Generating Offers with Cosine Similarity in Multi-Attribute Negotiation

Automated multi-attribute negotiation can be viewed as the search for a solution that satisfies two conflicting preferences. It is challenging not only because of the large multidimensional search space, but also when preferences of negotiating parties are kept private. Under such condition, negotiating agents must approximate their opponents’ preferences. To this end, we employ cosine similarity to generate offers approximately close to the opponent preference. Offers received from the opponent are regarded as its partial preference. Hence, the more similar an offer to the opponent’s most recent offer, the greater the chance that the offer will be accepted. To find an offer close to the opponent preference, the agent searches for the offer within its accepted utility range that has the highest cosine similarity to the opponent’s last offer. Our experiments show that generating offers with cosine similarity leads to high agreement rate and mutual gain for the negotiating agents. Keywords—Multi-attribute negotiation, agents, cosine