Abstract: Visualization is the most effective way to unleash the insights of data. However, discovering the relevant subgroup of large multidimensional data sets and mapping it into appropriate visualizations often require large amount of sophisticated iterative analysis. In order to make it easy to obtain better visualization automatically, this paper proposes DRVisSys, a system that can produce suitable visualizations according to Canonical Correlation Analysis（CCA） based attributes relevance. Furthermore, considering that the weight of attribute is different in the real world, the system uses cascaded Hidden Markov Model（HMM） to calculate the weight of attributes, which regards weight as one of criteria of non-trivial set of attributes. Besides, the system can update the visual recommendation model based on users’ feedback in order to make the recommended visualization meet the needs of users. The experiment results show that DRVisSys can quickly analyze the data and recommend the appropriate visualization to users.

Key words: data visualization, Canonical Correlation Analysis（CCA）, non-trivial set of attributes, cascaded Hidden Markov Model(HMM), visual recommendation model

摘要： 数据可视化通常是展示数据价值最有效的方式。针对大规模复杂多维数据，对相关数据子集进行分析并将分析结果自动映射成合适的可视化展现模式，是一项需要大量迭代计算的复杂技术工作。设计并实现了DRVisSys系统，该系统根据属性关联分析技术推荐出合适的可视化展现模式；其对于非平凡属性组合的选择，采用典型关联算法计算出更优的属性集。考虑到各属性权重在实际生活中是有区别的，采用层叠隐马尔可夫算法计算各属性权重，将属性权重作为非平凡属性组的评测标准之一。为使得推荐出的可视化展现模式能更好地满足用户需要，DRVisSys系统能根据用户反馈，更新可视化推荐模型。实验结果表明，DRVisSys能够快速进行数据分析并为用户推荐出合适的可视化展现模式。

关键词: 数据可视化, 典型关联算法, 非平凡属性组, 层叠隐马尔可夫模型, 可视化推荐模型