Commonsense in Parts: Mining Part-Whole Relations from the Web and Image Tags Niket Tandon1, Charles Hariman1, Jacopo Urbani1,2, Anna.

Slides:



Advertisements
Similar presentations
Writing constructed response items
Advertisements

Strategies for Success
CMo: When Less Is More Yevgen Borodin Jalal Mahmud I.V. Ramakrishnan Context-Directed Browsing for Mobiles.
Linking Entities in #Microposts ROMIL BANSAL, SANDEEP PANEM, PRIYA RADHAKRISHNAN, MANISH GUPTA, VASUDEVA VARMA INTERNATIONAL INSTITUTE OF INFORMATION TECHNOLOGY,
6-1 Chapter Goals Determine whether a problem is suitable for a computer solution Describe the computer problem-solving process and relate it to Polya’s.
Shashank Mohan ITC105 Part A – Power Point Presentation.
TEMPLATE DESIGN © Identifying Noun Product Features that Imply Opinions Lei Zhang Bing Liu Department of Computer Science,
Sunita Sarawagi.  Enables richer forms of queries  Facilitates source integration and queries spanning sources “Information Extraction refers to the.
A Web of Concepts Dalvi, et al. Presented by Andrew Zitzelberger.
DLSI Lexical Analysis Prof Brook Wu and Ph.D. student Xin Chen.
CMPUT 301: Lecture 25 Graphic Design Lecturer: Martin Jagersand Department of Computing Science University of Alberta Notes based on previous courses by.
Statistical Relational Learning for Link Prediction Alexandrin Popescul and Lyle H. Unger Presented by Ron Bjarnason 11 November 2003.
Chapter 6 Problem Solving and Algorithm Design Nell Dale John Lewis.
Specific Learning Difficulties: Dyslexia is one of many labels for a Specific Learning Difficulty. Other Labels for other Learning Difficulties include:
Bieber et al., NJIT © Slide 1 Digital Library Integration Masters Project and Masters Thesis Summer and Fall 2005 CIS 786 / CIS Fall.
Business Modeling Domain Modeling Source: Use Case Driven Object Modeling with UML – A Practical Approach By Doug Rosenberg ISBN:
12 Ways MAP Data Can Be Used in a School. 12 Ways To Use MAP Data Monitor Academic Growth Using National Norms Identify Individual Reading Pathway using.
Graphic Design.  Graphic Designers create designs using print, electronic, and film media.  About 26% of Graphic designers are self employed.  To be.
COMP423: Intelligent Agent Text Representation. Menu – Bag of words – Phrase – Semantics – Bag of concepts – Semantic distance between two words.
Survey of Semantic Annotation Platforms
A hybrid method for Mining Concepts from text CSCE 566 semester project.
Date : 2014/09/18 Author : Niket Tandon, Gerard de Melo, Fabian Suchanek, Gerhard Weikum Source : WSDM’14 Advisor : Jia-ling Koh Speaker : Shao-Chun Peng.
Personalized Interaction with Web Resources First Sino-German Symposium on KNOWLEDGE HANDLING: REPRESENTATION, MANAGEMENT AND PERSONALIZED APPLICATION.
Learning Models for Object Recognition from Natural Language Descriptions Presenters: Sagardeep Mahapatra – Keerti Korrapati
Annotating Words using WordNet Semantic Glosses Julian Szymański Department of Computer Systems Architecture, Faculty of Electronics, Telecommunications.
Semantic Enrichment of Ontology Mappings: A Linguistic-based Approach Patrick Arnold, Erhard Rahm University of Leipzig, Germany 17th East-European Conference.
Intelligent Database Systems Lab 國立雲林科技大學 National Yunlin University of Science and Technology 1 Visualizing Ontology Components through Self-Organizing.
Natural Language Questions for the Web of Data Mohamed Yahya, Klaus Berberich, Gerhard Weikum Max Planck Institute for Informatics, Germany Shady Elbassuoni.
Natural Language Questions for the Web of Data Mohamed Yahya 1, Klaus Berberich 1, Shady Elbassuoni 2 Maya Ramanath 3, Volker Tresp 4, Gerhard Weikum 1.
Bootstrapping for Text Learning Tasks Ramya Nagarajan AIML Seminar March 6, 2001.
Google’s Deep-Web Crawl By Jayant Madhavan, David Ko, Lucja Kot, Vignesh Ganapathy, Alex Rasmussen, and Alon Halevy August 30, 2008 Speaker : Sahana Chiwane.
