Fusion of XAI and LLM: Automatic Topic Name Extraction from Text Data Using Approximate Inverse Model Explanations (AIME)
Takafumi Nakanishi · 2025
In this paper, we propose an automatic topic-name extraction method that combines explainable AI (XAI) and a large language model (LLM) for text datasets. In the past, topic models such as latent Dirichlet allocation and latent semantic analysis have been widely used to extract potential topics from text datasets; however, it is not easy to assign appropriate names to the extracted topics. Although these methods can indicate a set of words related to a topic, it is difficult to derive a specific topic name from the set. In this study, we propose an approach to derive the counter-important features of each cluster of text data using our XAI technique, approximate inverse model explanations. This approach enables us to understand the features of each cluster, and the results are input into the LLM for the automatic generation of summaries and topic names. This process automates topic naming and makes it easier to understand the topics of a text set. This method is positioned as a new application that integrates XAI and LLM, and it is expected to be useful as a text mining and data analysis tool.