a first approach to argument-based recommender systems based on defeasible logic programming

37
A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming Carlos I. Chesñevar Dept. of Computer Science Universidad de Lleida - Spain Ana G. Maguitman Computer Science Dept. Indiana University – USA Guillermo R. Simari Dept. of Computer Science and Eng. Universidad Nacional del Sur – Argentina

Upload: ivie

Post on 20-Jan-2016

61 views

Category:

Documents


0 download

DESCRIPTION

A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming. Outline. (1) Introduction and motivations. (2) Argumentation Framework DeLP. (3) Recommender Systems (RS). (4) Argument-Based RS. (5) An Argument-Based Search Engine. (6) Conclusions. Ongoing work. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

A First Approach to Argument-based Recommender Systems based on

Defeasible Logic Programming

Carlos I. ChesñevarDept. of Computer Science

Universidad de Lleida - Spain

Ana G. MaguitmanComputer Science Dept.

Indiana University – USA

Guillermo R. SimariDept. of Computer Science and Eng.

Universidad Nacional del Sur – Argentina

Page 2: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 3: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Recommender Systems address the problem of information overload by providing guidelines or hints.

The Problem: Information Overload

Page 4: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Limitations of Traditional Views

• Unable to perform qualitative inference on the recommendations.

• Unable to deal with the defeasible nature of user’s preferences.

• Unable to provide explanations: trustworthiness issues!

Page 5: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Our Proposal

• Integrate recommender system technologies with a defeasible argumentation framework.

• To enhance practical reasoning capabilities of current recommender systems

Page 6: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 7: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

flies(X) bird(X), broken_wing(X)

flies(X) bird(X)

bird(X) penguin(X)

bird(opus)

broken_wing(opus)

(,)

bird(opus)

flies(opus)

, flies(opus)

={ flies(X) bird(X) }

Extension of logic programming which allows to reason with tentative, defeasible information.

Argument ,L 1) L 2) P, P3) There is no such that

satisfies 1) and 2).

DeLP (1)

Page 8: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

bird(opus)

flies(opus)

bird(opus), broken_wing(opus)

flies(opus)

, flies(opus) , flies(opus)

Specificity is a syntax-based criterion used to define preference ( ) among arguments.

An argument , L defeats another argument , Q if

, L is in conflict with , Q

, Q is preferred over , L or is unrelated to , L

DeLP (2)

Page 9: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

U

UD

D

D

U

UD

L

An argument , L is warranted if the root of the associated tree is labelled as U.

In order to determine whether an argument , L is finally acceptable, a dialectical tree rooted in , L can be built.

Leaves are U-nodes.

Inner node U iff every children node is a D-node.

Inner node D iff at least one children node is a U-node.

DeLP (3)

Page 10: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

How DeLP works

DeLP Interpreter

Abstract Machine

?- flies(opus)

• YES, there exists a warranted argument , L )

• NO, there exists a warranted argument for , L

• UNDECIDED, none of the above cases hold.

Possible Answers to Query L

User Query Defeasible rules

Strict rules

Facts

DeLP Program P

Page 11: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 12: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Recommender Systems

Programs that create a model of the user’s preferences, or the user’s task, to help identify worthwhile items such as news, web pages, books, etc.

Page 13: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Goals for Recommender Systems

• Find what the user wants.

• Know what the user likes.

• Gain trustworthiness from the user.

Page 14: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Traditional Approaches

Collaborative Filtering Recommenders: Infer preferences of individual users based on behavior of multiple users.

Content-Based Recommenders: Infer preferences of individual users based on what the user liked in the past.

Hybrid Recommenders: Combine both.

Page 15: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Hybrid RS: outline

Page 16: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 17: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Argument-based RS

Model the users’ preference criteria in terms of a DeLP program built on top of a content-based search engine.

Users’ preference criteria are:

• Incomplete

• Potentially Inconsistent

Page 18: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Encoding Users’ Preferences

DeLP

Program

user: preferences and behavior of active user (facts, strict rules and defeasible rules)

pool: preferences and behavior from a pool of users (defeasible rules)

domain: domain background knowledge (facts, strict rules and defeasible rules)

Page 19: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Argument-Based RS Architecture

Page 20: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Prioritizing Recommendations

Recommendations can be prioritized according to their epistemic status:

• Sw warranted results

• Su undecided results

• Sd defeated results.

Page 21: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 22: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Argument-Based Search Engine

Page 23: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

A Case-Study: Solving Web Search Queries

Consider a journalist who wants to search for news articles about recent outbreaks of bird flu.

Outbreaks of bird flu

?

Page 24: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Querying a Conventional Search Engine

news regarding

bird flu

Too many results!

Page 25: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Applying Implicit KnowledgeArticles written by Bob Beak are reliable.Usually, if the journalist is trustworthy then the article is reliable.

