Pairwise issue modeling for negotiation counteroffer prediction using neural networks

doi:10.1016/j.dss.2010.11.002

Decision Support Systems

Volume 50, Issue 2, January 2011, Pages 449-459

https://doi.org/10.1016/j.dss.2010.11.002 Get rights and content

Abstract

Electronic negotiation systems can incorporate computational models and algorithms in order to help negotiators achieve their objectives. An important opportunity in this respect is the development of a component, which can assess an expected reaction by a counterpart to a given trial offer before it is submitted. This work proposes a pairwise modeling approach that provides the possibility of developing flexible and generic models for counteroffer prediction when the negotiation cases are similar. The key feature is that each negotiated issue is predicted while paired with each of the other issues and the permutations of issue pairs across all negotiation offers are confounded together. This data fusion permits extractions of common relationships across all issues, resulting in a type of pattern fusion. Experiments with electronic negotiation data demonstrated that the model's predictive performance is equivalent to case-specific models while offering a high degree of flexibility and generality even when predicting to a new issue.

Introduction

Business negotiations are an important type of exchange mechanism. The competency in conducting negotiations critically affects long-term business relationships, profitability, and reputations of businesses. Due to the rise of e-business, electronic negotiations have gained heightened importance lately [29], [32]. Electronic negotiations systems, which are web-based successors of negotiation support systems [27], allow parties located in various parts of the world to seek mutually acceptable agreements by exchanging offers over the networks in a structured or unstructured fashion. The organic involvement of the digital medium in these exchanges provides new opportunities for employing support and automation tools, such as preference modeling and software agents, for promoting effective decision-making.

The purpose of this paper is to investigate the feasibility of developing a generalized approach for empirically modeling an opponent's future offers. Consequently, the main contribution of this work is a pairwise modeling approach that is flexible with respect to the set of issues in a negotiation case and even to new unseen issues. In others words, the model has inputs and outputs that are independent of the particular issues of a specific negotiation case.

The approach is tested using data obtained from electronic negotiation experiments, which provide a rich source of information about the relationships between negotiators, their individual actions, and the negotiation dynamics. Advanced negotiation support tools equipped with adaptive capabilities to learn from past negotiations and assist in selecting appropriate negotiation tactics, can effectively utilize this information to help human negotiators in composing offers [2]. This work focuses entirely on the development and testing of the flexible predictive model, rather than its implementation as part of some electronic negotiation system. The model could be potentially incorporated into the analytical toolbox of any given electronic negotiation system, or it could be used by software agents for negotiation automation or for providing guidance to a human user.

As the development of the internet allowed for the natural involvement of computational models and methods in the electronic negotiation process, the question of adequately splitting of human vs. computer tasks had emerged. In addition to providing passive analytical facilities, computational models have been proposed for active negotiation support, as well as negotiation automation. The use of intelligent support tools for facilitating effective and efficient negotiations has been widely investigated in the past. Automated negotiations performed by software agents have been considered by many [3], [10], [28], [36]. Early work in this respect applied genetic algorithms to generate rules which relate the negotiators' current offers with the likely subsequent offers [37]. The algorithm evolves multiple classifiers by assigning a higher fitness to rules that more frequently contribute towards “compromise trajectories”. Kasbah is an agent-based marketplace in which various agents created by the users engage in bilateral negotiations on behalf of their principals (buyers and sellers) [11]. These agents, follow one of the three negotiation strategies defined by a price-concession curve over time [10]. More recently, a semantic web-based agent community was used to introduce the concept of pervasive negotiation support [33]. The system, called “SmartGuide”, included user, supplier, and negotiation agents to provide a flexible environment for context-aware automated negotiations.

The aforementioned overview of intelligent techniques used in negotiations is not comprehensive, but it provides insights into research on negotiation automation. In most business negotiation contexts however, humans need to be involved in the process with the intelligent software possibly playing a supporting role. Research in this area focuses on solutions for assisting human negotiators. An overview of electronic negotiations, negotiation support systems and negotiation software agents (NSA) includes discussion of the Aspire system, in which an agent uses an inference engine to provide recommendations based on inputs and previously encoded rules [30]. Another application of an agent assistant in commerce negotiations has been implemented in the eAgora marketplace [12], [13], where an agent watches over the shoulder of the negotiator and critiques trial offers. The agent also advises an action upon receiving the counterpart's offer and generates a package of adequate candidate offers for the consideration by the user. Regardless of the extent of computer involvement (automation or active/passive support) in composing offers, it would be useful to be able to predict potential reaction by a counterpart to a given trial offer. In a “toolbox” mode, a human user could perform “what-if” analysis before committing to an offer. If software agents are involved, the predictive model could help in the search for the most promising offer.

