Using Neural Networks to Visualize Patent Claim Support
A patent application must include a specification that enables one of ordinary skill in the art to make and use the claimed invention. In part, 35 U.S.C. § 112(a) states that a specification "shall contain a written description of the invention ... in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same."
Neural networks can be used to potentially identify issues and/or enablement support within a specification. For example, during drafting or prosecution, patent practitioners could use neural networks for comparing claim elements to paragraphs of the specification to determine whether the specification may properly enable the claim elements. The following image is a heat map illustrating the similarity of paragraphs of a specification and the claim elements of an independent claim:
The horizontal axis represents the paragraph numbers (labeled 0 through 126), and the vertical axis represents the claim elements (labeled 0 through 3). The heat map captures the similarity of the paragraphs with the claim elements. In other words, the darker red represents a greater similarity between the variables, and the light yellow represents dissimilarity between the variables. Based on the graph above, claim element 1 has the greatest similarity around paragraph 97 (discounting paragraph 7 - which is probably a direct recitation of the claim elements in the Summary section).
In conclusion, neural networks could be leveraged to determine possible 112(a) issues during drafting or prosecution.
ADAS Data at Ford Motor Company
4 年Do you have a link out to a more detailed description of the technique (e.g. a jupyter notebook). Are you using paragraph2vec or word2vec to make these numerical similarity estimates? What similarity metric are you using (eg. cosine similarity). If you are using a word2vec approach, have you pre-trained the vector representations on patent literature or a specific patent art unit(s)? Lastly, do you think the graph could be improved by plotting in aggregate rather than having an x-axis indicate individual patent applications since each patent is distinct from one another (not continuous or ordinal variable). Perhaps a violin/box plot could be more helpful although the multiple claims may need to be represented as a an average or hierarchically.