命名實體識別運用於產品同義詞擴增

Using Named Entity Recognition Increases the Synonym of Products

本研究提出產品名稱機率比對法，嘗試解決命名實體識別領域中，常被忽略的同義詞辨識問題。在意見探勘領域中，正確並使用相同的用詞用語描述一項產品或服務，能有效提升意見探勘或情感分析的成果。然而，口碑或意見為使用者自行產生的內容（UGC; user generated content），口碑文章缺乏一定的撰寫規則與統一的命名規定，不同作者甚至相同作者對於相同的產品，常會產生命名不一致的現象，導致口碑敘述的產品名稱和正式產品名稱不一致。此產品名稱不一致的現象，在搜尋或整理口碑文章時，則會產生資訊遺漏的問題。本研究提出產品名稱機率比對法，透過產品共現詞彙集、Word2vec語言模型，擴增產品的同義詞，並使用機率比對方式，嘗試找到適合的產品同義詞，以解決資訊遺漏的問題。根據本研究的初步實驗顯示，本研究所提出的方法對於同義詞辨識問題，有發展的潛力。

This research proposes the probability mapping approach for a product name to attempt to resolve the synonym identification problem, which is usually ignored in the field of Named Entity Recognition. Using the same name to describe a product or service may effectively improve the results of opinion mining or sentiment analysis. However, as WOM is a user generated content (UGC), different names may be used by the same or different users. Besides, there is no unified naming rule when writing the WOM. Even though the authors are the same or different, they may use different names to describe the same products. In this case, searching or organizing the WOM article without the consideration of the naming issue may lead to the problem of information loss. Thus, we propose the probability mapping approach via the co-occurrence naming dataset and the Word2vect language model in order to reduce the naming issue. According to our initially experimental results, the probability mapping approach for a product name present its potential in the naming issue.