A novel word similarity measure method for IoT-enabled Healthcare applications
Dehai Zhang,Xiaoqiang Xia,Yun Yang,Po Yang,Cheng Xie,Menglong Cui,Qing Liu
DOI: https://doi.org/10.1016/j.future.2020.07.053
IF: 7.307
2021-01-01
Future Generation Computer Systems
Abstract:With the development of the Internet of Things (IoT), Natural Language Processing(NLP) has become a key part of IoT applications in Healthcare. NLP is bringing a revolutionary shift to Healthcare, powered by rapid progress of NLP analytics techniques and increasing availability of Healthcare data. Therefore, using NLP solution for IoT enable Healthcare application is an urgent and valuable task. Word similarity measurement is the basis of semantic analysis, which can be applied to translation and disambiguation of medical terms, prescription analysis, medical question and answer systems, diagnostic assistance, etc. Previous similarity measures have mainly focused on either knowledge-graph-based or word-embedding-based methods, which suffer from two problems: (1) word-embedding-based methods have difficulty discriminating words with approximately the same surrounding context; and (2) knowledge-graph-based methods do not contain multiexpression words or named entities and cannot generally converge for large-scale and updated words. To solve these two problems, this paper proposes a novel method that combines knowledge-graph-based and word-embedding-based similarity measures via word entropy. An experiment is conducted on five public datasets (R&G, M&C, WS353, WS353-Sim and SimLex). The experimental results show that the proposed method achieves significant improvements over other word similarity measures in terms of the correlation coefficient.