R&D Trend Analysis of Wellbeing AI using Panoramic View Analytics

2017年3月に開催された"AAAI 2017 SPRING SYMPOSIA"で発表した資料です。

AAAI 2017 SPRING SYMPOSIA
https://www.aaai.org/Symposia/Spring/sss17.php

"WELLBEING AI: FROM MACHINE LEARNING TO SUBJECTIVITY ORIENTED COMPUTING"というセッションに参加し、Wellbeing AI関連の特許情報を使った技術動向分析を発表しました。

論文
https://www.aaai.org/ocs/index.php/SSS/SSS17/paper/view/15329/14621

Transcript

1. VALUENEX
   www.valuenex.net
   R&D Trend Analysis of Wellbeing AI
   using Panoramic View Analytics
   Mar. 29, 2017
   Takayoshi Hayashi
   AAAI 2017 SPRING SYMPOSIA
   

   View Slide

2. Introduction
   intellectual innovator
   1
   • To better understand the R&D trends of wellbeing AI, we analyze a massive
   amount of patent data.
   • We use Panoramic View Analytics that is our unique data visualization method.
   • We show the main technology areas, growing and fusion areas, as well as the
   effectiveness of panoramic view analytics.
   

   View Slide

3. Methodology: Panoramic View Analytics
   intellectual innovator
   2
   Proven scientific methods that allows the user to “see and understand” the entirety of
   large amounts of data rather than “search and read” individual pieces of data.
   A
   B
   C
   Clustering docs by calculating
   similarities among them
   Visualization of the
   similarity among docs
   Original indicators for
   mining insights
   Patent / Paper / SNS /
   News / Annual Report etc
   Big Data Clustering Visualization Analytics
   

   View Slide

4. Methodology: Panoramic View Analytics
   intellectual innovator
   3
   We can get insights for strategy-making by analyzing information (distance, density,
   white space, distribution) on the radar chart.
   The size of each cluster is
   proportional to the number of
   documents in it.
   The distance between clusters
   indicates the similarities
   between them.
   The axis has no meanings.
   Each circle is called a “cluster”.
   They contain similar documents.
   Sparse area
   Density
   area
   Far = Low
   similarity
   Near = High
   similarity
   Figure 1: Understanding the Radar Chart.
   

   View Slide

5. Methodology: Algorithm
   intellectual innovator
   4
   Morphological
   analysis
   Extract
   characteristic
   word
   Calculating
   Similarities
   between
   Documents
   Visualization
   The system divides
   sentence into words.
   Calculating each word’s weight based on the term
   frequency and the inverse document frequency.
   e.g.) The words included in documents
   Doc.A：aaa | aaa | bbb | ccc
   Doc.B：aaa | bbb | bbb | ddd
   Doc.C：aaa | bbb | ddd | ddd
   Word TF IDF Weight (TF*IDF)
   aaa 2 1/3 2/3
   bbb 1 1/3 1/3
   ccc 1 1/1 1
   Vectorizing each document based on the each word's weight,
   and calculating the similarities based on the inner product.
   Doc.A Vector
   aaa
   bbb
   ccc
   Doc.A Vector
   θ1
   Doc.C Vector
   Doc.B Vector
   θ2
   Visualizing by original algorithm based on Multi Dimensional
   Scaling method
   θ1
   θ2
   The system visualize
   the documents
   searched.
   Documents(1,2,3,…,n) Documents(1,2,3,…,n)
   The / system / visualize
   / the / documents /
   searched.
   Doc.A
   Doc.C
   Doc.B
   Doc.A Vector
   Doc.C Vector
   Doc.B Vector
   Word importance in Doc. A High-dimensional
   data representation
   (Tens of thousands)
   Modified MDS in
   order to express
   close relationship
   

   View Slide

6. Methodology: Patent Dataset of Analysis
   intellectual innovator
   5
   We made dataset S003 for analyzing Wellbeing /Healthcare and AI field.
   • Wellbeing / Healthcare: Keyword search
   • AI: IPC search (G06N)
   * IPC means International Patent Classification. And, G06N is “Computer systems based on specific
   computational models”. So, it is the IPC that represents the core technology area of AI. Also, it is used in JPO’s
   report on AI.
   No Content Search Type Search Formula
   Count
   (Rough estimate)
   S001
   Wellbeing /
   Healthcare
   Keyword search
   “Wellbeing” OR “well-being” OR
   “happy” OR “happiness” OR
   “healthcare” // Title, Abstract, Claims
   7,000
   S002 AI IPC search G06N 14,000
   S003 Total Logical expression S001 OR S002 21,000
   Search condition
   DB： PatentSQUARE
   Period： Jan. 1, 2001 – Oct. 26, 2017 (publication date)
   Authority： US (published application)
   Language： English
   