The rest of the paper is organized as follows. First, an overview of negotiation modeling and the proposed pairwise approach are presented, followed by the hypotheses. Subsequently, the negotiation case, which is the source of data for testing the hypotheses, is described. Using this negotiation case as the example, the pairwise modeling approach is explored in detail. Next, the experimental setup is presented followed by the results and relevant discussions.

Section snippets

Background

This paper builds upon the past work that proposed a model developed with neural networks for predicting opponents' counteroffers in the context of electronic negotiations [7], [8]. Although the results were encouraging, a major disadvantage was that a separate neural network had to be trained for each specific negotiation case. This was necessary because the model included inputs and outputs for all negotiation issues of the case. The pairwise modeling approach proposed in this paper

Hypotheses

Evaluating the performance of the proposed model requires reference benchmark models. The first benchmark proposed is the naïve model, where the next counteroffer for an issue is predicted as being the same as the most recent offer from the opponent on the specified issue. Although very simple, the naïve model is also flexible with respect to any set of negotiation issues, thus as a minimum requirement any proposed predictive model must outperform the naïve model. Therefore, the first

Development of the predictive model

The feasibility of the pairwise modeling approach is investigated using past data collected by the Inspire system [31]. The Inspire negotiation system is a web-based system, which permits two parties located anywhere in the world to negotiate on a chosen case. Negotiation issues and issue options are specified in advance for a specific case and each negotiator specifies a rating for each issue, and additionally indicates a rating for each option of an issue. All issue ratings as well as all

Prediction process

The starting point for the prediction process is a negotiation session that has multiple offers exchanged between a buyer and a seller, where each offer is for a set of negotiation issues. Either an offer will result in a counteroffer or the negotiation will end with an agreement or no agreement. The offer values for each issue are scaled to match a common distribution for all issues. An offer is then expanded into an observation for each pairwise permutation, which includes the relevant issue

Results and discussion

Although the Levenberg–Marquardt with Bayesian Regularization algorithm for training neural networks can reduce overfitting, it is still useful to try different levels of network sizes. If the neural network is too small, it will not be able to learn all the patterns. On the other hand, if the neural network is too large, it will be slow to train and will tend to overfit data as compared to a smaller network. Within the range of 1 to 30 neurons (Fig. 4), five neurons provided the lowest

Conclusions

The results of this research indicate that formulating a flexible pairwise counteroffer prediction model is technically feasible. The pairwise counteroffer model prediction error is shown not to be significantly higher than that of the equivalent model which considers all issues simultaneously. Furthermore, the error does not worsen for the prediction of new issues, which were excluded during the model training. It has also been demonstrated that the pairwise neural network based model

References (47)

R. Carbonneau et al.
Predicting opponent's moves in electronic negotiations using neural networks
Expert Systems with Applications
(2008)
S.P.M. Choi et al.
A genetic agent-based negotiation system
Computer Networks
(2001)
P. Faratin et al.
Using similarity criteria to make issue trade-offs in automated negotiations
Artificial Intelligence
(2002)
K. Hornik
Approximation capabilities of multilayer feedforward networks
Neural Networks
(1991)
G.E. Kersten et al.
WWW-based negotiation support: design, implementation, and use
Decision Support Systems
(1999)
O. Kwon et al.
Context-aware multi-agent approach to pervasive negotiation support systems
Expert Systems with Applications
(2006)
W.-P. Lee
Towards agent-based decision making in the electronic marketplace: interactive recommendation and automated negotiation
Expert Systems with Applications
(2004)
G. Ou et al.
Multi-class pattern classification using neural networks
Pattern Recognition
(2007)
D. Zeng et al.
Bayesian learning in negotiation
International Journal of Human-Computer Studies
(1998)
M.A. Abidi et al.
Data Fusion in Robotics and Machine Intelligence
(1992)

C. Bartolini et al.

A software framework for automated negotiation

C. Beam et al.

Automated negotiations: a survey of the state of the art

Wirtschaftsinformatik

(1997)

R.A. Bradley et al.

