The authors discuss the use of the Kohonen Self-Organizing Map (SOM) to classify notes from an electronic brainstorming session. Because such sessions can generate massive amounts of output, an automatic means of organizing and indexing the content seems necessary. The Kohonen SOM is compared with earlier work using a Hopfield neural network, and both are compared to a manually performed control.

The Kohonen SOM outperformed the Hopfield network and did as well as a human classifier at representing the output and recall of topics and concepts. However, it did not provide as much precision as the human model. The authors suggest areas to consider in order to improve processing by the Kohonen SOM. They conclude that the technique holds considerable promise, especially considering the minimal amount of time it requires, as compared with the 40 minutes required by the manual classifier. Multiple layers of maps and real-time user augmentation of concepts are two areas that the authors intend to explore.