   View Slide

7. Results: Radar chart
   intellectual innovator
   6
   Left and right side represent wellbeing / healthcare and AI, respectively.
   Medical (Prescription)
   information management
   Risk (finance,
   disease), Monitoring
   Medical fraud
   detection
   Rule base
   Ontology
   Inference
   Agent
   Recommendation
   Q&A system
   Neuromorphic chip
   Neural network
   Robot
   Wellbeing / Healthcare AI
   Classification
   Figure 2: Radar chart of wellbeing / healthcare and AI.
   

   View Slide

8. Results: Fusion area
   intellectual innovator
   7
   • Extracted the clusters containing 50% to 80% G06N patents in AI side
   • The below list shows examples of invention names included in the extracted clusters.
   No. Publication no. Title (example)
   A1 20150273697 ROBOT FOR MEDICAL ASSISTANCE
   A2 20160104486
   Methods and Systems for Communicating Content to Connected Vehicle
   Users Based Detected Tone/Mood in Voice Input
   A3 20070005621 Information system using healthcare ontology
   A4 20120290516 Habituation-compensated predictor of affective response
   A5 20140149128 HEALTHCARE FRAUD DETECTION WITH MACHINE LEARNING
   A6 20120278064
   SYSTEM AND METHOD FOR DETERMINING SENTIMENT FROM TEXT
   CONTENT
   A7 20120215555
   SYSTEMS AND METHODS FOR HEALTHCARE SERVICE DELIVERY
   LOCATION RELATIONSHIP MANAGEMENT
   A8 20160156781
   SECURELY AND EFFICIENTLY TRANSFERRING SENSITIVE
   INFORMATION VIA A TELEPHONE
   A9 20060129427 Systems and methods for predicting healthcare related risk events
   A10 20080294692 Synthetic Events For Real Time Patient Analysis
   Table 1: Fusion fields of list A.
   

   View Slide

9. Results: Fusion area
   intellectual innovator
   8
   • Extracted the clusters containing 20% to 50% G06N patents in wellbeing / healthcare side
   • The below list shows examples of invention names included in the extracted clusters.
   No. Publication no. Title (example)
   B1 20100185456 Medication Management System
   B2 20160156781
   SECURELY AND EFFICIENTLY TRANSFERRING SENSITIVE INFORMATION
   VIA A TELEPHONE
   B3 20150324693
   PREDICTING DRUG-DRUG INTERACTIONS BASED ON CLINICAL SIDE
   EFFECTS
   B4 20120190937
   EMOTION SCRIPT GENERATING , EXPERIENCING , AND EMOTION
   INTERACTION
   B5 20030050797 System and user interface for processing healthcare related event information
   B6 20080164998 Location Sensitive Healthcare Task Management System
   B7 20140108025 COLLECTING AND TRANSFERRING PHYSIOLOGICAL DATA
   B8 20090248744 TRANSACTIONAL STORAGE SYSTEM FOR HEALTHCARE INFORMATION
   B9 20100205001
   SYSTEM AND METHOD FOR ASSISTING IN THE HOME TREATMENT OF A
   MEDICAL CONDITION
   B10 20140249848 DEFINING PATIENT EPISODES BASED ON HEALTHCARE EVENTS
   Table 2: Fusion fields of list B.
   

   View Slide

10. Results: Fusion area
    intellectual innovator
    9
    There are many medical area, but several other applications.
    Robot
    Connected car
    Healthcare ontology
    Medical fraud detection
    Sentimental / emotion
    analysis
    Telemedicine /
    home care
    Medical information
    security
    Medication
    management
    Risk prediction
    Patient condition
    analysis
    Predicting drug-drug
    interactions
    Utilization of
    physiological data
    Medical Other application
    Keywords map in fusion area.
    

