In this paper, we present a time series structure analysis method of a meeting using text data and a method for the visualization of state transitions. Our method evaluates and visualizes the convergence/divergence of the meeting in a time series using text data from the meeting. It is important to facilitate and review meetings for improving efficiency. Therefore, it is important not only to review the final agreement and conclusion in the dialogue during the meeting but also to understand the dialogue process. We introduce two indicators: freshness and representativeness. Our system expresses the status of the meeting in four quadrants (“stagnation”, “exploration”, “deepening”, and “consensus building”) corresponding to the combination of degrees of freshness and representativeness. By conducting an analysis using these indicators, we can objectively find which parts of the dialogue stagnated or advanced the discussion. In addition, it is possible to clarify the meeting process as a structure for review and facilitation, thereby improving efficiency of the meeting. Thus, we implemented a system that realizes this method. Furthermore, we applied this system to real data gathered from meetings consisting of actual multi-company members and verified its effectiveness.

中文翻译：

---

使用文本数据的会议时间序列结构分析方法和状态转换的可视化方法

在本文中，我们提出了一种使用文本数据的会议的时间序列结构分析方法和一种状态转换的可视化方法。我们的方法使用会议的文本数据在时间序列中评估和可视化会议的收敛/发散。重要的是促进和审查会议以提高效率。因此，重要的是不仅要在会议期间审查对话中的最终协议和结论，还要了解对话过程。我们引入两个指标：新鲜度和代表性。我们的系统以新鲜度和代表性的组合对应四个象限（“停滞”、“探索”、“深化”和“共识建立”）来表达会议的状态。通过使用这些指标进行分析，我们可以客观地找出对话的哪些部分停滞或推进了讨论。此外，可以将会议过程明确化为审查和促进的结构，从而提高会议的效率。因此，我们实现了一个系统来实现这种方法。此外，我们将该系统应用于从由实际多公司成员组成的会议中收集的真实数据，并验证了其有效性。