Abstract

Negotiation on the Internet is a new business activity that emerged with the development of the Internet and the World Wide Web. In order to study the use of software tools in cross-cultural Internet negotiations, a project named InterNeg was initiated. The INSPIRE negotiation support system of this project collected data of Internet negotiations by allowing participants to negotiate a mock case. Empirical research was conducted on these data by applying different data mining methods. This is because there were few former studies and hypotheses, and the variable number is large and they are not obviously correlated. Three data mining methods were applied to find hidden behavior patterns of Internet negotiations: Tree Rule Induction (TRI), Artificial Neural Networks (ANN) and Logistic Regression Analysis (IRA). The results showed that the numbers of offers sent, especially during the early and middle negotiation stages, are positively related to reaching agreements, while sending offers at last minute has low chance to get the compromise. Other factors, such as gender and interval time between offers, can also affect Internet negotiation results. Comparisons of results from different data mining, especially on their prediction accuracies, were also conducted. The results revealed that the TRI method enjoys the highest prediction accuracy while consuming least processing time. The ANN method has the lowest prediction accuracy. Our research results also indicated that the layer number and hidden unit number in the layers could not affect the ANN method's prediction accuracy.