Abstract
Dynamic topic modeling is useful at discovering the development and change inlatent topics over time. However, present methodology relies on algorithms thatseparate document and word representations. This prevents the creation of ameaningful embedding space where changes in word usage and documents can bedirectly analyzed in a temporal context. This paper proposes an expansion ofthe compass-aligned temporal Word2Vec methodology into dynamic topic modeling.Such a method allows for the direct comparison of word and document embeddingsacross time in dynamic topics. This enables the creation of visualizations thatincorporate temporal word embeddings within the context of documents into topicvisualizations. In experiments against the current state-of-the-art, ourproposed method demonstrates overall competitive performance in topic relevancyand diversity across temporal datasets of varying size. Simultaneously, itprovides insightful visualizations focused on temporal word embeddings whilemaintaining the insights provided by global topic evolution, advancing ourunderstanding of how topics evolve over time.