ConceptNet 5.5: An Open Multilingual Graph of General Knowledge
Robyn Speer Luminoso Technologies, Inc. 675 Massachusetts Avenue Cambridge, MA 02139、Joshua Chin Union College 807 Union St. Schenectady, NY 12308、Catherine Havasi Luminoso Technologies, Inc. 675 Massachusetts Avenue Cambridge, MA 02139
Abstract
Machine learning about language can be improved by supplying it with specific knowledge and sources of external information. We present here a new version of the linked open data resource ConceptNet that is particularly well suited to be used with modern NLP techniques such as word embeddings.
中文速览
用知识图谱辅助机器学习理解自然语言，一直面临"知识来源单一、词向量语义不足"的瓶颈。研究者发布了新版知识图谱ConceptNet 5.5，它汇集了众包常识、维基词典、WordNet等多个来源，涵盖超过2100万条边、83种语言，并专门优化了与现代词向量（word embeddings）技术的兼容性。在此基础上，他们将ConceptNet与从大规模文本中学到的分布式语义词向量（如word2vec、GloVe）通过"retrofitting"方法融合，得到名为ConceptNet Numberbatch的混合词向量空间。该系统在词语相关性等内在评测及SAT类比题等下游任务上均达到当时最优水平，类比题准确率56.1%接近人类平均水平，证明将结构化常识知识与统计语言模型结合能显著提升NLP应用的语义理解能力。
原文 arXiv:1612.03975;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1612.03975v2