Cäcilia Zirn

Distinguishing between instances and classes in the wikipedia taxonomy

This paper presents an automatic method for differentiating between instances and classes in a large scale taxonomy induced from the Wikipedia category network. The method exploits characteristics of the category names and the structure of the network. The approach we present is the first attempt to make this distinction automatically in a large scale resource. In contrast, this distinction has been made in WordNet and Cyc based on manual annotations. The result of the process is evaluated against ResearchCyc. On the subnetwork shared by our taxonomy and ResearchCyc we report 84.52% accuracy.

Can we identify manipulative behavior and the corresponding suspects on review websites using supervised learning

Identification of manipulative behavior and the corresponding suspects is an essential task for maintaining robustness of reputation systems integrated by review websites. However, this task constitutes a great challenge. In this paper, we present an approach based on supervised learning to automatically detect suspicious behavior on travel websites. We distinguish between two types of manipulation, treating them as separate tasks: promoting manipulation, which is performed in order to push the reputation of a hotel, and demoting manipulation, which is used to demote competitors. Both tasks consist of three separate levels: detecting suspicious reviews (review level), suspicious reviewers (reviewer level) and suspicious objects of the reviews, i.e. hotels (object level). A separate classifier for each of the levels is trained on various sets of textual and non-textual features. We apply state-of-the-art machine learning algorithms like Support Vector Machines. The performance of our approach is evaluated on a new dataset that we created based on reviews taken from the platform TripAdvisor and which was carefully annotated by human judges. The results show that it is possible to identify manipulating reviewers and objects of manipulation with over 90% accuracy. Identifying suspicious reviews, however, seems to be a much harder task, for which our classifier achieves an accuracy of 68% detecting promoting manipulation and 84% detecting demoting manipulation. We argue that there is the need to identify more efficient features for the classification on review level. Finally, we analyze and discuss statistical characteristics of manipulative behavior based on the predictions of the reviewer and object level classifiers.

Analyzing Positions and Topics in Political Discussions of the German Bundestag

We present ongoing doctoral work on automatically understanding the positions of politicians with respect to those of the party they belong to. To this end, we use textual data, namely transcriptions of political speeches from meetings of the German Bundestag, and party manifestos, in order to automatically acquire the positions of political actors and parties, respectively. We discuss a variety of possible supervised and unsupervised approaches to determine the topics of interest and compare positions, and propose to explore an approach based on topic modeling techniques for these tasks.