Topic modeling is a popular technique for exploring large document collections. It has proven useful for this task, but its application poses a number of challenges. First, the comparison of available algorithms is anything but simple, as researchers use many different datasets and criteria for their evaluation. A second challenge is the choice of a suitable metric for evaluating the calculated results. The metrics used so far provide a mixed picture, making it difficult to verify the accuracy of topic modeling outputs. Altogether, the choice of an appropriate algorithm and the evaluation of the results remain unresolved issues. Although many studies have reported promising performance by various topic models, prior research has not yet systematically investigated the validity of the outcomes in a comprehensive manner, that is, using more than a small number of the available algorithms and metrics. Consequently, our study has two main objectives. First, we compare all commonly used, non-application-specific topic modeling algorithms and assess their relative performance. The comparison is made against a known clustering and thus enables an unbiased evaluation of results. Our findings show a clear ranking of the algorithms in terms of accuracy. Secondly, we analyze the relationship between existing metrics and the known clustering, and thus objectively determine under what conditions these algorithms may be utilized effectively. This way, we enable readers to gain a deeper understanding of the performance of topic modeling techniques and the interplay of performance and evaluation metrics.

Citation analysis has been applied to map the landscape of scientific disciplines and to assess the impact of publications. However, it is limited in that it assumes all citations to be of equal weight. Doing away with this assumption could make such studies even more insightful. Current developments in this regard focus on the evaluation of the syntactic and semantic qualities of the text that surrounds citations. Still lacking, however, are computational techniques to unpack the thematic context in which citations appear. It is against this backdrop that we propose a text clustering approach to derive contextual aspects of individual citations and the relationship between cited and citing work in an automated and scalable fashion. The method reveals a focal publication’s absorption and use within the scientific community. It can also facilitate impact assessments at all levels. In addition to analyzing individual publications, the method can also be extended to creating impact profiles for authors, institutions, disciplines, and regions. We illustrate our results based on a large corpus of full-text articles from the field of Information systems (IS) with the help of exemplary visualizations. In addition, we provide a case study, the scientific impact of the Technology acceptance model. This way, we not only show the usefulness of our method in comparison to existing techniques but also enhance the understanding of the field by providing an in-depth analysis of the absorption of a key IS theoretical base.

No abstract

Since data is often multi-faceted in its very nature, it might not adequately be summarized by just a single clustering. To better capture the data's complexity, methods aiming at the detection of multiple, alternative clusterings have been proposed. Independent of this research area, semi-supervised clustering techniques have shown to substantially improve clustering results for single-view clustering by integrating prior knowledge. In this paper, we join both research areas and present a solution for integrating prior knowledge in the process of detecting multiple clusterings.We propose a Bayesian framework modeling multiple clusterings of the data by multiple mixture distributions, each responsible for an individual set of relevant dimensions. In addition, our model is able to handle prior knowledge in the form of instance-level constraints indicating which objects should or should not be grouped together. Since a priori the assignment of constraints to specific views is not necessarily known, our technique automatically determines their membership. For efficient learning, we propose the algorithm SMVC using variational Bayesian methods. With experiments on various real-world data, we demonstrate SMVC's potential to detect multiple clustering views and its capability to improve the result by exploiting prior knowledge.