    View Slide

11. Results: Growing area
    intellectual innovator
    10
    We detect growing areas by counting patent amount in each mesh.
    Growing
    Count
    

    View Slide

12. Results: Growing area
    intellectual innovator
    11
    For example, in AI area, “Neural network”, “Neuromorphic chip” and “Q&A system” are
    detected as growing area.
    Medical (Prescription)
    information management
    Risk (finance,
    disease), Monitoring
    Medical fraud
    detection
    Rule base
    Ontology
    Inference
    Agent
    Recommendation
    Q&A system
    Neuromorphic chip
    Neural network
    Robot
    Wellbeing / Healthcare AI
    Classification
    Figure 2: Radar chart of wellbeing / healthcare and AI.
    

    View Slide

13. Results: Growing area - Neuromorphic chip
    intellectual innovator
    12
    n Keywords
    neuron, synaptic, firing, spike, neural,
    STDP, synapse, circuit, pulse, learning
    Year
    Count
    Count
    Assignee
    n Transition
    n Player
    n Example of title
    COMPUTED SYNAPSES FOR
    NEUROMORPHIC SYSTEMS
    n Example of figure
    

    View Slide

14. Results: Growing area in the fusion area
    intellectual innovator
    13
    Robot (A1), Sentimental analysis (A6), Telemedicine / home care (A7) and Medication
    management (B1), etc are detected in growing area in the fusion area.
    Medical (Prescription)
    information management
    Risk (finance,
    disease), Monitoring
    Medical fraud
    detection
    Rule base
    Ontology
    Inference
    Agent
    Recommendation
    Q&A system
    Neuromorphic chip
    Neural network
    Robot
    Wellbeing / Healthcare AI
    Classification
    A6
    A1
    A7
    B1
    Figure 2: Radar chart of wellbeing / healthcare and AI.
    

    View Slide

15. Results: Growing area – Sentimental analysis
    intellectual innovator
    14
    n Keywords
    noun, score, polarity, corpus, sentiment,
    topic, opinion, domain, content,
    advertisement
    Count
    Count
    Assignee
    n Transition
    n Player
    n Example of title
    System and method for determining
    sentiment from text content
    n Example of figure Year
    

    View Slide

16. Conclusion
    intellectual innovator
    15
    n Purpose
    We applied Panoramic View Analytics to massive amounts of patent data for analyzing
    R&D trend in wellbeing / healthcare and AI field.
    n Methodology
    Panoramic View Analytics is our unique data visualization method of massive amounts of
    document data. You can easily find trends and create R&D strategy.
    n Result
    The fusion and growing areas: “robotics”, “sentimental analysis”, “telemedicine / home
    care” and “medication management systems”, etc.
    n Future work
    • To analyze scientific paper data on wellbeing AI.
    • To analyze specific field discussed in this symposium such as sleep, body motion etc.
    

    View Slide

17. About VALUENEX
    intellectual innovator
    16
    n Analysis Topic
    1.Technology Trend Analysis
    2.SWOT Analysis
    3.Searching Application Areas for Client’s Technology
    4.Investigating Alliance Strategy (Technology Synergy Analysis)
    n ASP service
    Panoramic View Analytics software that can analyze not only patent but also research paper,
    marketing data, etc.
    n API service
    API that can be embedded in client’s web site, internal system, etc.
    n Consultation
    We provide R&D strategy-making consultation by using panoramic view analytics.
    n Coaching
    We support the development of human capital & organizations to utilize data science.
    

    View Slide

18. 17
    Electronics
    Transportation
    Equipment
    Chemicals
    Other
    Government, Public
    Research Organization,
    Universities
    Other Manufacturing
    23%
    15%
    15%
    20%
    13%
    14%
    Industry
    (2013-15)
    We provide analytics services (ASP and over 100 consulting projects/year,
    etc.) to leading companies.
    About VALUENEX
    To name a few:
    • Honda motors
    • Panasonic
    • OMRON
    • Otsuka HD
    • Mitsubishi UFJ Morgan Stanley
    • LIXIL
    • Sumitomo Bakelite
    • The Chugoku Electric Power
    • IPICS
    • National Graduate Institute for Policy
    Studies
    Much more
    150+
    

    View Slide

19. 18
    Takayoshi Hayashi
    hayashi@valuenex.com
    www.valuenex.net
    VALUENEX Inc.
    Thank you
    

    View Slide