Emerging Trends in IP Searching:
Machine/AI Based Search Strategies
Anoop Cheeran & Manish Badani
Philips IP&S/IP Analysis
September 27, 2016
September 27, 2016
Philips IP Analysis Team
IP&S IP Analysis
IP Search
IP Intelligence
Market Intelligence
2
3 IP&S IP AnalysisSeptember 27, 2016
Search Tasks
• Patentability search for Invention Disclosures
• Validity/Invalidity Searches
• Product Risk Assessment Searches
• Trade Mark Searches
• Landscape Searches
• Company Analysis
• Market Intelligence
September 27, 2016
Patentability Search
Building the IP Assets of Philips
IP&S IP Analysis
Input
• Abstract
• Invention Disclosure
• Prior art (if any)
Output
• Search Criteria
• Prior art
Challenges
• Large number of patent applications filed
each year
• Filings in several different languages
• Traditional keyword search is increasingly
inadequate
Stringent Quality
Checks
4
Source: http://ipstats.wipo.int/ipstatv2/
September 27, 2016
Artificial/Machine Intelligence
Application in IP Search
Artificial Intelligence
• Simulate human like intellect and/or behavior by a
machine
Natural Language Processing
A sub-field of Artificial Intelligence
Computers/Search Engines can interact with
human beings like human
Semantic Search
Understands searcher’s intent and context
Major web search engines like Google
incorporate some elements of Semantic
search
IP&S IP Analysis5
Source: http://www.slideshare.net/ecodrew/future-of-artificial-
intelligence-iftf-fujitsu-labs-2015
Semantic Search
Situations when it helps
6 IP&S IP AnalysisSeptember 27, 2016
Source: http://www.slideshare.net/fullscreen/ABLVienna/semantic-search-8159539/4
7 IP&S IP AnalysisSeptember 27, 2016
Keyword Search vs Semantic Search
Semantic search helps
• Reduce the time of running the search
• Improves RECALL and PRECISION
• Uncover relevant patents that would
otherwise be missed with keyword
search
Semantic Search - Orbit
Free Text
8 IP&S IP AnalysisSeptember 27, 2016
• Find similar patents based on freely entered text
• One paragraph is required for the search to run
• If the text is entered in non-English language, it gets translated by Google
9 IP&S IP AnalysisSeptember 27, 2016
Semantic Search - Orbit
Concept Selection
• Concepts are extracted and shown in the
selection window
• Additional concepts can be searched and
added to the list
• Based on the selected concepts relevant
patent families are searched and listed
Semantic Search - Relecura
10 IP&S IP AnalysisSeptember 27, 2016
• Semantic search
based on uploaded
text/file
• Suggested keywords
and concepts
• Seed Data to quickly
zoom into relevant
documents
• Search trail to track
selected concepts
and keywords
11 IP&S IP AnalysisSeptember 27, 2016
Semantic Search - AutoMatch
www.auto-match.se
• AutoMatch is a natural language text based search tool for patents and patent
applications
• Tool creates a finger print reflecting the given input textual content (for example, from
an abstract) that is adapted and matched against the tool’s IP publication repository
• Customization of technology areas (search index) possible
• Provides up to 100 hits per search
https://www.auto-match.se/Guide%20AutoMatch%20Evaluation.pdf
12 IP&S IP AnalysisSeptember 27, 2016
Can copy paste directly from
an invention disclosure or
can provide a descriptive
summary encompassing the
keywords as the search
query
Tool uses text in title and
the descriptive summary to
match and provides the
result set
https://www.automatch.se/Guide%20AutoMatch%20
Evaluation.pdf
Step 1: Register an idea: Concise summary
text (without truncations) about half a
page disclosing the subject area
Choose the technical index (default AM Master
covering all areas or the optimized index covering
specific technical fields) and run the query
Semantic Search - AutoMatch
September 27, 2016 IP&S IP Analysis13
Semantic Search - AutoMatch
Tool has an option of ranking
each of the retrieved
references to ‘match, relevant,
background & noise’
Can add annotations for each
document analyzed
Can add new text into the idea
summary and create a new
search
Can download matched results
to an Excel report
