面向脑电数据的知识建模和情感识别

EEG-data-oriented knowledge modeling and emotion recognition

心理科学研究依赖于对生理、心理数据的分析,情感是心理研究的重要内容.近年来随着认知神经科学研究技术的成熟,研究者利用脑电(electroencephalogram,EEG)等可以反映脑功能活动的生理信号,直接研究情感问题,如情感识别、情绪脑等.但是,生理信号将会产生TB级甚至PB级的数据量,认知研究和临床神经科学在过去几十年中已产生大量生理数据,对这些大数据的表示和情感知识挖掘需要更高级的工具.构建能够表示数据含义和情感相关知识的模型,能够给心理研究者提供一个知识共享平台,以便使用这些大数据进行情感方面的科学研究.本文构建一个可以表示EEG数据语义和被试者上下文信息的本体模型,并基于该模型使用推理引擎进行基于EEG生理信号数据的自动情感识别.实验结果表明,模型在e NTERFACE 2006数据集上能够以99.11%的平均准确率识别被试者的情感状态,并从实验结果分析发现基于EEG数据情感识别最关键的特征是Beta波与Theta波的绝对功率比.

Psychological research relies on physical and psychological data, and emotion has always been an important subject in the field of psychology. Recently, with the development of cognitive neuroscience technology, researchers can study topics such as emotion recognition and the emotional brain using electroencephalogram （EEG） and other physiological signals that reflect brain activity. These physiological signals can generate terabytes or even petabytes of data, and in fact, cognitive research and clinical neuroscience has already accumulated a wealth of data over the past several decades. Knowledge representation of this large amount of data and mining it for information regarding emotion requires more advanced tools. Making a data model that can clearly represent the meaning of data associated with emotion information would create a knowledge-sharing platform for psychological researchers to access the vast amount of data for further scientific research related to emotion. This paper provides such an ontology model that represents the semantics of EEG data with contextual information about the subjects. We used the model in conjunction with a reasoning engine to perform automatic emotion recognition based on EEG signals. Experimental results show that the ontology model reaches an average accuracy of 99.11% in identifying the emotional state of the subjects. Analysis of the results suggests that the most critical characteristic of EEG-based emotion recognition is the absolute power ratio between beta and theta waves.