Multi-agent sentiment analysis using abstraction-based methodology

In this paper we want to develop sentiment analysis as multi-agent system (MAS), thus provides it with abstraction and modularity as a powerful handling against its complexity. Instead of reinventing every possibility of combination to refine the result of sentiment analysis, a number of sentiment analysis agents are released into the system. These agents possess intelligence and will autonomously interact with each other and determine their appropriate behavior to achieve system´s goal, which is determining polarity of online opinions. In this paper we propose an alternative approach to develop a multi-agent system using role abstraction, which we call Input-Process-Output (I-P-O). Different from some established methodologies like Gaia, Tropos, or MaSE, I-P-O is considered simpler and faster. I-P-O in our concept is not a traditional information processing structure in which data flow sequentially. It encased roles, dividing it into three niches, each with its own behavior combination that determines how agent communicates and how it is constructed. In this paper we also create agent taxonomy, a hierarchical structure that helps categorizing agents into I-P-O niches of abstraction based on their behavior similarities. I-P-O abstraction based approach has been applied and successfully used to transform object-oriented sentiment analysis into multi-agent system.