Summary
https://www.auto-match.se/Guide%20AutoMatch%20Evaluation.pdf
September 27, 2016 IP&S IP Analysis14
Citation Based Search - Ambercite
Citation based tool that has 7 generations of citations; 3 earlier generations, 3 later
generations and the seed document (citations of citations of citations)
The relevancy of a document is based on the direct and indirect connection of the
document with the seed document(s)
The citations are based on individual records and not patent family; all members of
patent families may need to be added as a seed document
http://www.ambercite.com/index.php/support/getting-started
Ambercite: Ranking Method
September 27, 2016 IP&S IP Analysis15
Direct Citations
In-Direct Citations
Higher number of
shared citations
(direct & indirect)
result in higher
similarity ranking
Indirect citation
analysis identifies
references otherwise
not obtained through
direct citation analysis
http://www.ambercite.com/index.php/support/getting-started
Citation Based Search - Ambercite
September 27, 2016 IP&S IP Analysis16
http://www.ambercite.com/index.php/support/getting-started
Citation Based Search - Ambercite
Searching based on relevant patent
numbers, avoiding need to make
predictions on likely keywords or
class codes
Listing of direct patent citations,
and indirect ‘unknown’ citations in
search results
Tool provides an ‘AmberScore’, a
similarity metric based on
citation(s) to each document
Recent patent/patent applications
may have a lower AmberScore
Option for ‘hiding’ known
documents from search results and
a ‘like’ button to categorize
documents
Summary
Summary
17 IP&S IP AnalysisSeptember 27, 2016
• Large number of patent applications in various languages have increased
challenges for searching prior art
• Several patent search tools are now available that utilize natural language
processing systems
• Machine intelligence based on citations is also used in identifying related patents
• The AI/MI based search tools can significantly reduce the search time and improve
the search accuracy
• At the moment, there is still room for improvement of these tools for them to
exactly mimic a human
Until then, there will always be a need for the traditional IP Search Analyst ☺
Emerging trends in IP searching: Machine/AI based searching - Anoop Cheeran & Manish Badani

Emerging trends in IP searching: Machine/AI based searching - Anoop Cheeran & Manish Badani

  • 1.
    Emerging Trends inIP Searching: Machine/AI Based Search Strategies Anoop Cheeran & Manish Badani Philips IP&S/IP Analysis September 27, 2016
  • 2.
    September 27, 2016 PhilipsIP Analysis Team IP&S IP Analysis IP Search IP Intelligence Market Intelligence 2
  • 3.
    3 IP&S IPAnalysisSeptember 27, 2016 Search Tasks • Patentability search for Invention Disclosures • Validity/Invalidity Searches • Product Risk Assessment Searches • Trade Mark Searches • Landscape Searches • Company Analysis • Market Intelligence
  • 4.
    September 27, 2016 PatentabilitySearch Building the IP Assets of Philips IP&S IP Analysis Input • Abstract • Invention Disclosure • Prior art (if any) Output • Search Criteria • Prior art Challenges • Large number of patent applications filed each year • Filings in several different languages • Traditional keyword search is increasingly inadequate Stringent Quality Checks 4 Source: http://ipstats.wipo.int/ipstatv2/
  • 5.
    September 27, 2016 Artificial/MachineIntelligence Application in IP Search Artificial Intelligence • Simulate human like intellect and/or behavior by a machine Natural Language Processing A sub-field of Artificial Intelligence Computers/Search Engines can interact with human beings like human Semantic Search Understands searcher’s intent and context Major web search engines like Google incorporate some elements of Semantic search IP&S IP Analysis5 Source: http://www.slideshare.net/ecodrew/future-of-artificial- intelligence-iftf-fujitsu-labs-2015
  • 6.
    Semantic Search Situations whenit helps 6 IP&S IP AnalysisSeptember 27, 2016 Source: http://www.slideshare.net/fullscreen/ABLVienna/semantic-search-8159539/4
  • 7.
