In Cooperation with BMW Group

The goal of this Master’s Thesis is to develop an approach for measuring the similarity among docu-ments, stemming from various areas such as scientific literature, belles letters, news, etc. that might be written in different styles and languages. In traditional text clustering methods this is done through the “bag of words ” concept. This method calculates the similarity of these documents based on the frequencies of each term found in them, but exhibits the drawback of ignoring the semantic relation-ship among the words. Consequently, if two documents, representing the same topic use different terms or synonyms, they will be falsely classified as distant. In order to overcome this problem some external knowledge has to be used. Wikipedia is an appro-priate example for a good external dictionary- it is multilingual and written collaboratively by more than 10000 regular editing contributors. Each article describes a single topic. If equivalent concepts, e.g. synonyms or alternative names exist, they are redirected to the same article, section of an article, or page usually from alternative article, describing the main concept. Every Wikipedia page belongs to at least one category. Furthermore, there are links to other languages associated to each topic. Therefore, the information provided by Wikipedia is applied to enhance the "bag of words " techniqu