人-服务机器人交互中自然语言理解研究

摘要

Aiming at unconstrained natural language instructions for interacting with a service robot, a classification and understand-ing method based on maximum entropy is proposed. Firstly, a series of effective robot control commands are designed by collecting corpus on home service robot control. And then, the corpus is tagged by corresponding control instruction, so that natural language understanding is transformed into classification problem. Secondly, through extracting textual features and adding weights, a classi-fier based on maximum entropy is established to solve this problem. In the experiment, a multi-fold cross validation is performed on the established corpus, and it shows that the instruction matching accuracy is more than 90%.%针对与服务机器人交互过程中不受限的自然语言指令，提出了一种基于语料库的最大熵分类及理解算法。首先，通过收集家庭服务领域内的机器人控制语料，设计了一套有效的机器人控制指令体系。然后，对语料库进行对应控制指令的类别标注，从而把自然语言指令理解问题转化为分类问题。其次，通过对库中文本特征的提取和附加权重，建立最大熵分类器求解分类问题。最后，通过在所建控制指令语料库上进行多重交叉训练测试，控制指令理解的准确率可以达到90%以上。