Task and Product Selection .

Uploaded on:
Category: News / Events
04/01/2003. 2. Diagram. On Integrating CatalogsA Hierarchical Constraint Satisfaction Approach to Product Selection for Electronic Shopping SupportA Multiple Attribute Utility Theory Approach to Ranking and Selection . 04/01/2003. 3. On Integrating CatalogsRakesh Agrawal and Ramakrishnan SrikantIBM Almaden Research Center.
Slide 1

Venture and Product Selection by He Jiang Department of Management University of Utah April 1 st , 2003

Slide 2

Outline On Integrating Catalogs A Hierarchical Constraint Satisfaction Approach to Product Selection for Electronic Shopping Support A Multiple Attribute Utility Theory Approach to Ranking and Selection

Slide 3

On Integrating Catalogs Rakesh Agrawal and Ramakrishnan Srikant IBM Almaden Research Center

Slide 4

Summary Problem: coordinating archives from various sources into an ace inventory. Holes: Many information sources have their own particular orders; verifiable similitude data in these source lists might be overlooked. Approaches: Naïve Bayes order Contribution: arrangement exactness can be enhanced by join the certain comparability data display in these source classifications

Slide 5

Problem — Why Integration? B2C shops need to incorporate inventories from numerous merchants ( Amazon); B2B entries converged into one organization (Chipcenter & Questlink �� eChips); Information gateways arrange records into classifications (Google & Yahoo!). Corporate entrances Merge intra-organization and outside data into a uniform order

Slide 6

Problem Identification — Model Building Problem recognizable proof: arrangement issue. Ace inventory M with classes C1, C2, … , Cn; Source list N with classifications S1, S2, … , Sm; Merge archives in N into M.

Slide 7

Question How to Integrate?

Slide 8

Straightforward Approach: Completely overlook N\'s arrangement, put each of N\'s item into M\'s classification as per M\'s characterization run the show.

Slide 9

Enhanced Approach join the understood classification data exhibit in N into M.

Slide 10

Assumptions and Limitations M and N may are homogeneous and have huge cover; M and N utilize similar vocabularies (Larkey, 1999). List chains of command is straightened and is dealt with as an arrangement of categories(Good 1965 & Chakrabarti 1997) Different progressive system levels (if M>N, can help recognize classes that M doesn\'t have; if N>M, NBHC can be connected.

Slide 11

Related Works and Gaps Naïve-Bayes classifiers are precise and fast(Chakrabarti et al 1997, … ), so we pick Bayesian model; Folder frameworks, for example, email routing(Agrawal et al, 2000,… ), activity predicting(Maes, 1994 & Payne et al, 1997), inquiry sorting out utilizing content clustering(Sahami et al, 1998) and filings transferring(Dolin et al 1999); But none of this frameworks address the errand of consolidating pecking orders The Athena framework incorporates the office of revamping envelope chain of importance into another progression (Agrawal et al, 2000); But no data from the old chain of importance is utilized as a part of either building the model or directing the archives.

Slide 12

Straightforward Approach

Slide 13

Straightforward Approach — Continued

Slide 14

Enhanced Bayes Classification

Slide 15

Effect of Weight on Accuracy Weight can make contrast for a given M and N; Tune set strategy to choose a decent incentive for the weight. in which the archive will be accurately arranged or will never be effectively grouped The most noteworthy conceivable exactness achievable with the improved calculation is no more terrible than what can be accomplished with the essential calculation.

Slide 16

Experimental Results — Data Sets Used Synthetic list: getting source index N from M utilizing diverse distributions(e.g. Gaussian). Genuine Catalog: two certifiable inventories that have some normal archives; treat the main list less the regular records as M, the rest of the reports in the second inventory as N;

Slide 17

Experimental Results

Slide 18

Experimental Results

Slide 19

Experimental Results

Slide 20

Experimental Results — Catalog Size

Slide 21

Experimental Results — Catalog Size

Slide 22

Contributions and Future Research Directions Contributions: improving the standard Naive Bayes grouping by joining the classification data of the source lists; the most elevated exactness of the upgraded method can be no more awful than that can be accomplished by standard Naïve Bayes order. Future research: utilizing different classifiers, for example, SVM to joining the certain data of N requires additionally work

Slide 23

A Hierarchical Constraint Satisfaction Approach to Product Selection for Electronic Shopping Support Young U. Ryu IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and people Vol. 29, No. 6, November 1999

Slide 24

Summary Problem: proposing an item determination system for electronic shopping support; Approach: progressive requirement fulfillment (HCS) approach Gap: basic scientific classification hierarchy(STH) approach is imperfect in that the pursuit is led on a solitary non specific item order; HCS is more intense and adaptable than STH.

