Patent Search

Input Strategy: Semantic-Search Queries

Patent SearchFlashPoint IP has formed strategic partnerships with premium, IP search-engine providers (with patented technologies in the field of patent search) which enable us to provide advanced patent searches for complex undertakings such as IP analytics and competitive intelligence. Furthermore, technical fields in which innovation is rapid, dense, and often incremental do not lend themselves easily to screening the prior art effectively using informed search strings. Advanced, value-added searches provide clients with rich knowledge to reflect on in modifying the focus of their invention upon patent analysis, and adding aspects otherwise considered irrelevant during the patent-drafting stage.


Traditional search methods predominantly rely on a ranking or score based on keyword count. Even with high-score results, such word-matching may still produce documents that are irrelevant to the patent search. While data mining has become somewhat of a science in analyzing structured datasets, text mining of unstructured data remains piecemeal in its attempt to derive meaning from free-flowing text.


In addition to this, patent documents encompass deep technical details, reliance on the body of knowledge of the informed reader, and legal considerations in document structure – making it an even more complex challenge to decipher during patent analysis. Yet, unlike meandering prose (or an open-ended blog), patent documents do have a clear point (or several) to convey, making the challenge somewhat more approachable. Natural Language Processing (NLP) involves the extraction of meaningful information from natural language input.


Relying on a practical mix of statistical modeling, machine learning, and language analysis, our providers employ NLP-based algorithms which utilize semantic dimensions to represent a document. Literal word-matching becomes just one aspect of retrieving documents. For example, the word class can be a grouping or category, an indication of quality or sophistication, or a learning session. Merely matching the word class in a document is not an indication of its relevance to a given query containing the word class. Distinguishing between different word meanings and usages in a document query helps reduce the number of unrelated matches retrieved in a patent search.


Semantic searching differs greatly from conventional keyword searching. Keyword searching works best with a small number of search terms because longer queries become harder to structure to obtain meaningful results, while semantic searching works better with longer queries to clarify distinguishing features and context.


The text input for a semantic search can be based on an entire invention disclosure document, a published patent document or part thereof (such as the claims), a technical journal article or whitepaper, or even a well-crafted summary of a discussion (prior to commencing the patent-drafting process) between the inventor/entrepreneur and the search analyst (with the advantages over known prior art clearly articulated).


The length of such text input is practically unlimited without the need for intensive linguistic rules – pages of rich content which compare and contrast the invention to what is known in the art is well within the search-type capability of patent analysis. FlashPoint IP’s semantic searches allow a few iterations of searching (with interim search results provided to the client) to be performed to optimize the input text in order to improve the search results. Contact us to find out more about FPIP patent searches.