We describe a web access log analysis of an e-commerce site as visualized by FACT-Graph with a sequential probability ratio test. Web access log analysis is an important task, and is termed as web usage mining. A variety of studies have been conducted on this subject. However, there is no known study that has focused on understanding the trends and relationships in the analysis considering the structural change of the access trends, especially in visualization fields. To solve this problem, we propose an analysis using FACT-Graph with a sequential probability ratio test. FACT-Graph has been used for the trend visualization of text mining, while the sequential probability ratio test has been used for quality control. We utilize the sequential probability test to detect structural changes and FACT-Graph for visualization by considering the session and accessed pages in the access log as articles and words in text. We can visualize data by using FACT-Graph in an experiment using 1.6 million access logs generated between July 2010 and June 2011 on the basis of 13 structural change points detected by the sequential probability ratio test.

No abstract

Abstract-This study describes a chance discovery method for network that use betweeness centrality and similarity. In prior research of chance discovery, in the chance discovery process, it is required that analysts infer chance from visualized network, because it is difficult that to solve problem like to guess the cause from the data such as non-parametric problem. However, this reasoning process has problem that chance discovery is difficult because chance discovery depends on experience or background knowledge of analysts. Therefore, to solve this problem, we pay attention the mathematical element with the network, and propose chance index that is index of network. Chance index have three calculation methods: the sum of the reciprocal, the product of the reciprocal, and the average reciprocal. Using the proposal method on three kinds of data, results show that proposal method is useful method and chance index that use average reciprocal is most useful calculation method.

This paper describes an improved chance index for chance discovery. A chance is an important event or circumstance that can be used by analysts to make decisions. Discovery chance, i.e., chance discovery, is important for knowledge to be used effectively in understanding the background and causes hidden in a dataset. However, chance discovery depends on analyst's inference. Therefore, we propose a chance index that quantitatively evaluates chance. The method is based on betweenness centrality and the strength of co-occurrence. This study improves the accuracy of chance index by considering cluster information.

SUMMARY This paper proposes chance index that estimates whether a node is chance in a co‐occurrence network. Recently, chance discovery researches are attractive for several domains. By using chance discovery, we can develop new business or predict earthquake. However, there is a problem that chance discovery requires analysts’ inference from visualized network so that success and failure of chance discovery depend on analysists. In order to solve this problem, we analyzed the features from previous chance discovery researches and build the two hypotheses: (1) chance nodes have high betweenness centrality and (2) chance nodes connect to others with weak links. Based on the hypotheses, chance index is formulated by two terms about betweenness centrality and the strength of links. We confirm the usability of chance index from verification experiments, using Bush network, questionnaire network, interview network, and editorial network.