Rank analysis of incomplete block designs: I. The method of paired comparisons

Biometrika

(1952)

L. Breiman

Random forests

Machine Learning

(2001)

J. Brzostowski et al.

Adaptive negotiation with on-line prediction of opponent behaviour in agent-based negotiations

International Conference on Intelligent Agent Technology

(2006)

R. Carbonneau

Machine learning algorithms for negotiation counter-offer modeling

Proceedings of the Group Decision and Negotiation (GDN)

(2007)

R. Carbonneau et al.

Forecasting supply chain demand using machine learning algorithms

A. Chavez et al.

A real-life experiment in creating an agent marketplace

A. Chavez et al.

Kasbah: an agent marketplace for buying and selling goods

First International Conference on the Practical Application of Intelligent Agents and Multi-agent Technology

(1996)

E. Chen et al.

An E-marketplace for agent-supported commerce negotiations

E. Chen et al.

Agent-supported negotiations in the e-marketplace

International Journal of Electronic Business

(2005)

R.J.P. de Figueiredo

Implications and applications of Kolmogorov's superposition theorem

IEEE Transactions on Automatic Control

(1980)

H. Demuth et al.

Neural Network Toolbox 6 User's Guide

(2008)

Cited by (29)

Modeling and prediction of human negotiation behavior in human-computer negotiation
2021, Electronic Commerce Research and Applications
Citation Excerpt :
However, offer prediction involves uncertainty and is not an easy task, especially when concerning human behavior. That is the reason why most offer prediction studies are applied to and tested in CtC negotiation context (Brzostowski and Kowalczyk, 2006; Cao et al., 2015; Carbonneau et al., 2011; Ji et al., 2015; Lee and Ou-Yang, 2009; Ren et al., 2012), leaving the HtC negotiation relatively under-researched. More importantly, predicting the opponent’s future offer is not commonly done by the human negotiator.
With increasing e-commerce activities, the human–computer negotiation mechanism for online transactions emerges. This study proposes a segmented linear regression strategy (SLR) model, which can predict the opponent's future offer patterns and dynamically adjust the model parameters according to the opponent's behavior. We use the hope-based (behavior-oriented) and game-based (technical-oriented) negotiation theories as the theory ground to explicate the logic of the agent's negotiation strategy. Upon which we propose a theoretical model of bilateral negotiation as the framework for designing the SLR model. We then propose the testable hypotheses to evaluate the utility of the proposed model against the benchmark results. We build a prototype system and empirically test the hypotheses using computer-to-computer (CtC) and human-to-computer (HtC) negotiation experiments. The empirical results show that the proposed system outperforms the benchmark system in both CtC and HtC negotiations regarding deal price and utility. The proposed model provides an automated negotiation solution for online platforms that can improve the efficiency and effectiveness of their negotiation capability.
Bilateral negotiations for electricity market by adaptive agent-tracking strategy
2020, Electric Power Systems Research
Citation Excerpt :
As another example, once opponents have been grouped according to their previous negotiation traces by a clustering method, a Bayesian learning algorithm can identify an opponent's preferences from its current negotiation actions [31]. If opponent dynamically switches among a number of strategies during negotiation then it is challenging but possible to learn its bidding strategy by time series forecasting using Neural Networks [32]. In this paper, multi-strategy negotiations are facilitated as negotiating agents switch to a behavior-dependent strategy if they experience adamant opponent.
Bilateral transactions hedge both sides against uncertain price and volume risks of day-ahead auction and make up major portions of trading in electricity markets. Peer-to-peer bilateral transactions avoid broker fees but involve challenges of balancing between cooperative and competitive strategies for multi-round negotiations. To solve these challenges, this paper develops novel utility-based and adaptive agent-tracking strategies for bilateral negotiations. Relying on bilateral transaction volume and utility curves determined over a price range during unilateral pre-negotiation, utility-based strategies are developed for generation company (GenCo) agent (load serving entity (LSE) agent) to offer (bid) volumes and prices during multi-round bilateral negotiations. GenCo agent is also equipped with a new adaptive agent-tracking strategy that estimates reservation price of each LSE agent by Bayesian learning and updates the estimates in each round. The adaptive agent-tracking strategy facilitates cooperative yet competitive responses. Integration of new bilateral negotiation strategies with existing day-ahead auction in a renowned agent-based platform also enables combined simulation of the two market types. The case study demonstrates that the adaptive agent-tracking strategy empowers GenCoagents to swing bilateral negotiation results in their favor and yield 7% more payoff than the utility-based strategy, while achieving 100% improvement in frequency of failure of negotiation.
Real-time opponent learning in automated negotiation using recursive Bayesian filtering
2019, Expert Systems with Applications
