Takayasu Fushimi

Functional cluster extraction from large spatial networks

We address a problem of extracting functionally similar regions in urban streets regarded as spatial networks. Such characteristics of regions will play important roles for developing and planning city promotion, travel tours and so on, as well as understanding and improving the usage of urban streets. In order to analyze such functionally similar regions, we propose an acceleration method of the FCE (functional cluster extraction) algorithm equipped with the lazy evaluation and pivot pruning techniques, which enables to efficiently deal with several large-scale networks. In our experiments using urban streets of six cities, we show that our proposed method achieved a reasonably high acceleration performance. Then, we show that functional cluster produced by our method are useful for understanding the properties of areas in a series of visualization results.

Constructing and visualizing topic forests for text streams

A great deal of such texts as news and blog articles, web pages, and scientific literature are posted on the web as time goes by, and are generally called time-series documents or text streams. For each document, some strongly or weakly relevant texts exist. Although such relevance is represented as citations among scientific literatures, trackback among blog articles, hyperlinks among Wikipedia articles or web pages and so on, the relevance among news articles is not always clearly specified. One easy way to build a similarity network is by calculating the similarity among news articles and making links among similar articles; however, adding information about the posted times of articles to a similarity network is difficult. To overcome this problem, we propose a framework that consists of two parts: 1) tree structures called Topic Forests and 2) their visualization. Topic Forests are constructed by semantically and temporally linking cohesive texts while preserving their posted order. We provide effective access for users to text streams by embedding Topic Forests over the polar coordinates with a technique called Polar Coordinate Embedding. From experimental evaluations using the actual text streams of news articles, we confirm that Topic Forests semantically and temporally maintain cohesiveness, and Polar Coordinate Embedding achieves effective accessibility.

Fast Extraction Method of Functional Clusters from Large-Scale Spatial Networks Based on Transfer Learning

In this paper, we treat the road network of each city as a network and attempt to accelerate extracting functional clusters which means areas that perform similar functions in road network. As a method of extracting a group of nodes having similar functions from the network, we have proposed Functional Cluster Extraction method. In this method, high dimensional vectors based on random walks are clustered by the greedy solution of the K-medoids method, and K functional clusters are extracted. However, it is difficult to hold a similarity matrix of all node pairs for a large network with a large number of nodes like a road network. On the other hand, it has been discovered that the structure of the road network has a similar structure even if the area is different. In this paper, we propose a fast clustering method by extracting approximate medoids from the target network, using the medoid set of networks already clustered, and execute the Voronoi tessellation based on them. Using the actual road network, we evaluate the proposed method from the viewpoint of the correct answer rate (accuracy) and the calculation speed of the approximate solution.

Extraction method of typical purchase patterns based on motif analysis of directed graphs

As online stores continue to expand their businesses, a huge number of customer purchase data can be obtained. When users purchase a product, almost all users tend to post reviews on it. Therefore review data can be treated as propinquity of purchase data. In this paper, we propose a novel method that extracts typical purchase patterns based on motif analysis of a directed graph constructed from review history data. We first construct a directed graph called a purchase history graph (PHG), where a node stands for an item and a directional edge is added between successively purchased two items in chronological order. Second we decompose all of the item nodes of PHG into weakly connected components (WCCs) and expect that each WCC consists of items that are sold by the same store. For each WCC, to extract frequently appearing local edge structures, we enumerate the number of 3-node motif patterns, which is a well-known notion in complex network science. These only express theoretic patterns; the actual typical ones are slightly more complicated. Thus, we construct motif vectors, which stand for how many individual patterns are contained in each WCC. Finally, we divide all of the WCCs into K clusters based on the similarity of the motif vectors. In the above procedure, we extract the typical purchase patterns of users. From our experimental evaluation using real review dataset, we confirm the validity of each step of our proposed method and discuss the results obtained from it.

Dynamic Visualization of Citation Networks and Detection of Influential Node Addition

In this paper, to effectively visualize the browsing order of scientific articles, we propose a visualization method for citation networks focusing on the directed acyclic graph (DAG) structure. In our method, all article nodes are embedded into polar coordinate plane, where angular and radial coordinates express the citation relations and order relations among articles, respectively. Furthermore, the proposed method is equipped with a dynamic property to update coordinates of all nodes at low cost when a new article node and citation links are added to the citation network. From experimental evaluations using real citation networks, we confirm that our method explicitly reflects citation relations and browsing order compared with existing methods. Furthermore, focusing on changes in visualization results when new nodes and links are added to the citation network, our method can detect influential node and links addition by angular displacement of each node.

