Conducting the disambiguation dialogues between software agent sellers and human buyers

In the buying and selling interaction in e-commerce, one of the important dialogues is the discourse of resolving the ambiguities. That is to say that both selling and buying agents may have to conduct disambiguating dialogues to some extent in order to resolve the ambiguities and infer the true intents of the other agents. In the paper, we assume buyers are human agents while sellers are software agents and thus the seller agents will construct dialogues to resolve the ambiguities from the buyer agents. To resolve ambiguities, agents rely on four levels of domain knowledge: the world model, the mental model, the language model, and the rational model. In addition, four kinds of disambiguation strategies for the seller agent are implemented: (1) Guessing (2) Filtering (3) Recommending and (4) Asking more hints. Experiments are conducted also to measure the performance of the dialogue system against different parameter settings of the disambiguation strategies. We find that by optimal parameter setting and suitable strategy combination, the seller will result in a shorter dialogue without sacrificing much the optimal profit.