Searching patent documents

A system for identifying patent documents similar to a search query, the system comprising:

• i. one or more processors, and
• ii. one or more computer readable storage devices,
• iii. the one or more storage devices storing computer instructions which when implemented cause the one or more processors to:
• iv. determine first similarity values between the natural language content of the search query and patent documents of a set of patent documents by comparing a query vector embedding determined from the natural language content of the search query using an embedding machine learning model to document vector embeddings determined from respective patent documents of the set,
• v. wherein the document vector embedding determined from a respective patent document of the set is a weighted average of separate vector embeddings of the natural language content of a set of one or more claims and of one or more other sections of that patent document, each determined using the embedding model;
• vi. select a first subset of the set of patent documents including patent documents with first similarity values above a first similarity threshold;
• vii. determine a second similarity value between the natural language content of the search query and each of the first subset of patent documents using a cross-encoder machine learning model;
• and
• viii. identify a second subset of the first subset of patent documents with second similarity values above a second similarity threshold.

A method for identifying patent documents similar to a search query, the search query comprising natural language content, the method comprising:

• a. generating, by an embedding machine learning model, a document vector embedding for each patent document of a set of patent documents, wherein the document vector embedding for a respective patent document is a weighted sum of separate vector embeddings of a claims section and of one or more other sections selected from an abstract, a description and a title of the patent document, each of the separate vector embeddings being produced by the embedding machine learning model, and wherein weights of the weighted sum are trainable parameters learned via backpropagation using patent-specific similarity labels comprising at least one of CPC or IPC proximity metrics or citation-derived labels;
• b. generating a query vector embedding from the natural language content of the search query using the embedding machine learning model;
• c. determining first similarity values between the query vector embedding and the document vector embeddings for the set of patent documents;
• d. determining a first similarity threshold as an adaptive threshold computed from a dispersion-based statistic of a top-k subset of the first similarity values;
• e. selecting a first subset of the set of patent documents including patent documents with first similarity values above the first similarity threshold;
• f. determining, for each patent document in the first subset, a second similarity value between the natural language content of the search query and the patent document using a cross-encoder machine learning model configured to receive a concatenation of the natural language content of the search query and at least part of the patent document separated by a separator token and to output a scalar similarity score;
• and
• g. identifying a second subset of the first subset of patent documents with second similarity values above a second similarity threshold.

A method for identifying patent documents similar to a search query, the search query comprising natural language content, the method comprising:

• a. generating, by an embedding machine learning model, a document vector embedding for each patent document of a set of patent documents, wherein the document vector embedding for a respective patent document is a weighted sum of separate vector embeddings of a claims section and of one or more other sections selected from an abstract, a description and a title of the patent document, each of the separate vector embeddings being produced by the embedding machine learning model, and wherein weights of the weighted sum are trainable parameters learned via backpropagation using patent-specific similarity labels comprising at least one of CPC or IPC proximity metrics or citation-derived labels;
• b. generating a query vector embedding from the natural language content of the search query using the embedding machine learning model;
• c. determining first similarity values between the query vector embedding and the document vector embeddings for the set of patent documents;
• d. determining a first similarity threshold as an adaptive threshold computed from a dispersion-based statistic of a top-k subset of the first similarity values;
• and
• e. selecting a first subset of the set of patent documents including patent documents with first similarity values above the first similarity threshold.