To better understand how Patentcloud’s Semantic Search works, let’s begin with the fact that Patentcloud has an extensive library of millions of patent documents, from patent offices that include the US, China, Japan, Europe, WIPO, and Korea.
During the Semantic Search’s pre-processing phase, this entire library is converted into vectors (converting words, sentences, or documents into vectors simply means changing them into real numbers, which then can be compared to the user’s query).
So, when a user makes a query using Semantic Search, the content in the query is processed in Patentcloud’s proprietary machine learning model, where the query itself is converted into vectors. This means that this content also becomes a representation in real numbers, with the occurrence of words as vectors and the context in which they occur of considerable importance.
Next, these vectors are used to calculate the similarity to Patentcloud’s entire patent document database, with the context, location, and frequency of the vectors playing key roles. From here, the patents with the highest similarity are selected and presented in the search results, with the most relevant results presented first.
All of this is completed in just seconds.
Unlike other vendors, Patentcloud’s machine learning model is trained and optimized exclusively on patent data, and this library is regularly updated with the latest patent documents. Perhaps even more importantly, as more patent data is fed into Patentcloud’s predictive model, this model becomes even more accurate.
This is how Patentcloud’s Semantic Search is able to find the patent documents that are most relevant to a user’s search—by using semantic similarity via a proprietary algorithm instead of simply matching keywords.