At the moment the pooling is merely a flattened list of concepts. Need to fix the pooling based on concept similarity and random selections.