Using Cleantech PatentEdge cloud software, we are able to deliver you phenomenal innovation visualizations based on user input. For example, after training our machine algorithms on a few sets of technologies, we can show companies competitors they were not aware of, by visualizing innovations which at very close in terms of technological use cases. This allows companies to adjust their competitive output, and also gives parties an opportunity to license stagnant patents.

The model uses artificial neural networks to try to predict individual word context, which consists of the words surrounding each single word in a patent. As the model sees more training text, it gradually adjusts its parameters so that the error is minimized when performing the context prediction. Once training is complete, the network is presented with a single word and the resulting internal activations of the network are used as that word’s semantic embedding vector (see also, Johnson & Whitehead, 2016).

Here is a sample machine-learning visualization output of wind energy motors:

In output (a), it is impossible to understand how slight variations in wind motors are distinguished. In model (b), one can surmise the dark-blue patent class of technologies stand apart, and it seems like they are unrelated to the rest. Meanwhile, in output (c), other slight variations are seen. Finally, after the machine has properly learned, we can see each of the 8-9 colors are an entirely new species of “wind energy motors”. What this tells us is exactly what companies are working on these variations, where and why this is the case. This also gives us predictive capabilities, a great insight for investors.


Another example output using this algorithm:

The top image shows solar technology variations, extrapolated from the spherical image below. Now we are ready to export technologies to hot regions of technological innovation, find a licensee for our new solar technologies, or embark on strategic partnerships with firms in different parts of the world.