EXPLOITING DYNAMIC VALIDATION FOR DOCUMENT LAYOUT CLASSIFICATION DURING METADATA EXTRACTION Kurt Maly Steven Zeil Mohammad Zubair WWW/Internet 2007 Vila.
Basic Concepts of Component- Based Software Development (CBSD) Model-Based Programming and Verification.
Natural Language Questions for the Web of Data 1 Mohamed Yahya, Klaus Berberich, Gerhard Weikum Max Planck Institute for Informatics, Germany 2 Shady Elbassuoni.
Multiple Intelligences Ways to learn. 2 Yesterday, we took a test to determine our “learning style” Yesterday, we took a test to determine our “learning.
Authors: Marius Pasca and Benjamin Van Durme Presented by Bonan Min Weakly-Supervised Acquisition of Open- Domain Classes and Class Attributes from Web.
Department of Software and Computing Systems Research Group of Language Processing and Information Systems The DLSIUAES Team’s Participation in the TAC.
Getting Boys into Reading Garth Stahl Bacons College 20/1/11.
Engaging students in Shakespeare Presented by Claire Amos
Natural Language Questions for the Web of Data Mohamed Yahya, Klaus Berberich, Gerhard Weikum Max Planck Institute for Informatics, Germany Shady Elbassuoni.
McGraw-Hill/Irwin © 2012 The McGraw-Hill Companies, Inc. All rights reserved. Obtaining Valid and Reliable Classroom Evidence Chapter 4:
Named Entity Disambiguation on an Ontology Enriched by Wikipedia Hien Thanh Nguyen 1, Tru Hoang Cao 2 1 Ton Duc Thang University, Vietnam 2 Ho Chi Minh.
Semantic web Bootstrapping & Annotation Hassan Sayyadi Semantic web research laboratory Computer department Sharif university of.
Commonsense Reasoning in and over Natural Language Hugo Liu, Push Singh Media Laboratory of MIT The 8 th International Conference on Knowledge- Based Intelligent.
Answer Mining by Combining Extraction Techniques with Abductive Reasoning Sanda Harabagiu, Dan Moldovan, Christine Clark, Mitchell Bowden, Jown Williams.
Acquisition of Categorized Named Entities for Web Search Marius Pasca Google Inc. from Conference on Information and Knowledge Management (CIKM) ’04.
Intelligent Database Systems Lab N.Y.U.S.T. I. M. 1 Mining knowledge from natural language texts using fuzzy associated concept mapping Presenter : Wu,
GoRelations: an Intuitive Query System for DBPedia Lushan Han and Tim Finin 15 November 2011
Sentiment Analysis Using Common- Sense and Context Information Basant Agarwal 1,2, Namita Mittal 2, Pooja Bansal 2, and Sonal Garg 2 1 Department of Computer.
Selection and Use of Supplementary Materials and Activities
COMP423: Intelligent Agent Text Representation. Menu – Bag of words – Phrase – Semantics Semantic distance between two words.
To my presentation about:  IELTS, meaning and it’s band scores.  The tests of the IELTS  Listening test.  Listening common challenges.  Reading.
SB English 12 Thursday, September 25 th, I Remember… Collect HW Display examples?
Elaboration: Iteration 2. Elaboration: Iteration 2 Basics Iteration 1 ends with : All the software has been tested: The idea in the UP is to do early,
The Great Gatsby Unit Plan Focusing on The American Dream.
1 Commonsense Reasoning in and over Natural Language Hugo Liu Push Singh Media Laboratory Massachusetts Institute of Technology Cambridge, MA 02139, USA.
Topics Beyond the imitation game Reading the web Turing test
HowToKB: Mining HowTo Knowledge from Online Communities
Exploring Scholarly Data with Rexplore
Commonsense Knowledge Acquisition and Applications
Knowlywood: Mining Activity Knowledge from Hollywood Narratives
Acquiring Comparative Commonsense Knowledge from the Web
Enriching Structured Knowledge with Open Information
Discovering Emerging Entities with Ambiguous Names
CSE 635 Multimedia Information Retrieval
Text Analytics and Machine Learning Workshop
Effective Entity Recognition and Typing by Relation Phrase-Based Clustering
Improving Machine Learning using Background Knowledge
Aishwarya Singh 03/18/2019 Acquiring Comparative Commonsense Knowledge from the Web Niket Tandon, Gerard de Melo, Gerhard Weikum.
Visual Grounding.
Presentation transcript:

