Advisor: Hsin-Hsi Chen Reporter: Chi-Hsin Yu Date: 2009.12.10 From K-CAP 2009 (ACM conference, 5 th Knowledge CAPture, Begins in 2001, biennially)

Method for Extracting Commonsense Knowledge

Advisor: Hsin-Hsi ChenReporter: Chi-Hsin YuDate: 2009.12.10

From K-CAP 2009 (ACM conference, 5th

Knowledge CAPture, Begins in 2001, biennially)

Outlines

Introduction Extended WordNet Knowledge Base Proposed Method for Extracting

Commonsense Knowledge Implementation and Results Applications of Commonsense

Knowledge Conclusion

Introduction

Commonsense › An average person is assumed to have

commonsense knowledge. Including spatial, physical, social, temporal,

psychological, and others

› This knowledge is not communicated most of the time. (assumption)

Introduction (conti.)

Related Works› Expert-build: Cyc› Collaborative efforts: OpenMind project

(ConceptNet), Mindpixel project, Verbosity Some proposals link the information

obtained by the collaborative effort to known ontologies to expand and structure the commonsense knowledge

Extended WordNet Knowledge Base (XWN-KB)

XWN (eXtended WordNet) › the words in the WordNet glosses

Semantically disambiguated syntactically parsed, Part-Of-Speech tagged transformed in logic forms

Extended WordNet Knowledge Base (XWN-KB)

XWN-KB is an upper ontology built on top of Extended WordNet (XWN).› semantically parsed and transformed

into semantic triples

Extended WordNet Knowledge Base (XWN-KB)

Examples

Timepiece:計時器

Opening:穴 , 孔 ; 通道

Proposed Method

Using › semantic information › Semi-automatic algorithm› 3 meta-rules › 27 commonsense rules

Proposed Method – Metarule

Hyponym: type-of, 下義詞

Proposed Method – Metarule

Proposed Method – Commonsense Rules

Obesity: 肥胖 ; 過胖

Proposed Method – The Algorithm

The array of commonsense rules, I[.]

Proposed Method – The Algorithm

Proposed Method – The Algorithm

Hyponym: type-of, 下義詞

Proposed Method – The Algorithm

Meronym: Part-Of

Implementation and Results

Implementation and Results

Implementation and Results

Applications of Commonsense Knowledge

Question Answering:

Text Entailment:

Specialty:專業 ; 專長

jimmies: 灑在冰淇淋上的巧克力片

Conclusion

The proposed method extracts high precision commonsense rules. › Limitation

Commonsense rule the incomplete semantic relations in

WordNet

My Opinions About Commonsense knowledge extraction

How do we define “commonsense knowledge” in definitional way? › Is it needed to have a definition? › Is it related to knowledge categorization? › 對知識分類 , 且是對沒有寫在紙上的知識分類

依假設 , 人有 commonsense, 但少明確說出來 How do we judge the extracted knowledge?

› The importance: in what perspectives?› The validness: in what context?› The usefulness : for what applications?

Should we apply operational criteria to judge the usefulness of extracted knowledge?

Thanks!!