest future directions to build a large-scale corpus and improve the automatic identification of knowledge of usable goods.