Commonsense in Parts: Mining Part-Whole Relations from the Web and Image Tags Niket Tandon1, Charles Hariman1, Jacopo Urbani1,2, Anna Rohrbach1, Marcus Rohrbach3, Gerhard Weikum1 1) Max Planck Institute, Germany 2) VU University, Amsterdam 3) ICSI, Berkeley

Motivation Humans can recognize objects easily.

Motivation handle partof cycle atom partof cycle For vision systems, commonsense background knowledge helps for structure prediction e.g. Having seen wheel and handle, increases the confidence of seeing cycle. External resources mapped to wordnet.

Motivation handle partof cycle atom partof cycle?? 1 handle partof cycle 2 tires partof cycle 3+ spokes partof tire 1 basket rack partof cycle … Bigger and richer WordNet.

Motivation We wish such a bigger and richer WordNet exists! handle partof cycle atom partof cycle?? 1 handle partof cycle 2 tires partof cycle 3+ spokes partof tire 1 basket rack partof cycle … Bigger and richer WordNet.

1 seat is a usually visible, physical part of a cycle What is the problem Other example: screen (projector for movies) screen (electronic display) screen (strainer to sieve) notebook (a book with blank pages) notebook (a small portable computer)

seat (center of authority) cycle (rhythm, oscillation) cycle (bicycle) 1 2 3+ ∞ visible invisible seat (reservation) seat (support to sit) seat (center of authority) cycle (rhythm, oscillation) cycle (bicycle) 1 seat is a usually visible, physical part of a cycle member of substance of physical part of exploration space very large other challenges other difficulties

1 2 3+ ∞ visible invisible seat (reservation) seat (support to sit) seat (center of authority) cycle (rhythm, oscillation) cycle (bicycle) 1 seat is a usually visible, physical part of a cycle member of substance of physical part of exploration space very large other challenges other difficulties Unstructured and ambiguous {cyclist, seat} is part of a {team, cycle} 1 seat, 2 brakes, 2 tires part of cycle!! the seat is visible while the atom is not!!

seat (center of authority) 1 seat is a usually visible, seat (reservation) seat (support to sit) seat (center of authority) 1 seat is a usually visible, physical part of a cycle cycle (rhythm, oscillation) cycle (bicycle) Problem statement

How can we fill this gap?

seat, cycle seat, sports cycle WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat Score Reason Enrich

seat, cycle seat, sports cycle WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat wheel, physical part of, cycle wheels are essential parts of a cycle wheels on sale along with the cycle * are essential parts of * seats are essential parts of a cycle dollars are an essential part of life snake is an essential part of machine Score Reason Enrich

Noun phrase aware wrapper over IMS: sports cycle -> cycle#6 * are essential parts * * on sale along with * PE are essential parts of PE seat, cycle dollars, life snake, machine Noun phrase aware wrapper over IMS: sports cycle -> cycle#6 Domain ( r ) Relation: r Range ( r ) Physical Entity Physical part of (P) wheel, cycle Physical + Abstract Entity Member of (M) cyclist, team Abstract Entity Substance Entity Substance of (S) rubber, wheel

Noun phrase aware wrapper over IMS: sports cycle -> cycle#6 * are essential parts * * on sale along with * PE are essential parts of PE seat, cycle dollars, life snake, machine Noun phrase aware wrapper over IMS: sports cycle -> cycle#6 Score of a candidate ak (either pattern or assertion) Seed Support: many distinct seed matches implies high score Strength: Penalize semantic drift of candidates that match seeds of different relations (P,M,S) Domain ( r ) Relation: r Range ( r ) Physical Entity Physical part of (P) wheel, cycle Physical + Abstract Entity Member of (M) cyclist, team Abstract Entity Substance Entity Substance of (S) rubber, wheel

seat, cycle seat, sports cycle WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat Score seat is part of a seat seat part of wheel wheel part of seat spoke part of wheel wheel part of cycle => spoke part of cycle sports cycle isA cycle wheel part of sports cycle Reason Enrich

seat, cycle seat, sports cycle WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat Score Reason cycle, fun, trip, go, seat, niket seat is visible part of cycle Are seats countable? [seat/seats] en: seats are part of cycles it: una sede è parte di un ciclo Enrich

Score Reason Enrich visible wheel, cycle seat, cycle seat, seat WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat Score Reason Enrich

PhysicalPart Member Substance Visibility Cardinality Overall 89 % 96 % 71 % 98 % 80 % 6.65 M 0.04 M 0.06 M 0.74 M 6.69 M 6.75 M Replacing WordNet by PWKB for image detection task, provides 10% relative increase in accuracy

from languages to languages (better cardinality estimates) WN visible wheel, cycle seat, cycle seat, seat seat, cycle seat, sports cycle 1 seat from text to concepts (helps reasoning) wheel part of cycle ^ sports cycle typeof cycle => wheel part of sports cycle from languages to languages (better cardinality estimates) from image to text (additionally quasi visual verification) from text to image (presented a use-case for better object detection) http://tinyurl.com/partwholekb