Increasing Topic Coherence by Aggregating Topic Models

In this paper, we introduce a novel method for aggregating multiple topic models to produce an aggregate model that contains topics with greater coherence than individual models. When generating a topic model a number of parameters must be specified. Depending on the parameters chosen the resulting topics can be very general or very specific. In this paper the process of aggregating multiple topic models generated using different parameters is investigated; the hypothesis being that combining the general and specific topics can increase topic coherence. The aggregate model is created using cosine similarity and Jensen-Shannon divergence to combine topics which are above a similarity threshold. The model is evaluated using evaluation methods to calculate the coherence of topics in the base models against those of the aggregated model. The results presented in this paper show that the aggregated model outperforms standard topic models at a statistically significant level in terms of topic coherence when evaluated against an external corpus.