基于特定领域知识的医疗问答系统信息质量预测

Information quality prediction of medical question-answering systems based on domain-specific knowledge

伴随着智能手机以及移动互联网的高速普及,健康消费者越来越倾向于随时随地地在线咨询疾病、健康信息。其中最流行的方式便是医疗问答系统,因为其作为一种典型的在线问诊平台,可以为广大健康消费者提供足不出户、高效率以及高性价比的专业医生诊断体验。然而由于缺乏有效的信息质量管控机制,当前的医疗问答系统仍然会出现医生回答质量参差不齐的状况,这不仅会误导健康消费者,而且会造成医生的重复努力,同时也导致了积累的医疗问答知识库无法被有效复用。因而,对医疗问答系统的信息质量进行自动化预测就显得迫在眉睫。为此,本文提出了一种基于特定领域知识视角、协同训练以及集成学习的医疗问答系统信息质量预测算法。通过俘获不同特定领域知识视角间的高度非线性关系,有效地挖掘出了嵌入在大量未标记医疗问答数据中的特定领域语义知识,显著地提升了信息质量的预测性能。

With the rapid adoption of smartphones and mobile internet,health consumers are increasingly inclined to consult disease and health information online anytime and anywhere. The most popular way is medical question-answering systems. As typical online inquiry platforms,they can provide health consumers with high efficiency and cost-effective professional doctor diagnosis experience without leaving home. However,due to the lack of effective information quality control mechanism,the current medical question-answering systems still present a situation in which the quality of the physicians’ answers varies greatly,which will not only mislead the health consumers,but also cause the repeated efforts of the doctors,and also lead to the accumulated medical question-answering knowledge base cannot be effectively reused. Therefore,it is extremely urgent to automatically predict the information quality of the medical question-answering systems. To this end,we propose a medical question-answering systems information quality prediction algorithm based on domain-specific knowledge views,co-training,and ensemble learning. By capturing the highly non-linear relationship between the different domain-specific knowledge views,we effectively mine domainspecific semantic knowledge embedded in a large amount of unlabeled medical question-answering data,which significantly improves the prediction performance of information quality.