Scraping Patent Citations for Prior Art Research with Python

#python #tutorial #webdev #programming

Patent citation analysis reveals technological lineage. Automated prior art search saves thousands in legal fees.

USPTO PatentsView API

import requests, time
from collections import defaultdict

class PatentScraper:
    def __init__(self):
        self.s = requests.Session()
        self.s.headers["User-Agent"] = "PatentResearch/1.0"

    def search(self, query, n=50):
        r = self.s.post("https://api.patentsview.org/patents/query", json={
            "q":{"_text_any":{"patent_abstract":query}},
            "f":["patent_number","patent_title","patent_date","patent_abstract",
                 "assignee_organization","cited_patent_number","citedby_patent_number"],
            "o":{"page":1,"per_page":n},"s":[{"patent_date":"desc"}]})
        return r.json().get("patents",[]) if r.status_code==200 else []

    def details(self, pn):
        r = self.s.post("https://api.patentsview.org/patents/query", json={
            "q":{"patent_number":pn},
            "f":["patent_number","patent_title","patent_date","patent_abstract",
                 "cited_patent_number","citedby_patent_number","assignee_organization"]})
        ps = r.json().get("patents",[]) if r.status_code==200 else []
        return ps[0] if ps else None

Citation Chain Builder

    def chain(self, pn, depth=2):
        visited = set()
        result = {"nodes":[],"edges":[]}
        def go(p, d):
            if p in visited or d>depth: return
            visited.add(p)
            det = self.details(p)
            if not det: return
            asgn = det.get("assignees",[{}])
            result["nodes"].append({"id":p,"title":det.get("patent_title",""),
                "assignee":asgn[0].get("assignee_organization","?") if asgn else "?"})
            for c in det.get("cited_patents",[])[:10]:
                cpn = c.get("cited_patent_number")
                if cpn:
                    result["edges"].append({"from":p,"to":cpn})
                    time.sleep(0.3); go(cpn, d+1)
        go(pn, 0)
        return result

    def prior_art(self, desc, before=None):
        qw = set(desc.lower().split())
        results = []
        for p in self.search(desc, 30):
            pd = p.get("patent_date","9999")
            if before and pd >= before: continue
            aw = set(p.get("patent_abstract","").lower().split())
            rel = len(qw&aw)/len(qw)*100 if qw else 0
            results.append({"number":p["patent_number"],"title":p["patent_title"],
                           "date":pd,"relevance":rel})
        return sorted(results, key=lambda x:x["relevance"], reverse=True)[:15]

s = PatentScraper()
c = s.chain("11023456", 1)
print(f"Nodes:{len(c['nodes'])} Edges:{len(c['edges'])}")
for p in s.prior_art("machine learning image classification")[:5]:
    print(f"  [{p['relevance']:.0f}%] {p['number']} - {p['title']}")