问题解决能力是21世纪的核心技能，且与科学素养有密切的关系，因此有必要对问题解决能力进行评估。PISA 2012首次将问题解决能力纳入基于计算机的测评中，公开的log数据库中存储了学生在解决任务中的大量过程数据可供分析。对过程数据中的行为序列进行挖掘与分析，可以帮助我们了解学生在解决问题中的认知过程，提供了作答结果更多的信息，因而能更好地评估学生问题解决能力的水平。同时，数据挖掘技术的快速发展，也为我们充分挖掘log数据提供了技术上的支持。但在使用数据挖掘技术挖掘行为序列的重要意义是获得具有心理学和教育学上有意义的发现，而这是目前相关研究所欠缺的。

本研究以PISA 2012问题解决任务中的“空调” (CLIMATE CONTROL, OECD, 2014) 一题为例，采用社会网络分析方法中的加权有效网络图的方式可视化和分析学生在解决该题过程中的行为序列，探究加权有向网络能否体现VOTAT策略使用与作答表现的关系，以及比较中国上海、新加坡和美国学生的行为序列。根据VOTAT策略使用与否将过程数据中的动作编码为策略性探索行为和非策略性探索行为，前者又根据认知负荷程度细分为优先级策略性探索行为和次级策略性探索行为。

本研究发现：

首先，加权有向网络图能清晰直观地展现不同得分组学生的行为序列是不一样的。0分组学生有较频繁的非策略性探索行为，1分组学生与2分组学生的行为序列较为相似，呈现出比较明显的策略性探索行为，且2分组学生的优先级策略性探索行为更加突出。其次，中国上海、新加坡和美国学生的网络图较为相似，都以优先级策略性探索行为为主。

进一步的描述统计和卡方检验支持了上述的观察结果。0分组与1分组和2分组学生的探索行为有显著性的差异，1分组与2分组学生的探索行为不存在显著性的差异；中国上海、新加坡和美国学生的探索行为不存在显著性差异。

使用社会网络分析方法中的整体网络测量指标，如网络密度和网络中心势指标，除了得到与前面较为一致的结论以外，还得到了更多的发现。网络中心势与学生得分存在正相关，表明得分越高的学生具有更加集中的行为路径。0分组和2分组学生的网络密度相对较大，表明这两组学生的行为探索更加丰富，结合网络中心势的结果，0分组学生最后没有形成正确的探索路径，而2分组学生找到了正确的探索路径并进行了集中探索。1分组学生的网络密度是三组中最小的，这可能是其没能正确解决问题的原因之一。另外，中国上海和新加坡学生的网络密度和网络中心势指标十分接近，且都略高于美国。表明中国上海和新加坡学生有更多的行为探索，并形成了更为集中的探索路径，这一方面是三个国家学生均分差异的体现，但也可能是存在一定的文化背景的影响。

根据本研究的结论，使用加权有向网络并结合学生的认知过程和行为序列进行分析，能得到一定的心理学和教育学的意义，同时也对题目设计提供了额外的效度检验。但当前对行为序列的挖掘还不够充分，未来的研究可以加入对完整行为序随时间展开的分析，以充分了解学生的动态行为过程。另外，我们应当对于成功使用策略但作答失败的这类学生基于更多的关注与研究，发现导致这类学生没能正确作答的障碍，帮助提高这类学生的问题解决力。

Problem-solving ability is the core skill of the 21st century, and it is closely related to scientific literacy, so it is necessary to evaluate the problem-solving ability. PISA 2012 is the first time to incorporate problem-solving tasks into the computer-based assessment. A large number of process data of students in problem solving is stored in the open log database for analysis. Mining and analyzing the behavior sequence derived from the process data can help us understand the cognitive process of students in problem solving, obtain more information beside the explicit answer results, and thus can better evaluate students' ability of problem-solving. At the same time, the rapid development of data mining technology also provides technical support for us to fully mine log data. However, the significance of using data mining technology to mine behavior sequence is to obtain meaningful findings in psychology and pedagogy, which is currently lacking in relevant studies. PISA 2012 problem-solving task, “Air Conditioning”, was took as an example in this study, using weighted directed network to visualize and analyze the behavior sequence of students in the process of solving the problem, to explore whether the weighted directed network can effectively reflect the relationship between the use of VOTAT strategy and students’ performance, and to compare the behavior sequence of students in Shanghai, Singapore and the United States. According to whether the VOTAT strategy is used or not, the behavior in the process data is coded into strategic exploratory behavior and non-strategic exploratory behavior. The former is subdivided into priority strategic exploratory behavior and secondary strategic exploratory behavior according to the degree of cognitive load.

The study found that:

First of all, the weighted directed network graph can clearly and intuitively show that the behavior sequence of different groups of students is various. Score 0 students have more frequent non-strategic exploration behavior, score 1 students and score 2 students' behavior sequence is similar, showing more obvious strategic exploration behavior, and score 2 students' priority strategic exploration behavior is more prominent. Secondly, the networks of students in Shanghai, Singapore and the United States are similar, and are mainly priority strategic exploratory.

Further descriptive statistics and chi-square test support the above observations. There is a significant difference in the exploratory behavior among score 0, score 1 and score 2, but there is no significant difference between score 1 and score 2; There is no significant difference in exploratory behavior among students in Shanghai, Singapore and the United States.

Using the whole network measurement indicators of social network analysis method, such as network weighted density and network centralization, we attain more findings than the previous conclusions. There is a positive correlation between weighted density and students' scores, which indicates that students with higher scores have more concentrated behavior paths. The network density of students in score 0 and score 2 is relatively high, which indicates that the behavior exploration of these two groups is more abundant. Combined with the results of network center potential, students in score 0 did not form the correct exploration path, while students in score 2 found the correct exploration path and carried out collective exploration. The network density of score 1 students is the smallest among the three groups, which may be one of the reasons why they can't solve the problem correctly. In addition, the network density and network centrality of students in Shanghai and Singapore are very close, and both are slightly higher than those in the United States. It shows that Chinese students in Shanghai and Singapore have more behavioral exploration, and form a more concentrated exploration path. On the one hand, this is the reflection of the differences in the average score of students in the three countries, but it may also be influenced by certain cultural background.

According to the conclusion of this study, using weighted directed network combined with students' cognitive process and behavior sequence analysis, can generate some psychological and educational insights, but also provides an additional validity test for item design. However, the current mining of behavior sequence is not enough. Future research can explore the development process of behavior sequence to fully understand the dynamic behavior of students. In addition, we should pay more attention to the students who successfully use strategies but fail to answer, find the obstacles that lead to the students' failure, and help them improve their problem-solving ability.