Automated negotiation is a toolset to model human interactions during a negotiation process with the aim of improving the efficiency and quality of decision-making using advanced information analytics. During the negotiation, the participants share their viewpoints and concerns about the negotiation issues. However, in reality, they usually do not reveal the details of their preferences to one another. Therefore, modeling and learning opponents’ behavior is a crucial component of automated negotiation. In this paper, we propose an estimation technique based on recursive Bayesian filtering to facilitate opponent-modeling and -learning in the context of multi-participant, multi-issue negotiations. In the proposed technique, opponents’ preference profiles are modeled using fuzzy functions, which are very close to the way humans evaluate alternatives. As the negotiation progresses, the agents can recursively learn the parameters of these models in real time. The only required information for this learning process includes the feedback and the arguments the participants may provide in support of their decisions. At each round, a probabilistic graphical model is also implemented that utilizes the learned preference limits of the participants to offer a new proposal with a high probability of satisfying the participants and reaching an agreement. The proposed methodology is examined in two different negotiation contexts: energy-system development and real estate service. The experiments show that the proposed opponent modeling/learning approach increases the efficiency of the negotiation up to 85% and facilitates reaching an agreement in fewer rounds of negotiation without requiring any prior understanding of the negotiation participants.
Unanimously acceptable agreements for negotiation teams in unpredictable domains
2014, Electronic Commerce Research and Applications
A negotiation team is a set of agents with common and possibly also conflicting preferences that forms one of the parties of a negotiation. A negotiation team is involved in two decision making processes simultaneously, a negotiation with the opponents, and an intra-team process to decide on the moves to make in the negotiation. This article focuses on negotiation team decision making for circumstances that require unanimity of team decisions. Existing agent-based approaches only guarantee unanimity in teams negotiating in domains exclusively composed of predictable and compatible issues. This article presents a model for negotiation teams that guarantees unanimous team decisions in domains consisting of predictable and compatible, and alsounpredictable issues. Moreover, the article explores the influence of using opponent, and team member models in the proposing strategies that team members use. Experimental results show that the team benefits if team members employ Bayesian learning to model their teammates’ preferences.
A utility concession curve data fitting model for quantitative analysis of negotiation styles
2014, Expert Systems with Applications
Citation Excerpt :
In Faratin, Sierra, and Jennings (2002) an approach that looked to generate offers that would be similar to those of the opponent was proposed and illustrated using simulations. In Carbonneau, Kersten, and Vahidov (2008), Carbonneau, Kersten, and Vahidov (2011) neural network based models have been developed using experimental data for predicting opponent’s next move. The modeling approach developed in this paper may also be helpful in determining opponent’s negotiation mode (e.g. competitive vs. collaborative).
A reciprocal function is proposed for defining the utility concession curve of a negotiation participant. The curve has only one free parameter and can fit the complete range of negotiation styles from extremely competitive to extremely collaborative. Various equations are derived, including the definition of a utility concession curve center which permits intuitive quantifying of a utility concession curve. Subsequently, an optimization model is proposed to fit the curve to a set of offers. Using the proposed model, a set of negotiations is mined for utility concession curves which are then used for clustering and hypothesis testing. Three negotiations styles seem to emerge from the data; slightly collaborative, neutral and quite competitive. It is also shown quantitatively that the level of competitiveness of the counterpart is negatively correlated with the agreement rate, and this is validated against the experimental treatment. Additionally, by the use of an experimental treatment, it is shown that the level of competitiveness of the counterpart has a positive causal impact on the negotiator’s style, causing him to become more competitive or collaborative. The data fitting model can also be used for incrementally fitting the curve in real-time during a negotiation to provide an estimate of the negotiation style which may help in the negotiation process.
A single issue negotiation model for agents bargaining in dynamic electronic markets
2014, Decision Support Systems
Electronic Commerce has been a significant commercial phenomenon in recent years, and brings more benefits to people by comparison with the traditional market in aspects of cost, convenience and efficiency. The use of agent technology in e-markets for automatic bargains between buyers and sellers further increases the advantages of the e-market. However, most of existing agent-based bargain strategies assume a fixed number of negotiation participants, which may fail to enlarge agents' profits or to lead a bargain to a success when these strategies are applied in the e-market-based agent negotiations straightway. Problems such as unexpected changes on negotiation participants, possible changes on agents' expected negotiation outcomes, and unexpected switching in-between the buyer's and seller's markets need to be considered in order to guarantee agents' benefits and the success of negotiations. This paper proposes a novel agent negotiation model to help agents to perform a more effective bargain in e-markets by considering the objectiveness of the e-markets and the subjectiveness of the agents. The e-market situation by considering the number of bargain participants is proposed to reflect the objectiveness of the e-markets, and the agents' negotiation attitudes is introduced to indicate agents' responses to possible changes of the e-markets. Both the objectiveness of e-markets and the subjectiveness of agents are taken into account in negotiation procedures such as offer evaluation, negotiation decision making and counter-offer generation. Experimental results on a simulated e-market illustrate the benefits and efficiency of the proposed negotiation model in handling agents bargain problem in complex and dynamic e-market environments.

