Methodology

In this version the unit of analysis is the patent record (one spreadsheet row). Relations are generated only within a record, never across records. Family-level deduplication is only possible when a reliable family identifier column exists; Lens ID and Application Number are never assumed to be family identifiers.

Occurrence is the number of distinct records containing an item. Repetitions of the same value inside one record are deduplicated and counted once.

Co-occurrence is the joint presence of two items in the same record.

A link is a connection between two nodes. Link strength is the frequency with which the two items appear related across the analyzed records. Total link strength is the sum of the strengths of all links of a node.

• IPCR × IPCR: number of documents in which both codes appear together.
• Applicant × Applicant: number of documents in which both applicants appear together.
• Jurisdiction × IPCR: number of documents in which the jurisdiction is associated with the IPC code.

In a same-field network, all unordered pairs of the unique values of a single field are generated: n(n − 1)/2 pairs per record. In a bipartite network, the Cartesian product of the unique values of two different fields is generated: nA × nB pairs per record. Node descriptions identify which side each node belongs to.

Full counting (default) adds 1 to every pair produced in a record. Record-normalized fractional counting adds 1/P to each pair, where P is the number of pairs produced in that record — 1/(nA × nB) for bipartite networks.

Codes are uppercased and stripped of internal spaces, keeping the “/” separator. From “H02J13/00” the application derives: section H, class H02, subclass H02J, main group H02J13 and the full code H02J13/00. Subclass is the recommended default for readable maps.

Every value is trimmed, has repeated spaces collapsed and empty values removed, then deduplicated within the record. Organization names are treated conservatively: legal suffixes such as LTD, INC, GMBH or UNIV are never removed automatically, because that would merge distinct entities. Use the normalization dictionary to merge variants explicitly.

Occurrences, Links, Total link strength, Average publication year and Average citations are exported in the map file as VOSviewer weights and scores. Columns whose source fields are absent from the dataset are omitted.

Your patent dataset is processed locally in your browser. No file is uploaded, stored or sent to any API.