The SVM Method for the Classification of the Data of the Internet portal of the Bavarian Government According to the Life Event Principle

摘要

The search engine of the internet portal of the Bavarian government (www.bayern.de) faced the challenge to make publicly available over 4.5 million documents on the websites of ministries, public municipalities, authorities, etc. In order to make the results list of a search more accessible to the user, in this case private persons and businesses, the "Life Event" approach was chosen. This means, that all documents were put into one or more of a set of 650 predefined categories that each correspond to a specific situation, for example moving, marriage or birth. This assignment was achieved using two methods, namely the definition of word profiles for the documents belonging to a specific class and the mathematical-statistical method "Supported Vector Machine (SVM)". In this paper, we will briefly present the concept of Live Events and the two employed classification methods. We examine whether they lead to comparable results. It turns out that the use of a combination of both methods provides the best results regarding a plausible classification in many cases.