Patent citations are often used to determine the impact of patents, or groups of patents on the interrelatedness between organizations based on their co-citation patterns. A valuable way of looking at patent citations is to utilize Network Analytics to map the relationships between documents. Based on a number of seminal patents in Virtual Reality, this post shows how Network Analytics can be used to visualize and analyze large sets of documents simultaneously, and gain insights which cannot be obtained using traditional forms of citation analysis.
The Case: Virtual Reality
One of the first patent applications in the Virtual Reality (VR) realm was filed in 1961 by cinematographer Morton Heilig, seen by many as one of the early pioneers in the field. The patent (US3050870) described the ‘Sensorama simulator’, a device which showed 3D movies, had stereo sound, a vibrating chair, and was even able to produce wind and smells to accompany the virtual experience. Unfortunately, Heilig was unable to secure the necessary financial backing for the project, and ultimately had to abandon it.
Next to the Sensorama, Heilig worked on various related projects as well. His ‘Experience Theater’ (patented under US3628829 and US3469837) was essentially a version of the Sensorama for a larger audience, while his first invention (dubbed the ‘Telesphere Mask’ and patented under US2955156A) was actually one of the first head-mounted displays.
Taken together, these patents received at least 275 citations over the years. Studying the importance of Heilig’s inventions for the VR field as a whole, let’s further investigate what the impact of these patents has been by applying Network Analytics to visualize and analyze their first and second generation forward citations.
Network Analytics Applied to Citation Data
Traditional visualizations of patent citations show a focal patent connected to its forward and backward citations in a simple tree-like diagram. Although these visualizations allow easy exploration of the directly connected patents of a focal patent, they fail to show relations between patents across multiple citation generations. A more suitable methodology to accomplish this is based on the principles of Social Network Analysis (SNA). SNA focuses on the analysis and visualization of relationships between entities in a specific context. Networks consist of nodes (entities) and edges (connections), and can vary in size from the relationships between just a few actors to the relationships between millions of actors sharing billions of connections.
We can apply this to patent citation data in a relatively straightforward way. We will define individual patent documents as the network’s nodes, and draw an edge between two documents whenever one document cites another. When doing this for multiple generations of citations, it is possible to obtain a visual overview of the evolution of patent clusters over time, which can be very helpful in locating additional prior art or getting a view on the technology areas adjacent to an initial area of interest.
Visualizing Citation Networks
We collected the direct (first generation) and indirect (second generation) forward citations for each of the patents mentioned above from the databases of the European Patent Office. The forward citations were then connected to each of the focal patents to obtain a network visualization of citation activity. The image below shows the focal patents, and their direct forward citations:
We clearly see Heilig’s patents as the large central nodes residing in the center of their clusters. Interestingly, there’s a clear division between the Experience Theater and the other two patents: the Sensorama and the Telesphere Mask (called ‘Stereoscopic Television Apparatus’ on the patent) are connected to each other through various shared forward citations, with the only connection to the ‘Experience Theater’ patent being through a citation by a patent related to a scent diffuser device. Indeed, both the Sensorama and the Experience Theater had this kind of functionality.
This visualization is quite easy to interpret, since it includes just the three focal patents and their direct forward citations. Apart from the fact that shared citations are also visible, this does not differ very significantly from what is possible using traditional citation tree analysis. The real power of Network Analytics becomes clear when data sets get larger and also include indirect citations. Consider for instance the table below, which shows how large a data set can get when including multiple generations of forward citations:
|Generation||Number of patents|
|0 (Focal patents)||3|
|1 (Direct forward citations)||278|
|2 (Indirect forward citations)||3,823|
|3 (Indirect forward citations)||36,258|
|4 (Indirect forward citations)||228,850|
In four generations, Heilig’s three patents reach just under 230,000 other patents. As a further example, let’s take a look at the network when we also include the second generation of forward citations:
A cluster detection algorithm, which is often used in SNA was used to assign colors to patents based on their proximity to each other in terms of their connections. We then used a force-directed layout algorithm to optimize the positions of the nodes and clearly show clusters of related patents based on their shared citations.
The network is quite vast and consists of over 3,800 patents, but shows a very distinct structure. The left part of the network deals with all kinds of technology clusters related to Heilig’s Experience Theater patent, which is situated in the center. The dark blue cluster at the top left consists of over 200 patents related to ventilated and air-conditioned seats, in line with the type of seat that Heilig envisioned for his theater. The pink cluster at the bottom left deals with all kinds of technology related to gaming machines – the lighter blue group of patents next to that holds patents in the area of amusement rides based on motion simulation:
Interestingly, there is one cluster which very clearly links the Experience Theater patent, and the clusters related to the Sensorama and Telesphere Mask at the right. In line with the earlier visual of the direct citations, this cluster is fully focused on scent delivery, a shared characteristic of Heilig’s inventions:
The right part of the network is made up of citations related to the Sensorama and Telesphere Mask patents, and is most in line with the current interest in Virtual Reality which focuses on head-mounted devices such as the Oculus Rift, Microsoft HoloLens or Sony Playstation VR. The larger nodes depicted here are those which generated most forward citations over the years across the two generations visualized here, and can be seen as key patents in this specific space. The Sensorama and Telesphere Mask patents are highlighted in the boxes:
Of course, this is just the tip of the iceberg. There’s a lot more to discover in these networks when a user is also able to interactively zoom, search and filter through the data, draw their own conclusions and export interesting results to a tabular format for follow-up actions. That’s why it’s always recommendable to provide clients with an interactive app which holds the results of a patent citation network analytics assignment, and go through the full results in a workshop format to get the most out of such an analysis.
The application of Network Analytics in the context of patent citations has significant potential. It provides the opportunity to view the relationships between multiple generations of citations, which is a significant advantage compared to traditional tree-like visualizations. It is possible to interactively visualize the evolution of a certain field over time, which allows distinguishing the relationships between clusters of related technology. Within these clusters, one may be able to find prior art which can serve as a complement to prior art located through traditional methods of patent search based on keywords or technology classifications.
Of course, depending on the question at hand, patent citation Network Analytics is most powerful when used together with other analysis angles. For instance, it would also be interesting to connect the applicants listed on the patents in this network to determine which companies are most related in terms of their (reciprocal) citations. Other angles could be to focus on the IPC or CPC codes in these patents to accurately show interlinked technology areas, or to investigate the networks of inventors responsible for these patents.
In any case, network analytics can provide a unique view on the actual relations residing within a patent dataset, serving as a strong addition to more traditional forms of analysis.
 *: these two patents share a priority date of March 9, 1966. We therefore interpreted them as part of the same patent family and merged their forward citations in the analyses.