基于概率距离的局部线性嵌入在上下文感知中的应用

APPLYING PROBABILISTIC DISTANCE-BASED LOCAL LINEAR EMBEDDING IN CONTEXT AWARENESS

上下文感知是近几年来研究的热点,主要采用机器学习的算法来进行推理。原始的LLE(Locally linear embedding)算法只能对单个流形进行采样处理,但是不能处理多流形的情况,不能得到正确的邻域。针对这一点对LLE算法进行改进,得到PLLE算法(Probabilistic LLE),并将改进的算法用UCI数据集进行验证。通过实验证明,该方法的分类效果较LLE算法、ISOMAP算法、PCA算法和KNN算法在一定的数据集上要好一些;最后将PLLE算法运用的上下文感知中,可以发现,PLLE算法能够得出较完整的上下文信息,比LLE算法要好。

Context awareness is the research focus in recent years, it mainly uses machine learning algorithm to infer. Original LLE （lo- cally linear embedding） algorithm can only sample and process single manifold, but can＇t deal with multiple manifolds, nor obtain correct neighbourhood. Aiming at this point, we improve LLE to from the PLLE （ probabilistic LLE）, and verify the improved one with UCI datasets. Through the result it proves that the classification effects of PLLE on some certain datasets are better than the algorithms of LLE, ISOMAP, PCA and KNN. Finally, the PLLE algorithm is applied to context awareness, it is found that the PLLE algorithm can get quite complete con- text information, in this raged it is better than LLE.