View all citing articles on Scopus

Réal A. Carbonneau is a PhD candidate at HEC Montréal, Canada and has spent 11 years implementing enterprise systems. His main research interests include investigating novel approaches to using machine learning in solving various problems characterized by partial and noisy information. Some of his more specific research interests are genetic algorithms, neural networks, support vector machines, random forests and software agents as well as enterprise systems and data warehouses as information sources and transactional target points. Réal has published papers in the International Journal of Intelligent Information Technologies, European Journal of Operational Research and Expert Systems with Applications as well as a book chapter.

Gregory E. Kersten is a Professor of MIS at the Department of Decision Sciences and MIS, John Molson School of Business, Concordia University (Montreal, Canada) and holds the Senior Concordia University Research Chair in Decision and Negotiation Systems. He is also the founder and first director of the InterNeg Research Centre (http://interneg.concordia.ca). He received his PhD in economic sciences and operation research; MSc in econometrics, Warsaw School of Economics. His research interests are conflict resolution in physical and virtual environments; individual and group decision making; Internet decision and negotiation systems; and culture and technology. He is the author and co-author of 3 books, 25 chapters in books, 82 articles in refereed journal publications, and 79 papers in refereed conference proceedings.

Rustam M. Vahidov is an Associate Professor of MIS at the Department of Decision Sciences and MIS, John Molson School of Business, Concordia University (Montreal, Canada). He received his PhD and MBA from Georgia State University. Dr. Vahidov has published papers in a number of academic journals, including Journal of MIS, Decision Support Systems, Information and Management, European Journal of Operational research, IEEE Transactions on Systems, Man and Cybernetics, Fuzzy Sets and Systems, and several others. His primary research interests include: decision support systems, supply chain management, e-commerce systems, distributed artificial intelligence and multi-agent systems, negotiation software agents, and soft computing.

View full text

Pairwise issue modeling for negotiation counteroffer prediction using neural networks

Abstract

Introduction

Section snippets

Background

Hypotheses

Development of the predictive model

Prediction process

Results and discussion

Conclusions

Expert Systems with Applications

Computer Networks

Artificial Intelligence

Neural Networks

Decision Support Systems

Expert Systems with Applications

Expert Systems with Applications

Pattern Recognition

International Journal of Human-Computer Studies

Data Fusion in Robotics and Machine Intelligence

A software framework for automated negotiation

Automated negotiations: a survey of the state of the art

Wirtschaftsinformatik

Rank analysis of incomplete block designs: I. The method of paired comparisons

Biometrika

Random forests

Machine Learning

Adaptive negotiation with on-line prediction of opponent behaviour in agent-based negotiations

International Conference on Intelligent Agent Technology

Machine learning algorithms for negotiation counter-offer modeling

Proceedings of the Group Decision and Negotiation (GDN)

Forecasting supply chain demand using machine learning algorithms

A real-life experiment in creating an agent marketplace

Kasbah: an agent marketplace for buying and selling goods

First International Conference on the Practical Application of Intelligent Agents and Multi-agent Technology

An E-marketplace for agent-supported commerce negotiations

Agent-supported negotiations in the e-marketplace

International Journal of Electronic Business

Implications and applications of Kolmogorov's superposition theorem

IEEE Transactions on Automatic Control

Neural Network Toolbox 6 User's Guide