The multi-layer cluster analysis specifies a hierarchical structure which consists of layers specified in advance, where each layer has clusters, and allocates brands to one of the clusters at each layer which minimizes a badness of fit measure based on the sum of squares of deviations in each cluster. The multi-layer cluster analysis is evaluated by analyzing brand switching data among eight soft drink brands practically. The result shows that the difference between two groups; one comprised of Coke and Pepsi which are non-diet cola brands, and the other comprised of all diet brands and non-diet lemon-lime brands, is most important in the brand switching for consumers. This tells that the difference between the two groups is the primary concern of the consumers in the brand switching, suggesting that a marketing strategy which focuses on the difference between the two groups is desirable in the marketing of these eight soft drink brands. This shows that the multi-layer cluster analysis is effective in data analysis practically such as analyzing brand switching data. The multiple-layer cluster analysis can represent relationships among objects of after brand switching based on brands of before brand switching. This seems useful in designing the display shelf. But the multiple-layer cluster analysis cannot directly represent asymmetric relationships among brands in brand switching, as asymmetric multidimensional scaling and asymmetric cluster analysis can.
The purpose of this study was to examine word clustering in detecting Twitter trending topics about new products based on specific sentiment or interest expressions. Thus, we collected Twitter entries about new products based on specific sentiment or interest expressions. Twitter is an online social networking and microblog service that enables its users to post and read text-based messages of up to 140 characters, known as tweets. Twitter has spread rapidly in Japan in recent years. To identify market trends, analysis of consumer tweet data has received much attention recently. It is important to consider time series variation of trending topics when we perform word clustering to detect trending topics on Twitter. Personal concerns will be influenced by new product strategies, such as marketing communication strategies, and will change as time passes. In the present study, we sought to detect time series variation in topics about new products by classifying words into clusters based on the co-occurrence of words in Twitter entries. Then, we classified the words extracted from the tweet data using non-negative matrix factorization for dimensionality reduction of the vector space model.
A novel functional clustering method for domain-dependent functional data is proposed. The functional data are defined on sequential domains and they might have particular clusters on each domain. Our approach, Moving Functional k-means Clustering, is to classify the peculiar domains based on each sequential set of clusters. A typical functional clustering with fixed number of clusters is applied to the domains, then we relabel the set of clusters to keep the previous labels as many as possible. We demonstrate our method with large scale sensing data of environmental radio activity level in Fukushima Prefecture.
The purpose of this study is to propose the expression for the skew normal distribution by specifying up to third-order moments independently.We developed the new density function which can specify mean, variance, and skewness directly by translating the parameters of skew normal distribution.This method enabled us to estimate skewness of latent variable directly and to compare distribution between groups.A simulation study indicated validity of parameter estimation, and the analysis of brand value data showed effectiveness of this method.Parameter estimation was performed via Hamiltonian Monte Carlo Method.