Idea Clustering/OG

From Open Pattern Repository for Online Learning Systems
Jump to navigation Jump to search

Idea Clustering
Contributors Christian Kohls
Last modification May 17, 2017
Source Kohls (2016)[1]
Pattern formats OPR Alexandrian
Learning domain


You have generated a larger number of ideas using Brainstorming and Ideation techniques. Many of these techniques require judgment and proposals are uttered in an unstructured way. Since any judgment or categorization was suspended you cannot see the relations, similarities and differences between the ideas.


A huge number of unstructured items is hard to manage. One does not see relations or get the big picture. If you have a huge pile of ideas you need to create order. But where to begin?


Some ideas are just similar. Some ideas are different. Some ideas are sub categories of other ideas. Some ideas are related to other ideas. But if we have no order yet how can we identify which ideas belong together?

We want to find similar ideas and put them into categories. This is like putting all apples in one bucket and all peas in another bucket. But we have no buckets yet. Good ideas are novel. Old ideas may come from many different and unrelated fields. New categories have to emerge.

Similar ideas are closer to each other. Different ideas are at a greater distance to each other. This proximity or distance needs to be visible.


Therefore, create clusters of ideas. You need visual representations for each idea. The representations need to be on atomic items that can be freely moved and rearranged. Either use sticky-notes, cards, or a digital tool. Place ideas that are somehow related close to each other. Start with any two ideas that somehow match and create a first cluster. If you find more similar ideas, add them to the cluster. If you find ideas that do not fit to the cluster, create new clusters. You may need to split some of the clusters if you feel that there are too many dissimilarities in it. You may merge clusters that have similar ideas. You can identify main ideas that work as headlines for the clusters. You can visually highlight these ideas or move them into the center or on top of the cluster. Once you have identified categories you can start to find more examples for each category.

  1. Each idea needs to be represented on an atomic item (sticky-note, card, paper snippet) that can be moved around.
  2. Identify items that “somehow” belong together. It could be similar properties (any properties work) or a relation between the two items. Think of a jigsaw puzzle where you have to look for items that match.
  3. Place matching items close together. Create visual proximity for items that have semantic proximity. If possible, let the items overlap (easy to do with cards or sticky-notes).
  4. Constantly re-evaluate the emerging clusters and move items or group of items between clusters.
  5. Strive for large homogeneity within each cluster and heterogeneity between clusters.

At any time you can move one idea from one cluster to another if it fits better into it. Similarity can be based on different properties and which properties are most relevant only emerges in the process of clustering. Clusters are no hard coded categories. They are fuzzy categories and the ideas you move into one cluster are “somehow” related. But the “somehow” may only be clear at the end of the clustering process.

Good clusters contain homogeneous items. That means each item fits better to the cluster it is placed in than to any other cluster. Items between different clusters should be heterogeneous. That means you should clearly see why any item is cluster A instead of cluster B. If it’s not clear you should rethink about where to put the item. You can move items back and forth all the time. Clusters only emerge.

If you find difference between the items of an otherwise homogenous group you can try to create sub-clusters. Likewise, if you find that between to heterogeneous clusters there are also similarities (maybe with regard to other properties), you can move the two clusters closer to each other indicating that they are sub-clusters of another category. Note that you cannot only move single items around but whole clusters! That’s especially easy if you use digital tools.

Adding new items to a cluster may change its character. For example, you may put an item to a cluster based on its similarity (e.g. similar target group of young adults). But this may also create new differences within the cluster (maybe the cluster had products for young adults but the new item is a marketing strategy specific for young adults). Don’t care about that immediately. Observe how the cluster further develops. You may have to split the cluster again or remove the item later. Putting an item into a cluster is just a working assumption. The agility of the method is that you try out different variations.

At the end of the clustering you have identified different categories. You have identified similar ideas and can tackle them together. You have clear alternatives you can chose between. You may identify gaps in the clusters and come up with new ideas to fill up the gaps.

The fuzziness of the clusters is a benefit but also a burden. Clusters are not tried and tested categories. But they are good enough approximations of meaningful categories. Some items might fit into several clusters any it can be unsatisfying to put them into one category only. One has to compromise sometimes. You could also duplicate an item and put it into two clusters but this adds additional complexity and redundancy.


  1. Kohls, C. (2016). The magic 5 of innovation: judgement patterns. In Proceedings of the 21st European Conference on Pattern Languages of Programs (EuroPLoP 2016) (p. 21). New York:ACM.