Old articles are not reliable.If a journalist never faked a report then she is reliably.

Thailandian and Japanese newspapers usually offer a biased viewpoint on bird flu outbreaks.The “Japanese Times” is non biased.Chin Yao Lin faked a report.

Page 26: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

DeLP Program

rel(X) author(X,A), trust(A).

rel(X) author(X,A), trust(A),

outdated(X).

trust(A) not faked-news(A).

rel(X) address(X, Url), biased(Url).

biased(Url) thailandian(Url).

biased(Url) japanese(Url).

biased(Url) japanese(Url), domain(Url,D),

D =“jpt.jp...”.

Defeasible Rules

Page 27: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

DeLP Program

rel(X) author(X,bob-beak).

outdated(X) date(X,D), getdate(Today),

(TodayD)>100.

thailandian(X) [Computed elsewhere]

japanese(X) [Computed elsewhere]

domain(Url, D) [Computed elsewhere]

getdate(T) [Computed elsewhere]

faked-news(chin-yao-lin)

Strict Rules

Page 28: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Search Results Facts author(s1, chin-yao-lin)

address(s1, “jpt.jp/...”)date(s1, 20031003)

author(s2, jane-doe)address(s2, “jpt.jp/...”)date(s2, 20031003)

author(s3, jane-truth)address(s3, “jpt.jp”)date(s3, 20031003)

author(s4, bob-beak)address(s4, “mynews.com/...”)date(s4, 20031003)

Page 29: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Is this Article Relevant?

author(s1,chin-yao-lin)address(s1,“jpt.jp/...”)date(s1, 20031003)

rel(s1)

author(s1,chin-yao-lin) trust(chin-yao-lin)

not faked-news(chin-yao-lin)

rel(s1)

address(s1, “jpt.jp/...”) biased(“jpt.jp/...”)

japanese(“jpt.jp/...”)

biased(“jpt.jp/...”)

japanese(“jpt.jp/...”) domain(“jpt.jp/...”; “jpt.jp/...”) (“jpt.jp” = “jpt.jp”)

faked-news(chin-yao-lin)

Page 30: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Is this Article Relevant? (cntd)author(s1, chin-yao-lin)address(s1, “jpt.jp/...”)date(s1, 20031003) rel(s1)

address(s1, “jpt.jp/...”) biased(“jpt.jp/...”)

japanese(“jpt.jp/...”)biased(“jpt.jp/...”)

japanese(“jpt.jp/...”) domain(“jpt.jp/...”, “jpt.jp/...”) (“jpt.jp”=“jpt.jp”)

Undecided

Page 31: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Is this Article Relevant?

author(s2, jen-doe)address(s2, “news.co.uk/...”)date(s2, 20001003)

rel(s2)

author(s2, jen-doe) trust(jen-doe)

not faked-news(jen-doe)

author(s2,jen-doe) trust(jen-doe) outdated(s2)

not faked-news(jen-doe)

rel(s2)

Warranted!

Page 32: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Is this Article Relevant?

author(s3, jane-truth)address(s3, “jpt.jp”)date(s3, 20031003)

rel(s3)

author(s3, jane-truth) trust(jane-truth)

not faked_news(jane-truth)

rel(s3)

address(s3,“jpt.jp/...”) biased(“jpt.jp/...”)

japanese(“jpt.jp/...”)

biased(“jpt.jp/...”)

japanese(“jpt.jp/...”) domain(“jpt.jp/...”;“jpt.jp/...”) (“jpt.jp” =“jpt.jp”)

Warranted!

Page 33: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Is this Article Relevant?

author(s4, bob-beak)address(s4, “mynews.com/...”)date(s4, 20031003)

Warranted!

rel(s4)

Page 34: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Outline

(3) Recommender Systems (RS)

(2) Argumentation Framework DeLP

(4) Argument-Based RS

(5) An Argument-Based Search Engine

(6) Conclusions. Ongoing work.

(1) Introduction and motivations

Page 35: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Conclusions

• Information needs are complex:– Users’ preferences are frequently inconsistent and

incomplete.– Domain knowledge is inconsistent and incomplete.

• Traditional recommender systems are unable to perform qualitative inference on the recommendations.

• We have proposed a novel way of enhancing recommendation technologies through the use of qualitative analysis using argumentation.

Page 36: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Ongoing Work

• Implementation! DeLP is fully implemented since 1996, and as a programming language since 1999.

• Extraction of relevant features from Web search results to encode them as part of a DeLP program.

• Represent semi-structured text through logical formulas.

• Defeasible rule discovery.• Integration with specialized argument

assistance tools.

Page 37: A First Approach to Argument-based Recommender Systems based on Defeasible Logic Programming

Questions?