Modeling and prediction of human negotiation behavior in human-computer negotiation

Highlights

• This study proposes a segmented linear regression Strategy (SLR) model, which can predict the opponent's future offer patterns.

• Segmented linear regression method can model complex concession behavior dynamically.

• A theoretical framework is built upon the hope-based and game-based negotiation theories.

• The model is validated empirically through computer-to-computer and human-to-computer negotiation experiments based on a prototype system.

Abstract

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.

Introduction

Ecommerce has been gaining popularity as more and more consumers shift away from in-store shopping to online shopping due to its convenience and price competitiveness (Soh et al., 2006). This trend has created many commercial opportunities for online merchants (Chua et al., 2007). Many e-commerce platforms provide diversified transaction mechanisms to improve sales and customer experience, from traditional fixed-price transactions to auctions and negotiations (Huang et al., 2013). Best Offer listings on eBay differ from standard auction listings and fixed-price listings, allowing for haggling between sellers and potential buyers¹. A study shows that Best Offer practice increases purchasing likelihood and results in higher sales than the fixed-price format.² However, the online negotiation process usually needs frequent communications between sellers and buyers, which requires communication tools supported by e-commerce platforms. Using Taobao, the largest C2C platform in China, as an example, the platform provides Ali Wangwang, which is an instant messenger tool developed by the Alibaba Group to support buyer–seller communication. A study by (Cai et al., 2018) surveyed 270 respondents to understand Ali Wangwang's role in customer's online purchase decision-making. The result shows that 78.62% of the respondents favor using Ali Wangwang to communicate with sellers before buying in Taobao, of which 39.17% were satisfied with the “price bargaining opportunity.” However, response speed and communication efficiency were the two most complained factors, indicating the need to develop intelligent instant messaging tools and sales chatbots to support online automated negotiation in e-commerce.³

The current online negotiation is mainly conducted by humans via instant messaging tools, while human negotiation is time-consuming and labor-intensive (Lai et al., 2010, Maréchal and Thöni, 2019). Thus, to improve the outcome and efficiency of negotiation, researchers have attempted to develop automated negotiation systems in the last two decades (Kersten and Lai, 2007, Kolomvatsos et al., 2016). Such systems include a software agent that is designed to negotiate on behalf of a human negotiator partially, via human-to-computer (HtC) negotiation, or entirely, via computer-to-computer (CtC) negotiation (Adomavicius et al., 2009, Hess et al., 2000, Ketter et al., 2012, Vahidov et al., 2014, Yang et al., 2012). The potential benefits of automated negotiation systems include reduced negotiation time and costs and less social confrontation (Lopes et al., 2008). It can also increase the likelihood of reaching a better deal by finding an optimal solution in the entire solution space, which is usually challenging to navigate for a human negotiator. The amount of calculation involved often extends beyond a person's cognitive capabilities (Luo et al., 2012). Thus, there is a demand for automated systems that can negotiate on behalf of the vendors with the consumers regarding price and service (Lin and Kraus, 2010, Vahidov et al., 2017, Yang et al., 2012).

Designing the negotiation strategy is a core issue in automated negotiation research (Jennings et al., 2001). A strategy defines how an agent makes concessions in response to the opponent’s offers (Kersten et al., 2013). During the past two decades, there have been many studies dedicating to this area. Heuristic-based strategy design has been widely used for designing counteroffer-making functions based on human’s daily negotiation experience (Cao et al., 2020, Cao et al., 2015, Faratin et al., 1998). To optimize the counteroffer and dominate the negotiation process, some machine learning methods, such as non-linear regression (NLR) and artificial neural network (ANN), are used to predict the opponent’s future offers (Carbonneau et al., 2011, Lee and Ou-Yang, 2009, Ren and Zhang, 2007, Ren et al., 2012). Both NLR-based and ANN-based prediction studies dedicate to improve the accuracy of prediction to enhance the negotiating capacity of the system. However, these prediction models have been mainly applied to CtC negotiations rather than HtC negotiations due to the difficulty in predicting human negotiation behaviors.

Understanding the behavior of the human negotiator is essential for the design of a negotiation agent that can recognize and react to the human negotiation strategy. One critical issue of HtC negotiation is how the agent (machine) develops a specific negotiation behavior. Therefore, machine behavior should be studied empirically from the angle of an engineering artifact and from the perspective of the inner mechanisms that generate the behaviors (Rahwan et al., 2019). For the managers to feel comfortable using the system, the agent’s negotiation behaviors should not be a black box algorithm. However, existing strategy designs in automated negotiation lack theoretical support from either natural or social sciences to explain the machine behavior of the negotiation systems.

In this study, we focus on explaining the strategic logic and the prediction rationale of the negotiation agent for both CtC and HtC negotiations. The contribution of our study is twofold. From the methodological perspective, taking advantage of both computational statistics and heuristic methods, we design a segmented linear regression (SLR) model that can model the opponent’s future offer pattern and deal with humans’ dynamically changing negotiation behaviors, either in good-faith or bad-faith manners. Unlike the existing NRL and ANN models, the SLR model does not depend on any predefined pattern functions and prior knowledge. From the theoretical perspective, our study is among the first, in both CtC and HtC automated negotiation literature, to combine the game-theoretic and behavioral negotiation theories in the proposed framework to explicate the strategic logic and prediction rationale of an automated negotiation agent. The proposed framework can provide theoretical guidance for future negotiation systems development.

The remainder of this paper is organized as follows. We first review the literature regarding the negotiation strategy design methods and prediction models, which lead to the SLR model we propose in this paper. We then present the theory underpinning the SLR model. In the artifact description section, we detail the search process of the design and the algorithm. In the evaluation section, we implement a prototype of the system and evaluate the core design features compared to benchmark systems using CtC and HtC negotiation experiments. In the theoretical and practical implication section, we discuss the theoretical and technical contributions from the perspectives of the automated negotiation theory and the e-commerce practice. Finally, the paper concludes with the limitations and possible future research directions.