A recommendation framework based on automated ranking for selecting negotiation agents. Application to a water market

Abstract

This thesis presents an approach which relies on automatic learning and data mining techniques in order to search the best group of items from a set, according to the behaviour observed in previous groups. The approach is applied to a framework of a water market system, which aims to develop negotiation processes, where trading tables are built in order to trade water rights from users. Our task will focus on predicting which agents will show the most appropriate behaviour when are invited to participate in a trading table, with the purpose of achieving the most bene cial agreement. This way, a model is developed and learns from past transactions occurred in the market. Then, when a new trading table is opened in order to trade a water right, the model predicts, taking into account the individual features of the trading table, the behaviour of all the agents that can be invited to join the negotiation, and thus, becoming potential buyers of the water right. Once the model has made the predictions for a trading table, the agents are ranked according to their probability (which has been assigned by the model) of becoming a buyer in that negotiation. Two di erent methods are proposed in the thesis for dealing with the ranked participants. Depending on the method used, from this ranking we can select the desired number of participants for making the group, or choose only the top user of the list and rebuild the model adding some aggregate information in order to throw a more detailed prediction.