在现实生活中，存在着大量的属性图，每个属性图既包含属性信息，又包含结构信息。随着时间的推移，一组属性图形成一个属性图序列。多属性图序列作为单属性图序列的推广，正在大量而迅速地出现。同时，区间值属性图序列作为数值属性图序列的推广，也在广泛地出现。数据所有者迫切需要挖掘隐藏在多属性图序列和区间值属性图序列的时序关联。但是，到目前为止还没有文献致力于多属性图序列或区间值属性图序列的时序关联规则挖掘研究。
本文致力于区间值属性图序列的时序关联规则挖掘及应用研究，得到以下研究成果：
（1）提出了多属性图序列时序关联规则的定义和挖掘算法
首先给出多属性图序列时序关联规则的定义，然后提出一种基于支持度的反单调性的多属性图序列时序关联规则的快速挖掘算法。提出的算法包括两个步骤，即挖掘频繁时序关联规则和验证频繁规则的置信度。算法采用新的连接和剪枝策略，具有较高的效率，而高效率是规则挖掘追求的主要目标。该成果填补了多属性图序列时序关联规则挖掘研究的空白。实验表明，提出的算法是有效的且高效。
（2）提出了区间值属性图序列时序关联规则的定义和挖掘算法
首先利用拓扑学中覆盖的概念，基于给定区间值属性图序列对应的覆盖值属性图序列给出区间值属性图序列时序关联规则的定义。然后提出一种挖掘这类规则的算法，算法分为四个步骤：将区间值属性图序列转换为延迟-项对集数据集；基于FP增长算法挖掘频繁延迟-项对集；将频繁延迟-项对集转换为规则；验证挖掘到的规则的置信度。FP增长算法的应用保证了提出的算法的效率。该成果填补了区间值属性图序列时序关联规则挖掘研究的空白。实验表明，提出的算法是有效的且高效。
（3）提出了基于时序关联规则的股票价格预测和投资组合推荐方法
提出一种基于时序关联规则的股票价格预测方法，并利用预测结果推荐投资组合。该方法分为三个步骤：建立时序事务数据库；挖掘时序关联规则；基于规则进行预测与投资组合推荐。实验表明，提出的预测方法具有较高的精确率，投资组合收益较高，风险较低。

In real life, there are a lot of attributed graphs, and each attributed graph contains both attribute information and structure information. Over time, a set of attributed graphs forms an attributed graph sequence. Multi-attributed graph sequences as a generalization of single-attributed graph sequences are appearing in large numbers and rapidly. At the same time, interval-valued attributed graph sequences as a generalization of numerical attributed graph sequences also appear widely. There is an urgent need for data owners to mine temporal associations hidden in multi-attributed graph sequences and interval-valued attributed graph sequences. However, no literature is devoted to mining temporal association rules in multi-attributed graph sequences or interval-valued attributed graph sequences.
This paper focuses on the mining and application of temporal association rules in an interval-valued attributed graph sequence and the following results are obtained:
(1) A definition and a mining algorithm of temporal association rules in a multi-attributed graph sequence are proposed.
Firstly, the definition of temporal association rules in a multi-attributed graph sequence is given, and then a fast algorithm for mining temporal association rules in a multi-attributed graph sequence is proposed based on the anti-monotonicity of support. The proposed algorithm consists of two steps, namely mining frequent temporal association rules and verifying the confidence of frequent rules. The algorithm adopts new joining and pruning strategies and has high efficiency which is the main goal of rule mining. The results fill the gap in mining temporal association rules in a multi-attributed graph sequence. Experiments show that the proposed algorithm is effective and efficient.
(2) A definition and a mining algorithm of temporal association rules in an interval-valued attributed graph sequence are proposed.
Firstly, with the concept of cover in topology, a definition of temporal association rules in an interval-valued attributed graph sequence is given based on the corresponding cover-valued attributed graph sequence of the given interval-valued attributed graph sequence. Then, an algorithm for mining this kind of rule is proposed. The algorithm is divided into four steps: converting the interval-valued attributed graph sequence into a delay-item pair set dataset, mining frequent delay-item pair sets based on the FP-growth algorithm, converting frequent delay-item pair sets to rules, and verifying the confidence of the mined rules. The application of the FP-growth algorithm ensures the efficiency of the algorithm. The results fill the gap in mining temporal association rules in an interval-valued attributed graph sequence. Experiments show that the proposed algorithm is effective and efficient.
(3) A stock price prediction and portfolio recommendation method based on temporal association rules is proposed.
A stock price prediction method based on temporal association rules is proposed, and the predicted results are used to recommend portfolios. The method is divided into three steps: establishing a temporal transactional database, mining temporal association rules, and predicting and portfolio recommendation based on rules. Experiments show that the proposed method has high accuracy, high portfolio returns, and lower risk.