基于多情境数据和半监督代价敏感模型的场所个性化语义识别

发布时间：2018-05-10 12:07

本文选题：普适计算 + 场所语义识别　；参考：《浙江大学》2017年硕士论文

【摘要】：识别场所的个性化语义是普适计算的重要研究方向。现有方法的问题有:1)大多仅考虑场所访问在时间和空间上的信息,未充分利用与场所语义密切相关的情境信息;2)注重整体识别准确率,未区别对待不同类别间错误识别的代价损失;3)基于有监督学习来建立识别模型,当训练数据有限时识别性能不佳。本文提出基于多情境数据和半监督代价敏感模型的场所语义识别方法,首先,根据多情境数据特点构建用户活动等特征。然后,基于半监督集成学习框架,引入未标注样本进行迭代式多分类器协同训练。同时,通过场所在语义上的相似性构建代价矩阵,使用代价敏感学习方法优化模型的代价损失。实验结果表明,该方法在场所语义识别的准确率和代价损失上相比传统方法都具有显著优势。
[Abstract]:Identifying the personalized semantics of places is an important research direction of pervasive computing. The problems of the existing methods are: (1) most of them only consider the information that places access to in time and space, and they do not make full use of the situation information closely related to the semantics of places. (2) pay attention to the accuracy of overall recognition. Based on supervised learning, the recognition model is established, and the recognition performance is poor when the training data is limited. In this paper, a method of site semantic recognition based on multi-situation data and semi-supervised cost sensitive model is proposed. Firstly, user activities are constructed according to the characteristics of multi-situation data. Then, based on semi-supervised ensemble learning framework, unlabeled samples are introduced for iterative multi-classifier collaborative training. At the same time, the cost matrix is constructed by the semantic similarity of the place, and the cost loss of the model is optimized by using the cost-sensitive learning method. The experimental results show that the proposed method has significant advantages over the traditional methods in terms of accuracy and cost loss.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【相似文献】