Skip to content

Coherence Regularization for Neural Topic Models

Neural topic models aim to predict the words of a document given the document itself. In such models, perplexity is used as a training criterion, whereas the final quality measure is topic coherence. In this work, we introduce a coherence regularization loss that penalizes incoherent topics during the training of the model. We analyze our approach using coherence and an additional metric – exclusivity, responsible for the uniqueness of the terms in topics. We argue that this combination of metrics is an adequate indicator of the model quality. Our results indicate the effectiveness of our loss and the potential to be used in the future neural topic models.

The paper will be published at the 16th International Symposium on Neural Networks taking place in Moscow. In the meantime, do not hesitate to contact our R&D department at to discuss how you can leverage neural topic models in your projects.

Katsiaryna Krasnashchok, Aymen Cherif, Coherence Regularization for Neural Topic Models. in 16th International Symposium on Neural Networks 2019 (ISNN 2019)

Click here to access the paper.

Releated Posts

Augment to Interpret: Unsupervised and Inherently Interpretable Graph Embeddings

In this paper, we study graph representation learning and show that data augmentation that preserves semantics can be learned and used to produce interpretations. Our framework, which we named INGENIOUS, creates inherently interpretable embeddings and eliminates the need for costly additional post-hoc analysis.
Read More

SANGEA: Scalable and Attributed Network Generation

In this paper, we present SANGEA, a sizeable synthetic graph generation framework that extends the applicability of any SGG to large graphs. By first splitting the large graph into communities, SANGEA trains one SGG per community, then links the community graphs back together to create a synthetic large graph.
Read More