Emotion-Age-Gender-Nationality Based Intention Understanding in Human–Robot Interaction Using Two-Layer Fuzzy Support Vector Regression
Emotion-Age-Gender-Nationality Based Intention Understanding in Human–Robot Interaction Using Two-Layer Fuzzy Support Vector Regression
复制标题
DOI：
10.1007/s12369-015-0290-2
复制
发表时间：
2015-02
影响因子：
4.7
通讯作者：
Luefeng Chen;Zhentao Liu;Min Wu;Min Ding;F. Dong;K. Hirota
中科院分区：
文献类型：
--
作者：
Luefeng Chen;Zhentao Liu;Min Wu;Min Ding;F. Dong;K. Hirota
提出了一种基于两层模糊支持向量回归的人机交互意图理解模型，该模型采用模糊c均值聚类对输入数据进行分类，意图理解主要通过情感来获得，并结合年龄、性别、国籍等身份信息.它的目的是通过了解顾客在酒吧的点餐意图，实现透明的沟通，使酒吧工作人员和顾客之间的社会关系变得顺畅。为了验证意向理解模型的适用性，本研究设计了情绪-年龄-性别-民族与订单意向的关系实验。实验结果表明，在聚类数为2/3/6的情况下，该方法分别获得了70%/72%/80%的意图理解准确率（按不同性别/年龄/民族），分别比支持向量回归（SVR）和反向传播神经网络（BPNN）高23%/26%/33%和35. 5%/37. 5%/45. 5%;在聚类数为2/3/6的情况下，该算法的计算时间分别为0.976 s/0.935 s/0.67 s，而支持向量回归算法的计算时间为1.889 s，BP神经网络的计算时间为3.505 s。此外，在开发的人机交互系统吉祥物机器人系统中进行了初步的应用实验，实验在“在酒吧喝酒”的场景中进行，结果表明，酒吧小姐机器人对客户的订单意图的理解准确率为77.8%，并获得了“满意”的满意度评价。根据初步申请，该提案正在扩展到酒吧中的订购系统，用于商业通信。
An intention understanding model based on two-layer fuzzy support vector regression is proposed in human–robot interaction, where fuzzy c-means clustering is used to classify the input data, and intention understanding is mainly obtained by emotion, with identification information such as age, gender, and nationality. It aims to realize the transparent communication by understanding customers’ order intentions at a bar, in such a way that the social relationship between bar staffs and customers becomes smooth. To demonstrate the aptness of intention understanding model, experiments are designed in term of relationship between emotion-age-gender-nationality and order intention. Results show that the proposal obtains an intention understanding accuracy of 70 %/72 %/80 % with clusters number2/3/6 (according to different genders/ages/nationalities), which is 23 %/26 %/33 % and 35.5 %/37.5 %/45.5 % higher than that of support vector regression (SVR) and back propagation neural networks (BPNN), respectively; the computational time of proposal is about 0.976 s/0.935 s/0.67 s with clusters number2/3/6, while 1.889 s for SVR and 3.505 s for BPNN. Additionally, the preliminary application experiment is performed in the developing human–robot interaction system, called mascot robot system, where the experiment is performed in a scenario of “drinking at a bar”, result shows that the bar lady robot obtains an accuracy of 77.8 % for understanding customers’ order intentions and receives a satisfaction evaluation of “satisfied”. According to the preliminary application, the proposal is being extended to an ordering system in the bar for business communication.