计算机科学
表达式(计算机科学)
自然语言处理
编码(集合论)
人工智能
萃取(化学)
程序设计语言
色谱法
化学
集合(抽象数据类型)
作者
Ern Yi Goh,Tong Ming Lim,Chi Wee Tan,Yuen Kei Khor
标识
DOI:10.1109/icdxa61007.2024.10470474
摘要
Multiword Expression (MWE) extraction has been popular research in Natural Language Processing (NLP) area. This research aims to extract different MWE types, including Chinese (ZH), English (EN), Chinese-English (ZH-EN) code-mixed, and new word MWEs, from Malaysian Expressions (MSEs) in the e-commerce platform. MSEs are multilingual and contain code-mixed text as well as Out-Of-Vocabulary (OOV). Thus, they are highly unstructured. A comparative study is conducted using various statistical models such as Branch Entropy (BE), Normalized Branch Entropy (NBE) and Accessor Variety (AV). Each statistical model's performance is assessed using Top N precision. The study reveals the characteristics of ZH-EN code-mixed MSEs in e-commerce platforms and the contribution of punctuation towards the Top N precision of statistical models. The experiment results show that AV outperforms BE variations, whereas NBE outperforms conventional BE.
科研通智能强力驱动
Strongly Powered by AbleSci AI