| A recent trend in machine learning is the development of models that can explain their own predictions. The emerging field of interpretable artificial intelligence (XAI) has so far concentrated primarily on supervised learning, and more specifically on deep neural network classifiers. In many practical problems, however, label information is not available, and the objective is to uncover the underlying structure of the data, for example in the form of clusters. Although there are powerful methods for extracting cluster structures, they usually do not provide an explanation of why a particular data point has been assigned to a specific cluster. To address this gap, a new framework has been proposed, that can explain cluster assignments in terms of the input features in an effective and reliable manner. This is achieved by reformulating clustering models as neural networks, a process we refer to as neuralization. In this work, we utilize the application of this method on text data and we propose a comprehensive method for topic extraction followed by interpretation of the results using large language models (LLMs). |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.