Abstract: Focusing on the problem that traditional information search method can not realize contents search for specific topics directly, this paper proposes a microblog search method based on convolutional neural networks to search security topics related content from microblogs with matching and ranking. The proposed method is composed of related microblog content filter and microblog content matching, both of which are based on deep convolutional neural networks. The content filter aims at selecting security topic related microblog content according to convolutional feature representations. The content matching outputs the sequence of results which are ranked by similarities through the processes of non-linear feature transformation. The content filter and the content matching operate the non-linear feature transformation of mined local semantic features based on convolution and max-pooling calculations to achieve similarity matching. Experimental results demonstrate that the proposed method performs better compared with those of the state-of-the-art methods， and also verify the effectiveness of the proposed method on microblog content search on specific topics.

Key words: microblog search, deep convolutional neural networks, deep learning, search ranking, information search

摘要： 针对由于微博文本的数据特性造成的传统信息搜索方法无法直接实现微博话题内容搜索的问题，提出了一种基于卷积神经网络的微博话题内容搜索方法，对微博安全话题内容进行搜索和匹配排序。该方法包括基于深度卷积神经网络的微博内容筛选和微博内容匹配两部分。微博内容筛选依据深度卷积特征表示进行微博内容筛选，微博内容匹配通过卷积特征非线性变换对筛选结果进行匹配排序。微博内容筛选和微博内容匹配对国民安全话题相关的微博文本内容局部特征进行处理，对筛选结果进行相似度计算从而实现相似度匹配。实验结果表明该方法在微博搜索性能上优于现有同类方法，并验证了所提出方法针对安全话题的微博文本内容搜索的有效性。

关键词: 微博搜索, 深度卷积神经网络, 深度学习, 搜索排序, 信息搜索