Slide 25

Simple scientific classification Hierarchy Approach

Slide 26

Question 1. How would we look for a without sugar decaffeinated cola? 2. On the off chance that there isn\'t a cola that fulfill every one of the prerequisites, i.e., cola, without sugar and decaffeinated. what\'s your suggestion?

Slide 27

Gaps Search is led on a solitary non specific item progressive system; There may exist an item that can\'t fulfill every one of the imperatives; An item might be assessed to be superior to another while there is no enormous contrasts between these two items.

Slide 28

Hierarchical Constraint Satisfaction Approach Constraint Satisfaction: a procedure deciding assignments of qualities to factors that are reliable with given imperative; Hierarchical Constraint Satisfaction: an augmentation of STH which minimizes the fulfillment blunders of progressively composed limitations in view of their significance; Value of HCS: can be connected to cases in which there isn\'t an answer that is predictable with given requirements because of clashing imperatives.

Slide 29

Concepts Introduced Constraint space change: change of a Boolean imperative to a number-crunching requirement; Tree area: is one whose components are organized as a tree; in this manner can be taken care of all the more adaptably; Indifference interim: beat a weakness of various leveled thinking when the contrast between two options is little;

Slide 30

Constraint Satisfaction Error Measures the level of fulfillment of a number-crunching compel c by the limitation fulfillment blunder work for Boolean requirement, change them into number-crunching imperatives; e.g.

Slide 31

Hierarchical thinking and apathy interim

Slide 32

Constraint Hierarchies

Slide 33

Example Shopping for wipes items utilizing progressive requirement fulfillment approach. Every item is depicted by the accompanying qualities: Cost: pennies per sheet Add-on materials: "preparing pop", "aloe vera", … ; Strength: measured by pressure(psi) that breaks a sheet; Dispenser sort: "take care of", "pop"; Added simulated aroma: unscented, common aloe scented, characteristic jasmine scented and substance fragrance scented; Product reason: "universally useful", "diaper change".

Slide 34

Example — Result

Slide 35

Contributions and Future Research Directions Contribution: the item seek component is seen as a fulfillment issue of progressively sorted out limitations over item qualities, along these lines it is more effective and adaptable than item choice in view of a solitary item scientific classification chain of command. Future research: Purchasing prerequisite detail or imperative pecking order elicitation; finish model usage of the HCS approach; real acquiring/deals exchange in view of discourse –act hypothesis, illocutionary rationale and between hierarchical movement coordination.

Slide 36

A Multiple Attribute Utility Theory Approach to positioning and Selection John Butler, Douglas J. Morrice and Peter W. Mullarkey Management Science, Vol. 47, No. 6, June 2001

Slide 37

Summary Problem: building up a positioning and choice strategy for making correlation of frameworks that have different execution measures; Approach: consolidating Multiple Attribute Utility Theory (MAUT) and factual positioning and determination (R&S) utilizing lack of concern zone; Gaps: costing methodology is imperfect in that exact cost information may not be accessible, and it might be hard to gauge execution utilizing costs.. Favorable circumstances: thorough; near business rehearse; more straightforward to execute; can gauge the quantity of recreations required; can evaluate the relative significance of criteria

Slide 38

Gaps Most of the R&S writing concentrated on systems that diminish the multivariate execution measures to a scalar exhibitions measure issue, yet these techniques may have a few detriments, e.g. precise cost information may not be accessible; it perhaps hard to precisely append a dollar incentive to immaterial factors; Current methods may require a confounded stride of assessing a covariance matrix(Gupta & Panchapakesan 1979); Previous work doesn\'t give a way to deal with gauge the quantity of recreations required to choose the best arrangements with an abnormal state of probability(Andijani 1998, Kim & Lin 1999). Past work does not have an exchange off instrument that permits the leader to consolidate unique execution measures.

Slide 39

Assumptions Decision creator\'s inclinations are precisely spoken to ( Clemen 1991, Keeney & Raiffa 1976); Performance measures that is changed over to "utils" can be changed over to important unit by picking an invertible utility capacity; There is a lack of interest zone for the chief on all the execution measures;

Slide 40

General Outline of the Procedure

Slide 41

Multilinear Utility Function

Slide 42

Multiplicative MAU Model

Slide 43

Additive MAU Model If shared utility added substance free, then Example for added substance autonomy:

Slide 44

Single Attribute Utility Function Used Methods for relegating weights: exchange off strategy; scientific hier

View more...