Ben Said et al 2002 - Overview

Summary

This paper presents CUBES (CUstomer BEhavior Simulator), a multi-agent simulation platform for modeling consumer behavior in competing markets. The model is built on behavioral attitudes (mistrust, opportunism, conditioning, innovativeness, imitation) activated through behavioral primitives with threshold mechanisms. A genetic algorithm calibrates agent populations to match real market data. Key results show emergent market phenomena including brand lock-in and cyclic market share competition.

Overview

Citation: Ben Said, L., Bouron, T., & Drogoul, A. (2002). Agent-based Interaction Analysis of Consumer Behavior. Proceedings of AAMAS ‘02, Bologna, Italy, pp. 184-190.

Research question: Can a consumer behavioral model based on psychological primitives (imitation, conditioning, innovativeness) reproduce realistic market evolutions including emergent collective phenomena?

Approach: Multi-agent simulation using the Swarm engine with several thousand consumer agents interacting in a virtual market with competing brands. Genetic algorithms calibrate agent characteristics to match observed market data.

Two Key Originalities

1. Simultaneous individual and collective levels: Concepts (opinion leaders, innovation) are treated both as individual properties and as emergent collective phenomena
2. Generic behavioral functions: Consumer cognitive functions are derived from generic behavioral components related to interaction, not narrowly defined as reasoning and data processing

Main Content

Model Architecture

CUBES consists of:

• Consumer agents: Each with a socio-demographic profile (age, education, profession, social class) and behavioral attitudes
• Brand agents: Fixed number of competing brands with marketing strategies
• Virtual market: The environment where agents interact

The simulation flow follows: Brands emit stimuli (promotions, innovations, recommendations) → Consumers perceive stimuli through behavioral primitives → Attitudes and opinions update → Purchase decisions occur → Market shares evolve.

See CUBES Simulator Architecture for detailed system design.

Core Behavioral Framework

Consumer behavior is driven by five behavioral attitudes (BA):

1. Mistrust — skepticism toward stimuli
2. Opportunism — responsiveness to deals
3. Conditioning — habit formation from repeated exposure
4. Innovativeness — openness to new products
5. Imitation — tendency to follow others’ choices

These attitudes are activated through behavioral primitives (BP) — threshold-based response mechanisms that filter and weight external stimuli. See Behavioral Attitudes in CUBES and Behavioral Primitives and Thresholds for formal specification.

Social Processes

Two social processes drive attitude dynamics:

1. Imitation process — agents copy behaviors of connected agents
2. Conditioning process — repeated exposure to stimuli reinforces attitudes

These are distinct from direct information transfer — they operate through behavioral attitude modification. See Imitation and Conditioning Processes.

Calibration via Genetic Algorithm

Agent population characteristics are calibrated using GA to match real market data. The GA evolves chromosomes encoding behavioral and socio-economic characteristics, evaluated by a Result-Analysis Module that compares simulated diffusion curves and market shares to observed data. See Genetic Algorithm Calibration for ABM and GA Fitness Evaluation and the RAM.

Key Results

1. Behavioral attitude convergence: Young populations (15-25) show unstable oscillation in attitudes; older populations (45-65) converge and stabilize after ~15 steps
2. Brand lock-in: A dominant brand (70% initial share) maintains dominance beyond 90 time steps despite competitor efforts — an emergent phenomenon
3. Cyclic competition: With initially equal market shares, brands cyclically trade customers — reproducing real market share dynamics

Connections

• CUBES implements the ABM Methodology and Principles framework with a focus on psychological realism
• The behavioral attitude framework connects to Agent Decision Rules and Bounded Rationality — CUBES uses threshold-based rules
• Market share results connect to Market Share Equilibrium and Lock-In
• The GA calibration approach connects to Genetic Algorithm Calibration for ABM

See Also

• CUBES Simulator Architecture — system design details
• Behavioral Attitudes in CUBES — the five behavioral attitudes
• Behavioral Primitives and Thresholds — the BP activation mechanism
• Imitation and Conditioning Processes — the two social processes driving attitude dynamics
• Genetic Algorithm Calibration for ABM — population calibration
• GA Fitness Evaluation and the RAM — Result-Analysis Module used to score GA fitness against market data
• Market Share Equilibrium and Lock-In — emergent brand lock-in and cyclic competition results