Value Charts Visualization and Interactive techniques to support

Value Charts: Visualization and Interactive techniques to support Multi-attribute Preferential Choice UBC Giuseppe Carenini, Postdoc Intelligent Decision Support David Poole, Professor, PI Challenge/Problem Visualization and interactive techniques enable users to effectively perceive relationships and manipulate datasets, by replacing more demanding cognitive operations with fewer and more efficient perceptual and motor operations. We are investigating visualization and interactive techniques the may considerably improve multi-attribute decision making. Our focus is on facilitating the specification of trade-offs among attributes and also on allowing the decision maker to perform an effective sensitivity analysis of her judgments. Support Preferential Choice • Frequent and basic decision task: selecting a preferred entity out of a set of alternatives by evaluating the alternatives with respect to their values on a set of attributes – House – Univ. course – Vacation destination – Electronic equipment • Difficult because… – – often no dominant alternative decision maker has to evaluate trade-offs among attributes Traditional Approach Decision Maker Preferences Expressed as: Value Charts: Visualization Content Selection and Organization Text Planner Text Plan Additive Multi-attribute Value Function (AMVF) • Decision Theory • Psychology (Consumer’s Behavior) OBJECTIVES AMVF application OBJECTIVES 0. 78+ 0. 4 COMPONENT VALUE FUNCTIONS Neighborhood Location House Value 0. 7 0. 6 Westend Park-Distance Amenities 0. 8 20 m 2 _ Deck-Size _ 0. 32 Location House Value 0. 7 Neighborhood 0. 4 0. 6 0. 5 km 0. 9+ 0. 64+ 0. 3 COMPONENT VALUE FUNCTIONS House-A 0. 6 + 36 m 2 0. 25 0. 2 Porch-Size 0. 6+ Park-Distance 0. 3 Amenities 0. 8 + _ Deck-Size 0. 2 Likes it Supporting and Opposing Evidence Does not like it Porch-Size 0. 4 Location Value of an entity is computed by applying the AMVF OBJECTIVES + 0. 6 + House Value + COMPONENT VALUE FUNCTIONS 0. 64 0. 3 0. 4 Location House Value 0. 64 0. 7 Neighborhood 0. 6 _ 0. 9 0. 3 Amenities 0. 8 _ _ 0. 32 0. 2 House-A 0. 32 0. 8 _ 0. 25 Porch-Size 0. 6 Park-Distance n 2 + _ Deck-Size 0. 25 20 m 2 _ 36 m 2 Porch-Size + 0. 6 0. 5 km 20 m 2 Deck-Size House-A 0. 5 km Westend Park-Distance + 0. 9+ Amenities 0. 78 Neighborhood 0. 6 0. 78 0. 7 + + + _ 36 m 2 Sentence Generator Likes it + _ Does not like it Measure of Importance 1 Based on already existing components 0. 4 • Apply large grammar of English (SURGE [Elhadad 93, Robin 94]) • Aggregation: combining multiple propositions in one single • Apply morphology Text Micro-planner sentence [Shaw 98] • Generate English sentences Location 0. 55 0. 7 + House Value + 0. 64 0. 3 • Scalar Adjectives (e. g. , nice, far, convenient) [Elhadad 93] Neighborhood 0. 24 0. 6 + 0. 2 0. 32 0. 8 _ 0. 2 + 0. 6 Park-Distance 0. 9+ + _ Deck-Size 0. 6 0. 25 _ Porch-Size 0. 12 Likes it Does not like it 0. 5 n 2 0. 5 km Amenities + _ 0 House-A 0. 54 _ • Pronominalization: deciding whether to use a pronoun to refer to an entity (centering [Grosz, Joshi and Weinstein 95]) • Discourse cues (e. g. , Although, because, in fact) [Knott 96; Di Eugenio, Moore and Paolucci 97] + 0. 78 + + _ Supporting Opposing + 0. 6 20 m 2 36 m 2 1 vo Supporting Opposing