Topic Modelling and Network Models

In preparation for a workshop at this year’s Academy of Management conference my colleague Tim Hannigan and I gathered data on all the sessions and workshops. We then processed them using R’s topic models package to extract “themes” from the abstracts.The purpose was to identify what the divisions at Academy were discussing and to look at the overlap between them. The image below is one result from this work

What you see is a bipartite graph which is essentially a network graph with two types of entities. In this case our entities are AOM divisions (in black text encased in the red blob) and topics (in blue text). The theme for this year’s conference is the “Power of Words”. Thus, it’s fortunate that the “Power Words” topic we identified from abstracts is squarely in the center of the red blob. This means it’s a central theme relative to the divisions. Thus, the divisions have done a good job of incorporating the theme. Another interesting feature is that there are divisions, such as HR/OB, TIM, OMT, that are central to topics. That means that they are diverse with respect to themes.

Outliers are located in complement of the intersection of the colored blobs. Thus, divisions such OM (Operations Management) could be considered outlying, and among topics, supply chain would be as well. In this case, supply chain is the theme most linked to OM. In contrast, there are outlying themes, such as FirmPerformance, that are strongly tied to central divisions. In this case, those divisions are TIM, OMT and BPS — but not much else.

Anyway, topic analysis is a cool way to identify themes in text and opens up a lot of exciting possibilities. For instance, one neat thing to do would be to look at how themes within Academy have changed over the years.