Concurrent Visualization of Relationships between Words and Topics in Topic Models

Analysis tools based on topic models are often used as a means to explore large amounts of unstructured data. Users often reason about the correctness of a model using relationships between words within the topics or topics within the model. We compute this useful contextual information as term co-occurrence and topic covariance and overlay it on top of standard topic model output via an intuitive interactive visualization. This is a work in progress with the end goal to combine the visual representation with interactions and online learning, so the users can directly explore (a) why a model may not align with their intuition and (b) modify the model as needed.