PatentSight’s Technology Clusters are sets of similar patent families created based on full-text patent data under consideration of IPC classifications by an automated machine learning technique. Built from all documents from all patent families and other IP rights in the PatentSight database irrespective of their legal status, Technology Clusters comprise four levels of hierarchy.
- First, a self-developed IPC mapping process is applied to allocate each patent to 135 Level 2 clusters. Patens can be mapped to multiple Level 2 Clusters at that point.
- The Level 2 clusters serve as basis for the creation of Level 4 clusters (word similarity is measured in the semantic context of each Level 2 cluster), which are created based on full-text data. Including titles, abstracts, claims and descriptions, the documents in each Level 2 cluster are processed into word stems. Taking into consideration the frequency of these word stems, the stems are mapped to vectors and clustered, thereby resulting in Level 4 clusters.
- Then, the Level 4 clusters are automatically aggregated to Level 3 clusters, which is also based on full-text data.
- Next, PatentSight's technology experts manually aggregate Level 2 clusters into a total of 13 Level 1 clusters.
- In a following step, patents are assigned to the most fitting Level 4 cluster. Based on this level 4 cluster the patent is also assigned in a cascading way to the overlying level 3, 2 and 1 cluster. I.e. while in step 1 patents could have been mapped to multiple of the 135 technology this is not the case anymore after the assignment of the patents to the clusters (no co-classification).
- Finally, the clusters are automatically labelled based on patent titles and abstracts.
Freedom of Overlaps
Since each patent family in the PatentSight database belongs to exactly one Technology Cluster on each level, portfolios can be analyzed without having to pay attention to technology overlaps (e.g., as in IPC classifications).
The continuous inflow of new patent families and other IP rights coming along with the weekly updates of the PatentSight database also results in ongoing updates of the Technology Clusters. These weekly updates include changes in the overall number of Level 3 and 4 clusters (with an increasing tendency), the reallocation of patent families to different clusters, and changes of the titles of the clusters.
Based on iterative training, PatentSight’s clustering approach allows to easily incorporate new data. In this, automated machine learning continuously improves the process of clustering.
How to Use Technology Clusters
You can use Technology Clusters as a search filter element to narrow down your search to a technology field of interest
Searching by Technology Clusters can be helpful in prior art searches, FTO searches, benchmarks and many more use cases!
Narrow down your search to a Technology Cluster of interest, e.g., to find out who owns portfolios in this field and how these portfolios have developed over time
Use the sunburst chart to get an overview of an owner’s portfolio grouped by Technology Clusters
Edit the sunburst chart according to your needs
Of course, you can use Technology Clusters also as grouping and attributes and measures in all other charts and tables available in the PatentSight BI
Use Technology Clusters to quickly find out in which technology areas competitors are active and own strong patent portfolios