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)
11 Oct 2022•0 j'aime•269 vues
Télécharger pour lire hors ligne
Signaler
Internet
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)
Similaire à 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)
Similaire à 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)(20)
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. Machine learning based patent categorization:
A success story in monitoring a complex technology
with high patenting activity
Susanne Tropf, Syngenta Crop Protection AG
AI-SDV 2022, Vienna, Austria
2. 2
Topics
❑ Genome editing – a complex technology with high patenting activity
❑ Will A.I. based automatic patent categorization using Averbis’ Patent Monitor
drive efficiency in monitoring genome editing technologies?
❑ Process of using A.I. based automatic patent categorization for weekly patent
monitoring of genome editing technologies
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
3. 3
Genome editing – a complex technology with high patenting
activity
Sept 6, 2022
CRISPR/Cas9
Discovered 2012; Nobel Prize 2020; efficient and highly selective
Search for ‘CRISPR Cas9’
Sept 6, 2022
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Genome Editing (genome engineering, gene editing)
Type of genetic engineering in which DNA is inserted, deleted, modified
or replaced in the genome of a living organism.
Genome editing introduces modifications at site specific locations in the genome.
Applications: agriculture, pharma, biotechnology
Tool Box: e.g. meganucleases, zinc finger nucleases, TALEN, CRISPR/Cas9
... about 332,000 results
... about 121,779 results
4. 4
Genome editing (GE) technologies from an IP Analyst’s
perspective
Syngenta Interest
❑ Plant Applications
❑ Platform Technologies
Challenges for IP Analyst
❑ Retrieval: Precision vs recall
‒ Broad technology coverage affords complex retrieval
strategies.
‒ Patent classes alone not sufficient.
‒ Platform technologies often with generic claims for
use.
❑ High Data Volume
❑ Technical Assessment: time consuming
‒ Similar claim language.
‒ Complexity of technology.
‒ Data Volume.
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
5. 5
Patent monitoring of GE technologies – the beginnings
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Professional
search
Minesoft
• 2 x / year
• precision < recall
IP
Land-
scape
• weekly
• precision > recall
Project
Alert
IP Analyst: value add
- Relevance assessment.
- Complex categorization based on a
hierarchical labeling scenario.
GE team: read
- Relevance & technical assessment.
- ~25 % out of scope.
Retrieval Knowledge
Information
6. 6
- 1.5 x increase: 2018 to 2022
- currently ~43 new patent
publications/week (> 2 k/year)
Patent monitoring of GE technologies – filing trends
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Professional
search
Minesoft
• 2 x / year
• precision < recall
IP
Land-
scape
• weekly
• precision > recall
Project
Alert
Retrieval Filing Trends
Priority Year (earliest)
2010-2019
7. 7
Will A.I. based automatic patent categorization using
Averbis’ Patent Monitor drive efficiency in monitoring
genome editing technologies?
Retrieval of patents disclosing genome editing technologies results in low precision at good recall.
At the same time, the number of new filings is steadily increasing.
Hence, analysis of documents is time consuming.
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
8. 8
A.I. based automatic patent categorization using Averbis’
Patent Monitor – Process
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Retrieval designed for recall.
Collection of search results
in PatBase (Minesoft).
Provide
training sets1 Train classifier
Predict
categories
Due to complexity of technology: two-step approach
1. ‘in scope’ vs ‘out of scope’
training set ‘in scope’: 2,548
training set ‘out of scope’: 1,205
2. ‘Platform Technologies’ vs ‘Plant Applications’
training set ‘Platform Technologies’: 1,740
training set ‘Plant Applications’: 808
1: Expert provides training sets, i.e. expert knowledge pre-requisite for set up.
9. 9
A.I. based automatic patent categorization using Averbis’
Patent Monitor – Analysis of labels assigned to document set1
1. Label: ‘out of scope’ vs ‘in
scope’
2. Label: ‘Platform Technology’ vs
‘Plant Application’
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
1: Analysis based on 2,295 patent families categorized in Patent Monitor.
in scope:
86 % confirmed
out of scope:
99 % confirmed
Platform Technology vs Plant Application:
93 % confirmed
10. 10
✓ Improved Recall.
✓ Reliable exclusion of documents categorized as ‘out of scope’.
✓ Within the document set ‘in scope’ > 80 % confirmed.
✓ High accuracy labeling by high level category ‘Plant
Applications’ vs ‘Platform Technologies’.
✓ Read 648 instead of 2,295 patents.
99 % precision
Learnings from A.I. based automatic patent categorization
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Retrieval Knowledge
Information
Minesoft IP Analyst
& GE team
Better Recall. Less to read. More time for value add & knowledge extraction.
14 % noise
instead of 72 %
> 70 % time
savings
11. 11
Process of using A.I. based automatic patent
categorization for weekly patent monitoring of genome
editing technologies
A classifier could successfully be implemented for selecting documents on genome editing
technologies.
Use this classifier for pre-selecting relevant documents for monitoring new patent publications on
genome editing technologies on a weekly basis.
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
12. 12
Monitor
• weekly
• optimized
recall
Categorize
• relevance
assessment
Publish
• read less
• more time for
knowledge
creation
Monday
Sunday
Process of using A.I. based automatic patent categorization for
weekly patent monitoring
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Data
Transfer
Data
Transfer
Retrieval Knowledge
Information
14. 14
Conclusions
❑ A.I. based automatic patent categorization has been successfully and reliably implemented
for monitoring a complex technology with high patent activity by excluding unrelevant
documents and labeling of relevant documents by top level category.
❑ Benefits:
✓ Achieve better recall.
✓ At the same time: Read less. Focus on value add & knowledge extraction.
❑ Next steps – weekly monitoring: Publish results in PatBase Express.
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022
Monitor Categorize Publish
15. 15
Thank you for your attention.
Questions?
Thanks to:
Averbis Team
Minesoft Team
Syngenta’s IP Analysis Team
Susanne Tropf, Syngenta Crop Protection AG AI-SDV, Vienna 2022