Aiming at enhancing geographical road network analysis from a network science perspective, we introduce a novel problem of analyzing the road network in a city, and consider providing a new network centrality metric that could be useful for that problem. The problem addresses vehicular evacuation in urban settings. During emergency and disaster scenarios of vehicular evacuation, the shortest distance routes might not always be the most optimal. Instead, routes that are easier to traverse can be more crucial, even if they involve detours. Also, destinations for evacuation do not necessarily have to be restricted to traditional facilities; broad and well-maintained streets might also serve as suitable alternatives. We focus on the streets as the basic units of the road network to be investigated, and consider a scenario in which people efficiently move from starting intersections around their current places to designated goal streets, following the routes of easiest traversal. For the road network, we employ its topological representation, where vertices and edges correspond to streets and intersections between them, respectively. We thus represent the road network as a vertex-weighted graph, where the weight of each vertex reflects its ease of traversal. By appropriately extending the recently developed edge-centrality metric, “salience”, to this vertex-weighted graph, we construct a new network centrality metric to detect critical streets for the newly introduced problem. Using a toy model of road network and real-world urban road networks obtained from OpenStreetMap, we experimentally reveal its distinctive characteristics by comparing it with several baselines. Moreover, we demonstrate that the proposed network centrality metric can successfully find critical streets for vehicular evacuation, which are difficult to detect using baseline methods.

There has been a lot of research on using data from Wikipedia and other sources as a knowledge graph to generate questions for learning history and other subjects. These knowledge graphs consist of entities (words) and relations (links) between the entities, and the existing methods generated questions by extracting small subgraphs from the knowledge graphs and hiding target words (correct answer words). However, questions generated by existing methods can be solved with narrow knowledge, so they do not contribute to the development of panoramic ability that has been increasingly demanded in recent years. While increasing the size of the extracted subgraph enhances the panoramic of the question, if the subgraph is too large, it becomes difficult to understand and time-consuming to learn. Therefore, in this paper, our goal is to enhance the panoramic while keeping the subgraph small. Specifically, we prioritize extracting entities within the subgraph that are semantically distant from the correct answer word. Furthermore, we propose a method to add bypass links based on the inference rules to ensure that the extracted entities are connected to the correct answer word with minimal hops from the perspective of temporal and spatial panoramic knowledge. Since KGs based on Wikipedia do not represent all common knowledge, we utilize inference rules to complement the correct relations without contradictions. As a result of conducting subjective evaluation experiments with participants and objective evaluation experiments about the traversal of temporal and spatial knowledge from history subjects, it was confirmed that the proposed method can generate more panoramic and comprehensive questions in both temporal and spatial dimensions, at a similar scale to existing methods.

In this research, I conducted a network analysis to elucidate the relationship between the return characteristics of commodity futures prices and macroeconomic trends. First, I examined the time series characteristics of 19 types of commodity futures prices and classified them into four clusters using a time series clustering method based on TimeSeriesKMeans. Furthermore, I performed a network analysis using a correlation matrix based on the weekly returns of each commodity to identify commodities exhibiting central behaviors. By constructing a network and analyzing it using centrality measures (degree centrality, eigenvector centrality, betweenness centrality, and closeness centrality), I found that copper exhibited the highest centrality across all metrics. Additionally, this research showed a correlation analysis between copper futures prices and the Composite Leading Indicators (CLI) of various OECD countries. The results indicated that copper futures prices had a higher correlation with the CLI compared to other commodity prices, particularly showing significant correlation with the CLI of countries such as the United States and Germany. Considering the extensive industrial applications of copper in production activities, it can be inferred that macroeconomic conditions, including consumption activities and demand forecasts, are reflected in the futures prices of commodities centered around copper.

There have been active attempts to construct agents that can work in conjunction with large language model (LLM) and apply them to various intelligent systems. In this paper, we describe an attempt to incorporate an agent model using LLM into multi-agent social simulation (MAS). In the implementation of MAS, a problem has been how to construct a computational model to simulate human behavior in the target social environment. Building a model to extract and reproduce the individual characteristics of a wide variety of people has been difficult, including implementation costs. We propose a method to generate a wide variety of likely behavior characteristics from LLM, a kind of collective knowledge, and to realize decision making according to the surrounding environment. We construct a human flow simulation incorporating an agent based on the proposed method, and verify the validity of the behavior.