面向正样本和未标记样本学习的算法研究及其应用

Positive and unlabeled Learning（PU learning) is an important area in the machine learning and data mining communities. Different from the traditional supervised learning and semi-supervised learning, the training set of PU learning consists of a small set of labeled positive examples and a large number of unlabeled examples, and there may exist positive examples and negative examples in the unlabeled data. PU learning aims at constructing a classifier based on the labeled positive examples and the unlabeled examples to justify whether a test data belongs to positive class or not. This project aims at designing support vector machine-based methods to resolve the research problems in the PU learning. First, to resolve the problem of multi-task-based positive and unlabeled learning, this project proposes similarity metric-based multi-task model to help the reciprocal learning in the multi-task problem so that we can improve the learning ability of the PU learning classifier. Second, to resolve the problem of multi-view-based positive and unlabeled learning, this project puts forward the constrains inequality to restrict the hyper-planes in each view space, and then builds the objective model to obtain the PU learning classifier. Finally, to resolve the problem of user’s concept summarization learning in the positive and unlabeled data stream, this project first constructs support vector data description machine to obtain the support vectors which can describe the data distribution of the data in each time window, and then proposes class-based clustering method to summarize user’s concepts in the data stream. This project aims at delivering research ideas to the area of positive and unlabeled learning.

正样本和未标记样本学习是机器学习和数据挖掘中的重要研究领域,与传统有监督和半监督学习不同，该学习的训练集包含少量标记的正类样本和大量未标记的样本,其中可能包含未标记的正类样本和负类样本，并构造分类器预测未知样本是否属于正类。本项目拟基于支持向量机方法对正样本和未标记样本学习的问题进行研究。首先，针对面向多任务的正样本和未标记样本学习问题，本项目构建基于相似度度量的多任务模型，实现相关源任务之间的互助学习。其次，针对面向多视角的正样本和未标记样本学习问题，本项目通过构建多视角空间约束不等式，约束不同视角空间的分类平面，并建立目标方程获得分类器。最后，针对面向正样本和未标记样本数据流的个人兴趣总结问题，本项目构建支持向量描述机获得描述时间窗数据的支持向量样本，并提出基于时间窗的类聚类算法对用户个人兴趣进行总结。本项目旨在通过以上问题的解决为正样本和未标记样本学习的研究提供思路。

正样本和未标记样本学习是机器学习和数据挖掘中的重要研究领域,与传统有监督和半监督学习不同，该学习的训练集包含少量标记的正类样本和大量未标记的样本,其中可能包含未标记的正类样本和负类样本，并构造分类器预测未知样本是否属于正类。本项目拟基于支持向量机方法对正样本和未标记样本学习的问题进行研究。首先，针对面向多任务的正样本和未标记样本学习问题，本项目构建基于相似度度量的多任务模型，实现相关源任务之间的互助学习。其次，针对面向多视角的正样本和未标记样本学习问题，本项目通过构建多视角空间约束不等式，约束不同视角空间的分类平面，并建立目标方程获得分类器。最后，针对面向正样本和未标记样本数据流的个人兴趣总结问题，本项目构建支持向量描述机获得描述时间窗数据的支持向量样本，并提出基于时间窗的类聚类算法对用户个人兴趣进行总结。本项目旨在通过以上问题的解决为正样本和未标记样本学习的研究提供思路。