If you have paid license of Viscovery SOMine Cluster and Classify, you can use powerful Profile Analysis features.
As expained in the page “The Ugly Duckling Theorem and SOM“, the purpose of clustering is to discover new useful “segmentations”. In the philosophy of science, there is a concept called ”instrumentalism,” which is the standpoint that scientific theories are formal tools for organizing and predicting observable phenomena rather than “truths”. Considering this philosophy, it may be easier to understand the purpose of clustering.
Therefore, even in cluster analysis before SOM, it is more realistic to focus on whether the clustering results are useful, rather than whether they are correct. Even in traditional cluster analysis, there is a “profile analysis” that analyzes the statistical characteristics of clusters. Cluster analysis must be completed with profile analysis. However, it may not be taught much in universities.
Although Viscovery SOMine allows for powerful clustering, you can even use clustering as a “tool” and freely tune it without getting too attached to its results. Practical people should focus their efforts on finding useful segmentations.
Group Profile
Select ”View｜Group Profile”, then “Group Profile” window appear. By default, The Group Range is set to “Cluster”. Thus, the chart in the “Group Profile” window shows the characteristics of the cluseter where caret is in.
The graph shows attributes that have a significant difference in mean value when comparing the data in a cluster to the entire data set, and attributes with bars on the right (for example, keywords) indicate that the values in that cluster are Indicates that it is higher than the whole. The left is the opposite. Bar length is in standard deviation.
One way is to increase the number of clusters to see more details. Alternatively, as mentioned in “How to access the underlying data”, you can change the “Group Range” to “Nearest Nodes” and click in different places on the map to see features in different areas of the map, you can investigate efficiently.
The reduced version of the map contains 70 to 80 keywords, and the full version of the map contains thousands of keywords. Please note that the reduced version allows you to “experience” the above analysis, but the full version is required for more serious analysis.
Furthermore, it cannot be denied that there are advanced concepts that cannot be grasped simply by the appearance of keywords. Direct reading of texts is also required in map areas of interest. As mentioned in “How to access the underlying data”, it is also a good idea to have the LLM chatbot read the text and explain the concepts.
<< How to read the relationships btween Attributes.
How to create alternative models. >>
^Tutorials top