AI-SDV 2022: Machine learning based patent categorization: A success story in monitoring a complex technology with high patenting activity Susanne Tropf (Syngenta, Switzerland) Kornel Marko (Averbis, Germany)

AI-SDV 2022: Machine learning based patent categorization: A success story in monitoring a complex technology with high patenting activity Susanne Tropf (Syngenta, Switzerland) Kornel Marko (Averbis, Germany)

• 1.
  Averbis | ©2022
• 2.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Patent Monitor is a machine-learning based patent categorization application analyzes a large number of patents (and NPL) … … and automatically classifies documents into freely definable categories
• 3.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 2 Patent & Literature Categorization Pat_1 Category 2 Pat_2 Category 1 Pat_1 Category … n Pat_i
• 4.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 3 Machine Learning for Patent Identification Example Patents (Category 1) Example Patents (Category 2) Example Patents (Category 3) … Computer Machine Learning Patent Collection Model Patent result list, sorted in ✓ Category 1 ✓ Category 2 ✓ Category 3 ✓ … Model Computer
• 5.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 4 Main Benefits of Machine Learning for Patent Identification General problem with (complex) keyword searches formulated too broadly → many irrelevant hits formulated too specifically → miss potentially relevant documents ML-based models can be trained very specifically and applied to large set of documents, thus, both increasing precision and recall. This saves a substantial amount of time and increases the quality of work! (ATI=( plant% OR crop% OR cereal% OR flower% OR grass or grasses OR Arabidopsis OR (A thaliana) OR (A halleri) OR (A lyrata) OR nicotiana OR (N tabacum) OR tobacco OR tobaco OR Physcomitrella OR (P patens) OR corn% OR maize% OR Zea or (Z mays) or (Z vulgaris) OR rice% OR oryza OR oryzeae OR (O sativa) OR (O australiensis) OR (O glaberrima) OR paddy OR wildrice OR riz OR soybean% OR soya OR (Glycine max) OR (glycine hispida) OR (glycine soja) OR (G max) OR (G hispida) OR (G soja) OR (phaseolus max) OR (P max) OR (S dolichos) OR (S hispida) OR soja OR sojabean% OR sojbean% OR soyabean% OR soia OR soiabean% OR soy OR triticum OR wheat OR (T compactum) OR (T sativum) OR (T vulgare) OR (T aestivum) OR (T durum) OR (T trugidum) OR brassica OR (B napus) OR (B oleifera) OR (B campestri%) OR (B rapa) OR (B napobrassica) OR rape% OR rapeseed% OR colza% OR rapa OR canola% OR potato* OR (solanum near tuberosum) OR (solanum near esculentum) OR (lycopersicon near tuberosum) OR (lycopersicon near esculentum) OR poppy OR papaver OR medicago OR (M near truncatula) OR (M near sativa) OR (M near vulgaris) OR tomato* OR (lycopersic* near esculentum) OR (lycopersic* near lycopersicum) OR (solanum near lycopersicum) OR cotton% OR gossypium OR (g near arboreum) OR (g near barbadense) OR (g near herbaceum) OR (g near hirsutum) OR cucurbit* OR barley% OR (hordeum near vulgare) OR (hordeum near sativum) OR oat OR oats OR (avena near sativa) OR esculentum OR rye OR secale OR bean OR beans OR (phaseolus vulgaris) OR (faba vicia) OR (faba bona) OR (faba vulgaris) OR beet% OR sugarbeet% OR (beta vulgaris) OR (beta esculenta) OR cabbage% OR carrot% OR daucus OR carota OR lettuce% OR salad% or (lactuca sativa) OR (lactuca capitata) OR spinach* OR (spinacia oleracea) OR (spinacia glabra) OR (spinacia domestica) OR paprika% OR (pepper red) or (pepper sweet) or (pepper bell) OR (pepper bullnose) OR paprica% OR (sweet chillies) OR (capsicum annuum) OR (C annuum) OR (pea% near garden) OR (pea common) OR (pea green) OR (pea shelling) OR (pea field) OR (pea grey) OR (pisum speciosum) OR (pisum arvense) OR (pea sativum) OR grape% OR grapevine% OR vine% OR (vitis vinifera) OR mustard% OR sinapis OR (s alba) OR (s hirta) OR cacao OR caotree% OR (theobroma cacao) OR tea OR teas OR (thea sinensis) OR (camellia sinensis) OR coffee OR coffeetree% OR (coffea arabica) OR (coffea vulgaris) OR (coffea canephora) OR (coffea robusta) OR (coffea laurentzi) OR (coffea liberica) OR sugarcane% OR …
• 6.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 How do I integrate Patent Monitor into my daily work? Stand-alone application Integration in your IT environment Full Patbase/PatKM connectivity Full integration in CENTREDOC Rapid5
• 7.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 6 ➢ Add training examples in PatBase folders ➢ (Directly) Import folders in Patent Monitor ➢ Train the classifier ➢ Add patents to be categorized to PatBase folder ➢ (Directly) Import folder in Patent Monitor ➢ Classify documents and inspect/export results ➢ Create Alert in PatKM ➢ Configure Workflow in Patent Monitor to create weekly inspections for categories ➢ View weekly inspections directly in PatKM (!) Step-by-step workflow with PatBase @Syngenta Setup Ad-hoc categorization Weekly alerting
• 8.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Setup: Create folders in PatBase
• 9.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Setup: Import folders from PatBase
• 10.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 9 Setup
• 11.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 10 ➢ Add training examples in PatBase folders ➢ (Directly) Import folders in Patent Monitor ➢ Train the classifier ➢ Add patents to be categorized to PatBase folder ➢ (Directly) Import folder in Patent Monitor ➢ Classify documents and inspect/export results ➢ Create Alert in PatKM ➢ Configure Workflow in Patent Monitor to create weekly inspections for categories ➢ View weekly inspections directly in PatKM (!) Step-by-step workflow with PatBase @Syngenta Setup Ad-hoc categorization Weekly alerting
• 12.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Ad-hoc categorization
• 13.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 12 ➢ Add training examples in PatBase folders ➢ (Directly) Import folders in Patent Monitor ➢ Train the classifier ➢ Add patents to be categorized to PatBase folder ➢ (Directly) Import folder in Patent Monitor ➢ Classify documents and inspect/export results ➢ Create Alert in PatKM ➢ Configure Workflow in Patent Monitor to create weekly inspections, sorted in categories ➢ View weekly inspections directly in PatKM (!) Step-by-step workflow with PatBase @Syngenta Setup Ad-hoc categorization Weekly alerting
• 14.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 13 Weekly alerting PatKM Alert (weekly/monthly) PatKM Archive
• 15.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 14 Weekly alerting PatKM Alert (weekly/monthly) PatKM Archive
• 16.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 15 Weekly alerting
• 17.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 16 PatKM Archive … publishing classification results in PatBase Express will come soon!
• 18.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Patent Monitor reduces manual effort by 90% to identify relevant patents, with an accuracy of > 97%
• 19.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 18 Summary & Take Away Machine learning for categorization is a great improvement for patent identification less irrelevant hits, more relevant hits pre-sorting of results 80% of time saving in average Same accuracy compared to manual categorization by IP experts Seamless integration into PatBase and PatKM workflows (but other data sources supported as well) Also integrated in CENTREDOCs RAPID5 Outlook: PatBase Express integration Save time and focus on relevant documents only ! Increase the quality of your work !
• 20.
  Machine Learning basedPatent & Literature Categorization averbis.com 2022 Find. Understand. Predict. Interested? Get in Touch Kornél Markó [email protected]