Transcript 4.2: Customer Model

Customer Personalization
Gio Wiederhold
Presented as Part of EPFL DL talk
Spring 2000
Customer models
Customer is defined to be {a person
• arranging a vacation trip
6
one specific task}
• activity/interests  location town  days  hotel by grade 
flight / tour bus  public transport  rented car
• arranging a business trip
• location & date  hotel by corp. plan  flight 
taxi, limo, or rented car
• getting a computer for Joe Cheap
• search CPU by price  modem  display
• getting a computer for Peter Fast
• search CPU by speed  storage  display  network
 A customer model is Hierarchical  computable, unambiguous
 alternatives at each level ( evaluate, closure, commit, rollback )
Example: Result modes for ranking
Databases:
• Completeness
• All the answers
Customer:
• wants choices
Prolog
• Correctness
• The first answer
Optimization
• The best choice
• Assumes all
factors are known,
no human decision
also (but rarely invoked)
• explanation for trust
• provider background
Ranking
Qualitative Significant Differences:
in terms of the customer model
Plan 1. UA59 dep.Wash.Dulles 17:10, arr. LAX 19:49
Plan 2. AA75 dep.Wash.Dulles 18:00, arr. LAX 20:10
Plan 3. UA119 dep.Wash.Dulles 9:25, arr. LAX 12:00
Busy
Joe:
Speedy
Mike:
Greedy
Pete:
P1= P2, P3
P2, P1=P3
P1=P3, P2
Personal vs. Customer Model
Actual Person has multiple roles
 how to switch
 explicitly - awkward
 implicitly - hard to
perform fast
 keep past contexts
return to prior local state
Switching rate will differ
• work versus fun
• adequacy of models
Concept not yet proven
• experimentally
• in practice
* Combining the models
Identify articulations
• Match customer and resource terms
• semantic mismatches
• thesauri, matching rules
Match level of detail
• Match customer and resource values,
summarize numbers, result ranks
• completeness, unit mismatches, text
• indicate constraints in models
• textual abstraction
• input for visualization