Analysis of Web Survey Data based on Similarity of Fuzzy Clusters

Abstract Large amounts and varieties of data have gained the interest of various industrial companies as well as fed the ever expanding need of the public for information exchange. At the same time, internet sites that are evaluated by target audiences have increased in number, so the public can visit the targeted web sites and evaluate them based upon their firsthand opinions. Since the public rather than experts are making these assessments, there will inevitably be evaluators with views contrary to other evaluators. In order to use the information from such preference assessments of web sites effectively, it is important to consider the accuracy of the estimation of this public opinion observed through web-based surveys. Therefore, we capture the latent features of the information, categorize subjects based on their preferences, and identify the obtained latent features to the categorized clusters. We propose a method to capture this latent structure of the evaluation data as fuzzy clusters, and through the fuzzy clusters to identify the features of the various categorized subjects. In addition, using the same scales of degree of belongingness of subjects to fuzzy clusters, temporal difference over the different industries are captured through the similarity of fuzzy clusters. We show a better performance by using numerical examples.