What do we learn by applying multiple methods in topic detection? A comparative analysis on a large online dataset about mobility electrification

Identifying the topics covered in a corpus is one of the central issues in automatic text analysis. The objective of our paper is to contribute to the comparative analysis of different methods. In particular, we compare the results obtained through the use of the most common methods for topic identification, applied to the same corpus. The analysis is performed on a large original textual database created from an e-mobility newsletter. To compare the results between the methods, we refer to two criteria. First of all, the semantic consistency of the different models is evaluated by applying the UMass score and Pointwise mutual information. Secondly, the degree of association between the topics identified by the different models is processed using a heat-map and Cramer's V.