Although recent related reviews focused on general knowledge, they were unaware of the importance of weakly supervised relation extraction. So, on the basis of science and technology information domain, this work analyzed the logical relations among semi-supervised, distant supervised and unsupervised methods by reviewing, and the resulting advantages and disadvantages were summarized. It's believed that weakly supervised learning could partially solve the scarcity of labeled data. Especially for specific technology information applications, integrated weakly supervised methods may make giant improvement. But in the near future, the main challenge for weakly supervised relation extraction is still the problem of precision.%许多相关综述工作往往侧重于通用领域的实体关系抽取,没有充分体现弱监督学习实体关系抽取的重要性.文章立足于科技情报领域,在综述大量文献的基础上,剖析半监督、远程监督和无监督 3种弱监督实体关系抽取方法的逻辑关系,并总结了各自的优缺点.基于弱监督学习的实体关系抽取方法可以部分解决标注数量不足的问题.特别是,针对不同的科技情报应用,多种弱监督实体关系抽取方法的综合运用可以取得显著效果,但是精确度偏低的问题在未来相当长的时间内仍然是弱监督学习实体关系抽取的主要挑战.
展开▼
