CATI: An Extensible Platform Supporting Assisted Classification of Large Datasets

More and more, researchers in humanities and companies need large classified document data-sets. These users are not familiar with information retrieval or data science notions. For data scientists, there is also often a need for those classified document data-sets as ground truth. There are multiple tools that allow users to carry out this classification task on large data-sets, involving always a quite expert level in computer and data science. More over, these tools are not usually oriented to the domain of micro-blogs or do not always take into account meta data and attached images as additional dimensions to improve the classification. In this work, we present a platform to enable end users to classify large document collections of several hundred thousands documents in an assisted way, within a humanly acceptable number of clicks, with no coding and without having data science and information retrieval expert knowledge. The system includes a graphical user interface with several classification assistants doing text- and image-based event detection, geographical filtering, image clustering, search services with rich visual metaphors to visualize their results and finally Active Learning (AL) with different sampling strategies. We also present a comparative study on the impact of using different and interchangeable AL components on the number of clicks needed to reach a stable level of accuracy.