Case-Based Recommendation

Presented byChul-Hwan Lee

Barry Smyth

AgendaAgenda

Introduction1.

Retrieval & Recommendation2.

Similarity & Diversity3.

Conversation Technique4.

Personalization Technique5.

Origins of Case-Based RecommendationOrigins of Case-Based Recommendation

Introduction1.

Case-based Reasoning(CBR)A model of reasoning that incorporates problem solving, understanding,and learning, and integrates all of them with memory processes.

<- Soft Computing Methodology, Cognitive Science

[CBR prototype samples] Cyrus, Mediator, Persuader, Chef, Julia, Casey, and Protos, CLAVIER

Case-based Recommendation: A form of content-based recommendation that is especially well suited to many product recommendation Domains.

Case-base -> Well-structured Concepts DescriptionContent-base -> Less Structured Textual Item Description

Origins of Case-Based RecommendationOrigins of Case-Based Recommendation

Introduction1.

CaseA contextualized piece of knowledge representing anexperience that teaches a lesson fundamental to achieving the goals of the system.

Case-based SystemOperates through a process of remembering one or a small set of concrete instances or cases and basing decisions on comparisons between the new and old situations. – medical diagnosis, legal interpretation

Related DisciplinesFuzzy Logic(FL), Neural Network(NN), Evolutionary Computing(EC), Probabilistic Reasoning(PR), Belief Networks, Chaos Theory, Parts of Learning Theory

Successful TasksPlanning, Design, Diagnosis, Configuration, Classification, Prediction

Successful DomainsManufacturing, Medical, Help-desks, Sales Support

Major Components of a CBR systemMajor Components of a CBR system

Introduction1.

CBR CycleCBR Cycle

Introduction1.

1. Retrieving similar previously experienced cases whose problem is judged to be similar

2. Reusing the cases by copying or integrating the solutions from the cases retrieved

3. Revising or adapting the solution(s) retrieved in an attempt to solve the new problem

4. Retaining the new solution once it has been confirmed or validated

CBR CycleCBR Cycle

Introduction1.

Major Tasks of CBRMajor Tasks of CBR

Introduction1.

Guidelines for the Use of CBRGuidelines for the Use of CBR

Introduction1.

1. Does the domain have an underlying model?

2. Are there exceptions and novel cases?

3. Do cases recur?

4. Is there significant benefit in adapting past solutions?

5. Are relevant previous cases obtainable?

Advantages of Using CBRAdvantages of Using CBR

Introduction1.

1. Reducing the knowledge acquisition task.

2. Avoiding repeating mistakes made in the past.

3. Providing flexibility in knowledge modeling.

4. Reasoning in domains that have not been fully understood, defined, or modeled.

5. Making predictions of the probable success of a proffered solution.

6. Learning over time.

7. Reasoning in a domain with a small body of knowledge.

8. Reasoning with incomplete or imprecise data and concepts.

9. Avoiding repeating all the steps that need to be taken to arrive at a solution.

10. Providing a means of explanation.

11. Extending to many different purposes.

12. Extending to a broad range of domains.

13. Reflecting human reasoning.

Case RepresentationCase Representation

Content-base Recommendation: unstructured or semi-structured manner, using

keyword-based content analysis techniques

Case-based Recommendation: structured representations, using attribute-value

representations techniques – fit into e-commerce domains

Retrieval & Recommendation2.

Numeric

Nominal

Similarity-based RetrievalSimilarity-based Retrieval

Similarity Assumption: most similar to the target problem, more sophisticated

similarity metrics that are based on an explicit mapping of case features and

the availability of specialised feature-level similarity knowledge.

Retrieval & Recommendation2.

t=target query, c=case, w=weight

Inverse relative difference (numerical)

For nominal data1) Simple exact match metric (1 or 0)2) Use domain knowledge (similarity tables or similarity trees by domain knowledge expert)

Single-Shot Recommendation Problem: can be achieved by personalized

recommendation through extended dialog with the user.

Similarity & Diversity3.

Similarity & Diversity3.

Bounded random selection & bounded greedy selectionBounded random selection & bounded greedy selection

Similarity & Diversity3.

Alternative Diversity-Preserving ApproachAlternative Diversity-Preserving Approach

Similarity Layers

A set of cases, ranked by their similarity to the target query are partitioned into

similarity layers, such that all cases in a given layer have the same similarity

value to the query.

Order-based Retrival

constructs an ordering relation from the query provided

by the user and applies this relation to the case-base of products returning the

k items at the top of the ordering.

Compromise-driven Approach

the most similar cases to the users query are often not representative of

compromises that the user may be prepared to accept.

Navigation by AskingNavigation by AskingEmploying Natural Language Processing(NLP), originated from

conversational case-based reasoning systems.

Conversation Technique4.

An example conversational dialog between a user (the inquirer) and the AdaptivePlace Advisor recommender system (the advisor) in which the user is tryingto decide on a restaurant for dinner

Navigation by ProposingNavigation by ProposingPreference-based feedback, ratings-based feedback, critiquing-based

feedback(dynamic compound critiques)

Conversation Technique4.

Entree recommends restaurants to users and solicits feedback in the formof feature critiques. The screenshot a recommendation for Planet Hollywood along witha variety of fixed critiques (less$$, nicer, cuisine etc.) over features such as the price,ambiance and cuisine of the restaurant.

Conversation Technique4.

A screenshot of the digital camera recommender system evaluated in which solicits feedback in the form of fixed unit critiques and a set of dynamicallygenerated compound critiques. The former are indicated on either side of theindividual camera features, while the latter are presented beneath the main productrecommendation as a set of 3 alternatives.

User ModelingUser Modeling

User’s personal preference

User’s learned preference

Weak personalization to Strong personalization

Long term user preference information

This can be achieved by questionnaires, user ratings or usage traces, asking the

user to weight a variety of areas of interest, normal online behavior patterns

Case-based Profiling

Profiling Feature Preferences

Personalization Technique5.

Conclusions & References

[1] Pal, S. & Shiu, S. Foundations of Soft Case-Based Reasoning, Wiley Series on Intelligent System, 2004.

[2] Smyth, B. & Cotter, P. A personalized Television Listing Service, Communications of the ACM, 2000, 107-111.

[3] Goker, M. & Thompson, C. Personalized Conversational Case-Based Recommendation, Journal of Artificial Intelligence Research, 2004, 1-36.

[4] Ricci, F., Cavada, D., Mirzadeh, N., & Venturini, A. Case-Based Travel Recommendations, 2005, Retrieved from http://ectrl.itc.it:8080/home/publications/2005/cbr-cabv3.pdf on Sep. 2005.

Towards to Hybrid Recommendation System, Be~~tter System!Explain the results of their reasoningJustify their recommendationsIn/With Product Space

Final Goal is to provide good system to the world!That’s why we are here!

Question or Comments?Question or Comments?

http://www.sis.pitt.edu/~chleechlee56@pitt.educhlee56@pitt.edu