Abstract
This chapter documents the model development process and the resulting conceptual model for the simulation of automated negotiation, covering the steps outlined in Chapter 2. As mentioned in Chapter 3 many studies investigating automated negotiation make use of simulation techniques, however they do not document the model development process (e.g. validation procedures, etc.) though this is an important prerequisite for the transparency and the traceability of a study. We therefore follow the steps outlined in Chapter 2, making and justifying necessary decisions, together with a detailed description of the resulting model.1
Preview
Unable to display preview. Download preview PDF.
References
Considerations concerning experimentation are not discussed in this chapter but in Chapter 5, where the experimental design for the computer experiments, as well as the dependent and independent variables are determined.
The system configuration — the static structure of an automated negotiation system — actually only needs to be static for the time it takes to conduct the specific negotiation at hand. For this time, i.e. the duration of the focal negotiation, the interaction protocol and the software agents can be conceived as permanent entities of the system for automated negotiation. Clearly if the automated negotiation system is an open system software agents can freely join or leave the system for different negotiation instances and alternative interaction protocols can be chosen.
We showed in Chapter 3.1.1 that e.g. different priorities for the issues, shared interests concerning options for issues, concave partial utility functions, etc. are sufficient.
Furthermore the results of these experiments will act as a benchmark for the evaluation of automated negotiation systems (combinations of software agents and an interaction protocol) in Chapter 6, in which we also discuss these results of human negotiation experiments.
http://www.interneg.org — last accessed on 17.03.09
Therefore it is necessary to determine which preferences of the users to employ as input to the simulation. We decided to use the latest preferences indicated by the subjects in the experiments as this preference information probably is free of errors and therefore representing the actual preferences of the user best.
Consult the ‘research papers’ and ‘publications’ sections of the InterNeg web page at http:// interneg.concordia.ca/interneg/research/papers/ — last accessed on 17.03.09 — for a sample of these studies.
Note that a single party actually can only select its own agent autonomously and cannot determine the software agent choice of the other party. Furthermore the interaction protocol has to be chose jointly by the parties.
However, termination of the negotiation could be the final result of sending a reject message if the opponent does the same as discussed in the section on termination criteria.
Though this is a very strong restriction of the interaction between software agents imposed by the protocol, it might be useful in case that reaching an agreement is obligatory. On the other hand, if this is not the case the reached agreements might be inferior to those reached when exploitation can be circumvented by reject messages like in protocol 3, or negotiations can be broken off by quit messages to circumvent unfavorable outcomes like in protocol 2. So there are benefits and drawbacks of such a protocol and maybe trade-offs between outcome dimensions in negotiations — e.g. the probability of reaching an agreement versus the quality of agreements reached — which have to be evaluated.
Scholars argue that the game theoretic notion of’ strategy’ as a fixed plan of action for all possible situations during a game is more applicable to software than to human agents as their code must explicitly and unchangeably determine all actions to be performed in all possible situations [21, 170, 197].
This random choice of one out of the offers between which the user is indifferent, together with the random choice which agent receives the call for offer and therefore has to start the negotiation, constitute the only stochastic components of the otherwise deterministic negotiation procedure and causes the conceptual model to be a stochastic discrete event model as discussed in Section 4.3.
Clearly for learning agents it is a necessary and valuable information, how the negotiation ended — with a break off or with an agreement — which can be utilized together with information on the course of the focal negotiation for later negotiations, however, in their present form the software agents in this study do not learn for future negotiations but this remains an aspect to be investigated in subsequent studies.
Note that some of the agents do not follow strict concession making, but if possible propose offers of same utility level, like the subsequently discussed monotonic concession strategy, however if there are no such offers of same utility level these agents also reverse to concession making, so their general strategy allows to classify them as concession strategies.
Though the majority of scholars view this exchange of messages as the most important factors that change the state of the negotiation, there also exist few models that consider continuous concession and resistance forces or impatience of negotiators [9], which would call for a system dynamic rather than a discrete-event model.
In case of the multiple identical preference functions indicated in Table 4.8 the subjects often chose quite apparent distributions of partial utility values over the options of an issue, like linear or nearly linear distribution of the partial values over the options, or giving all issues the same weight and only the most preferred option the full score for the issue and all other options in that issue a score of zero.
The joint conference of the INFORMS section on Group Decision and Negotiation, the EURO Working Group on Decision and Negotiation Support, and the EURO Working Group on Decision Support Systems.
A χ2 goodness-of-fit test indicates that the observed proportions of starting the negotiations by the two parties do not differ significantly from the expected equal distribution; χ2 = 0.74, d f = 1, p = 0.39. Furthermore also for the results of the simulation runs no differences in individual utility — the only outcome measure at the individual level, while all other outcome measures are at the dyad level as discussed in Chapter 5 — were found between first mover and second mover in the simulations.
An actual negotiation protocol should, besides informing the agents which messages are possible, also control that the agents follow this advice, however, for simplicity we programmed the negotiation protocol function to just indicate the possible messages and the agents to strictly follow this instruction.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2010 Springer-Verlag/Wien
About this chapter
Cite this chapter
Filzmoser, M. (2010). Conceptual Model. In: Simulation of Automated Negotiation. Springer, Vienna. https://doi.org/10.1007/978-3-7091-0133-9_4
Download citation
DOI: https://doi.org/10.1007/978-3-7091-0133-9_4
Publisher Name: Springer, Vienna
Print ISBN: 978-3-7091-0132-2
Online ISBN: 978-3-7091-0133-9