摘要
问答服务系统的一个重要功能是问题检索,即根据用户的提问,在已有的问答对数据中查找与用户提问相似的其他问题,将这些问题的答案直接返回给用户。问题检索任务所面临的主要困难是如何计算两个问句之间的语义相似度,提出利用链接预测模型计算问句之间的关联程度,将链接预测模型与语言模型相结合,设计出一种新的问句检索方法。通过在真实问答对数据上进行实验,表明该方法可以有效计算问句之间的语义相似度,其性能优于传统的计算方法。
In Q&A service, one of the important tasks is question retrieval. It means finding questions in the archive that are semantically similar to a user’s question that can satisfy the users’need. The main challenge here is how to measure the semantic similarity between questions. In this paper, the link-prediction models are used to measure the latent relevance between questions. A model that combines the language model and the link-prediction model is pro-posed. The experiment results on a real Q&A data set show that it is possible to measure the semantic relationship between questions, and the approach outperforms the traditional calculating approaches.
出处
《计算机工程与应用》
CSCD
2012年第10期132-136,163,共6页
Computer Engineering and Applications
基金
安徽省高等学校优秀青年人才基金项目(No.2010SQRL192)
安徽省高校自然科学研究一般项目(No.KJ2011B173)
关键词
问答系统
链接预测
信息检索
Q&A system
link prediction
information retrieval
