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：2019
卷号：2239
页码：276-278
出版社：Newswood and International Association of Engineers
摘要：We propose a new method to extract paraphrases
of inappropriate expressions using long short-term memory
(LSTM) as one of the time series deep learning methods. Inappropriate
expressions are often described indirectly. To extract
inappropriate expressions described indirectly, the meanings of
expressions must be identified. The meanings of expressions
may vary depending on the domain and context where the
expressions are used. The proposed method uses LSTM to
learn the series of responses on thread on a bulletin board
system. LSTM obtains a model for detecting responses containing
inappropriate expressions. When the model evaluates a
response as inappropriate, the method extracts words from the
response. If a word appears frequently in responses evaluated as
inappropriate, the method evaluates the word as a paraphrase
for an inappropriate expression. We conducted preliminary
experiments. It was confirmed that the method could extract
paraphrases for inappropriate expressions.
关键词：Paraphrases for inappropriate expressions;
Time series deep learning; Bulletin board system; Word vector
with distributed representation
