Concept Extraction Methods for Agricultural Product–Related Texts Based on Natural Language Processing Techniques: A Case Study of Online Vegetable-Product Reviews

In this paper, we evaluated adaptation problems to general natural language processing (NLP) when applied to online vegetable- product reviews. Some keywords (nouns, verbs, and adjectives) extracted from vegetable- product reviews did not adequately represent their concepts because of generally low accuracy related to chunking, synonyms, a lack of applicable target nouns, and negative concepts. We proposed concept extraction methods based on NLP to solve these problems, including (1) morpheme analysis by reference to an additional custom dictionary; (2) combination of the verb “する” with the preceding noun; (3) acquisition of negative meaning, conversion of the auxiliary verb “ぬ” to the verb “無い” or the suffix “ない”, combination of the prefix “無”, “不”, “低”, “未”, or “非” with the following noun or adjective, and a combination of the suffix “ない” with the preceding adjective or verb; (4) synonym substitution; and (5) identification of target nouns that have related adjectives.