    7 IP&S IPAnalysisSeptember 27, 2016 Keyword Search vs Semantic Search Semantic search helps • Reduce the time of running the search • Improves RECALL and PRECISION • Uncover relevant patents that would otherwise be missed with keyword search
  • 8.
    Semantic Search -Orbit Free Text 8 IP&S IP AnalysisSeptember 27, 2016 • Find similar patents based on freely entered text • One paragraph is required for the search to run • If the text is entered in non-English language, it gets translated by Google
  • 9.
    9 IP&S IPAnalysisSeptember 27, 2016 Semantic Search - Orbit Concept Selection • Concepts are extracted and shown in the selection window • Additional concepts can be searched and added to the list • Based on the selected concepts relevant patent families are searched and listed
  • 10.
    Semantic Search -Relecura 10 IP&S IP AnalysisSeptember 27, 2016 • Semantic search based on uploaded text/file • Suggested keywords and concepts • Seed Data to quickly zoom into relevant documents • Search trail to track selected concepts and keywords
  • 11.
    11 IP&S IPAnalysisSeptember 27, 2016 Semantic Search - AutoMatch www.auto-match.se • AutoMatch is a natural language text based search tool for patents and patent applications • Tool creates a finger print reflecting the given input textual content (for example, from an abstract) that is adapted and matched against the tool’s IP publication repository • Customization of technology areas (search index) possible • Provides up to 100 hits per search https://www.auto-match.se/Guide%20AutoMatch%20Evaluation.pdf
  • 12.
    12 IP&S IPAnalysisSeptember 27, 2016 Can copy paste directly from an invention disclosure or can provide a descriptive summary encompassing the keywords as the search query Tool uses text in title and the descriptive summary to match and provides the result set https://www.automatch.se/Guide%20AutoMatch%20 Evaluation.pdf Step 1: Register an idea: Concise summary text (without truncations) about half a page disclosing the subject area Choose the technical index (default AM Master covering all areas or the optimized index covering specific technical fields) and run the query Semantic Search - AutoMatch
  • 13.
    September 27, 2016IP&S IP Analysis13 Semantic Search - AutoMatch Tool has an option of ranking each of the retrieved references to ‘match, relevant, background & noise’ Can add annotations for each document analyzed Can add new text into the idea summary and create a new search Can download matched results to an Excel report Summary https://www.auto-match.se/Guide%20AutoMatch%20Evaluation.pdf
  • 14.
    September 27, 2016IP&S IP Analysis14 Citation Based Search - Ambercite Citation based tool that has 7 generations of citations; 3 earlier generations, 3 later generations and the seed document (citations of citations of citations) The relevancy of a document is based on the direct and indirect connection of the document with the seed document(s) The citations are based on individual records and not patent family; all members of patent families may need to be added as a seed document http://www.ambercite.com/index.php/support/getting-started
  • 15.
    Ambercite: Ranking Method September27, 2016 IP&S IP Analysis15 Direct Citations In-Direct Citations Higher number of shared citations (direct & indirect) result in higher similarity ranking Indirect citation analysis identifies references otherwise not obtained through direct citation analysis http://www.ambercite.com/index.php/support/getting-started Citation Based Search - Ambercite
  • 16.
    September 27, 2016IP&S IP Analysis16 http://www.ambercite.com/index.php/support/getting-started Citation Based Search - Ambercite Searching based on relevant patent numbers, avoiding need to make predictions on likely keywords or class codes Listing of direct patent citations, and indirect ‘unknown’ citations in search results Tool provides an ‘AmberScore’, a similarity metric based on citation(s) to each document Recent patent/patent applications may have a lower AmberScore Option for ‘hiding’ known documents from search results and a ‘like’ button to categorize documents Summary
  • 17.
    Summary 17 IP&S IPAnalysisSeptember 27, 2016 • Large number of patent applications in various languages have increased challenges for searching prior art • Several patent search tools are now available that utilize natural language processing systems • Machine intelligence based on citations is also used in identifying related patents • The AI/MI based search tools can significantly reduce the search time and improve the search accuracy • At the moment, there is still room for improvement of these tools for them to exactly mimic a human Until then, there will always be a need for the traditional IP Search Analyst ☺