This talk will be presented by Marek Furka (TU Vienna) on Tuesday, May 3rd at 3.30 PM in a hybrid format. Marek wrote his Master’s thesis in collaboration with CSH.
If you want to join the talk online, please send us an email.
Title: Exhibit rating prediction and visitor path prediction in a museum setting
Rating prediction and path prediction tasks in a museum/exhibition setting are important problems, yet not well explored. In this study, we aim to close the existing research gap by giving a comprehensive comparison of state-of-the-art techniques for predicting user ratings and paths, which we test in an exhibition setting. The whole study is based on a dataset consisting of explicit museum visitor ratings collected using a web application in an exhibition in Rome. We select numerous relevant approaches and baselines, adapt them to our setting if needed, propose several modifications of the approaches and compare them using a comprehensive evaluation framework. Our proposed evaluation framework considers the order of the items in which the visitors visited the items that we argue is necessary for the physical setting.
We also analyze the performance of the approaches in terms of visitor path depth (the number of ratings a user has already given). In the case of the rating prediction, we focus on the ranking metric precision at k, and in the case of path prediction, we focus on the subsequence accuracy metric. In the evaluation for the rating prediction problem we observe that the approaches over-perform the baselines more, the more ratings the user gives and perform worse if we rank a higher number of items. In addition, we discovered that a state-of-the-art technique created for a large dataset from another domain unexpectedly performed best in general. For the path prediction task, we conclude a deep learning approach with our proposed modification performs the best in general unexpectedly on such a relatively small dataset.