Improving approximate extraction of functional similar regions from large-scale spatial networks based on greedy selection of representative nodes of different areas

Dividing a geographical region into some subregions with common characteristics is an important research topic, and has been studied in many research fields such as urban planning and transportation planning. In this paper, by network analysis approach, we attempt to extract functionally similar regions, each of which consists of functionally similar nodes of a road network.For this purpose, we previously proposed the Functional Cluster Extraction method, which takes a large amount of computation time to output clustering results because it treats too many high-dimensional vectors. To overcome this difficulty, we also previously proposed a transfer learning-based clustering method that selects approximate medoids from the target network using the K medoids of a previously clustered network and divides all the nodes into K clusters. If we select an appropriate network with similar structural characteristics, this method produces highly accurate clustering results. However it is difficult to preliminarily know which network is appropriate. In this paper, we extend this method to ensure accuracy using the K medoids of multiple networks rather than a specific network. Using actual urban streets, we evaluate our proposed method from the viewpoint of the improvement degree of clustering accuracy and computation time.

Accelerating Greedy K-Medoids Clustering Algorithm with \(L_1\) Distance by Pivot Generation

With the explosive increase of multimedia objects represented as high-dimensional vectors, clustering techniques for these objects have received much attention in recent years. However, clustering methods usually require a large amount of computational cost when calculating the distances between these objects. In this paper, for accelerating the greedy K-medoids clustering algorithm with \(L_1\) distance, we propose a new method consisting of the fast first medoid selection, lazy evaluation, and pivot pruning techniques, where the efficiency of the pivot construction is enhanced by our new pivot generation method called PGM2. In our experiments using real image datasets where each object is represented as a high-dimensional vector and \(L_1\) distance is recommended as their dissimilarity, we show that our proposed method achieved a reasonably high acceleration performance.

Category reformation using purchase logs

In item searches on shopping sites, customers must find the items they want from among many products. As one method of narrowing down products, the customer currently selects a category and posts queries related to the item in it. However, in such narrowing-down method searches for products, the customer needs to know the category to narrow down the shopping sites. In this paper, after a customer posts a query related to a product, the shopping site recommends a word that is closely related to a new category generated from a query log, and we propose a new search method to combine category and keyword searches. Our evaluation experiment using actual data shows that the categories proposed by recommendations provide sufficient information quantity.

Clustering and Visualizing Functionally Similar Regions in Large-Scale Spatial Networks

We address the problem of extracting functionally similar regions in urban streets and regard such regions as spatial networks. For this purpose, based on our previous algorithm called the FCE method that extracted functional clusters for each network, we propose a new method that efficiently deals with several large-scale networks by accelerating our previous algorithm using lazy evaluation and pivot pruning techniques. Then we present our new techniques for simultaneously comparing the extracted functional clusters of several networks and an effective way of visualizing these clusters by focusing on the fact that the maximum degree of the nodes in spatial networks is restricted to relatively small numbers. In our experiments using urban streets extracted from the OpenStreetMap data of four worldwide cities, we show that our proposed method achieved a reasonably high acceleration performance. Then we show that the functional clusters extracted by it are useful for understanding the properties of areas in a series of visualization results and empirically confirm that our results are substantially different from those obtained by representative centrality measures. These region characteristics will play important roles for developing and planning city promotion and travel tours as well as understanding and improving the usage of urban streets.

Comparison of influence measures on structural changes focused on node functions

The structures of some real-world networks are dynamic in nature as time goes by. These changes consist of the addition or deletion of nodes or links and the rewiring of links. Even if link rewiring occurs, the influence degree tends to differ depending on the location in which it occurs, the nature of the nodes, and so forth. In this paper, by quantifying the influence degree of each node, we attempt to extract the influential structural changes wherein each node in a large population changes its function. Concretely, we define the node function as the PageRank convergence curve of the node and the influence degree affecting the node as distance based on a correlation coefficient of convergence curves before and after change occurs. We then propose the Structural Change Influence Measure (SCIM), which is the average value of the influence degree of all nodes. Based on experimental evaluation using several synthetic and real networks, we found five promising properties of our proposed measure. Our method indicates a higher value for changes in: 1) number of link rewirings; 2) concentrated link deleting; 3) link addition between distant nodes; 4) link addition between important and unimportant nodes; and 5) link deletion between communities.