Yoshiaki Yasumura

Negotiation strategy of agents in the MONOPOLY game

In this paper, we discuss negotiation strategy of agents that play the board game MONOPOLY. First, we developed MONOPOLY server for the purpose of making agents confront on network and a MONOPOLY agent which negotiates with the other agents. Its negotiation strategy is based on a game theory using an evaluation function. This function includes pattern of color groups, money and position of tokens. The agent compares the proposal candidates for trading by using this function and selects the best one. Finally, the experimental results show that this agent system can be benchmark for studying negotiation strategy.

A tool for animated agents in network-based negotiation

In this paper, we describe a tool for developing animated agents with facial expressions in negotiation through a computer network. The tool learns a users tendency to select facial expressions of the animated agent, and generates facial expressions instead of human. In order to estimate facial expressions, the tool has an emotional model constructed by Bayesian network. We can easily develop animated agents if we use this tool as a component. And we describe the estimation of an opponents emotional state, based on observed data, by using the Bayesian network.

Case based online training support system for ADR mediator

This paper describes an overview of an online training support system for ADR mediators. To educate good mediators, much training is necessary, which is not easy for supervisors. As supervisors have to take care of many students, they cannot spare much time for a specific student. To train students effectively, some support system is needed.This system provides an environment for online disputation. Using this system, the supervisor and students can participate in the mediation process even if they are outside of the University. Furthermore, this system stores many disputation records in the form of XML documents as a case base, and this case base is used to navigate the mediation process. During the disputation, users can retrieve old similar scenes of disputation, and they can construct proper arguments by referring to similar scenes. Furthermore, by comparing records of disputation or by analyzing them statistically, we can get the information that help to evaluate the mediation skill.

Acquisition of a concession strategy in multi-issue negotiation

This paper presents a method for acquiring a concession strategy of an agent in multi-issue negotiation. This method learns how to make a concession to an opponent for realizing win-win negotiation. To learn the concession strategy, we adopt reinforcement learning. First, an agent receives a proposal from an opponent. The agent recognizes a negotiation state using the difference between their proposals and the difference between their concessions. According to the state, the agent makes a proposal by reinforcement learning. A reward of the learning is a profit of an agreement and a punishment of negotiation breakdown. The experimental results showed that the agents could acquire the negotiation strategy that avoids negotiation breakdown and increases profits of an agreement. As a result, agents can acquire the action policy that strikes a balance between cooperation and competition.

Integration of Bagging and Boosting with a New Reweighting Technique

We propose a novel ensemble learning method, IBB (integration of boosting and bagging). This method creates initial classifiers by bagging, and then builds base classifiers by boosting using the previously created classifiers. IBB has two new techniques, a reweighting technique and data adaptation. The reweighting technique increases a weight of a sample which is misclassified by both the ensemble classifier and previously created base classifier. The data adaptation is realized by controlling the number of iteration in boosting. Experimental results using the datasets of UCI machine learning repository show that IBB resulted better accuracy than the other ensemble learning methods on several datasets and on average

Potential topics discovery from topic frequency transition with semi-supervised learning

This paper presents a method for potential topic discovery from blogsphere. A potential topic is defined as an unpopular phrase that has potential to spread through many blogs. To discover potential topics, this method learns from topic frequency transitions in blog articles. Though this learning requires sufficient amount of labeled data, labeled data is costly and time consuming. Therefore this method employs a semi-supervised learning to reduce labeling cost. First, this method extracts candidates of potential topics from categorized blog articles. To detect potential topics from the candidates, a classifier is built from topic frequency transition data. Experimental results with real world data show the effectiveness of the proposed method.

Semi-supervised Gaussian process regression and its feedback design

Semi-supervised learning has received considerable attention in the machine learning literature due to its potential in reducing the need for expensive labeled data. The majority of the proposed algorithms, however, have been applied to the classification task. In this paper we present a graph-based semi-supervised algorithm for solving regression problem. Our method incorporates an adjacent graph, which is built on labeled and unlabeled data, with the standard Gaussian process (GP) prior to infer the new training and predicting distribution for semi-supervised GP regression (GPr). Additionally, in semi-supervised regression, the prediction of unlabeled data could contain some valuable information. For example, it can be seen as labeled data paired with the unlabeled data, and under some metrics, they can help to construct more accurate model. Therefore, we also describe a feedback algorithm, which can choose the useful prediction of unlabeled data for feedback to re-train the model iteratively. Experimental results show that our work achieves comparable performance to standard GPr.

Strategy acquisition on multi-issue negotiation without estimating opponent's preference

In multi-issue negotiation, an opponents preference is rarely open. Under this environment, it is difficult to acquire a negotiation result that realizes win-win negotiation. In this paper, we present a novel method for realizing win-win negotiation although an opponents preference is not open. In this method, an agent learns how to make a concession to an opponent. To learn the concession strategy, we adopt reinforcement learning. In reinforcement learning, the agent recognizes a negotiation state to each issue in negotiation. According to the state, the agent makes a proposal to increase own profit. A reward of the learning is a profit of an agreement and punishment of negotiation breakdown. Experimental results showed that agents could acquire a negotiation strategy that avoids negotiation breakdown and increases profits of both sides. Finally, the agents can acquire the action policy that strikes a balance between cooperation and competition.

Giving advice in negotiation using physiological information

The authors propose a method for giving advice to participants of negotiation using physiological information. We apply Newcombs A-B-X model (Isamu Saito, 1987) as a human relation model to negotiation. Based on the model, we adopt the users preference of a proposal, an impression of an opponents attitude and a degree of the users emotional disturbance in the advice generating tool. We measure the users perspiration and pulse rate to extract the users emotional disturbance. The tool learns the relation between the users preferences and the advice. The experimental results show that the tool can improve the quality of advice.

Quick adaptation to changing concepts by sensitive detection

In mining data streams, one of the most challenging tasks is adapting to concept change, that is change over time of the underlying concept in the data. In this paper, we propose a novel ensemble framework for mining concept-changing data streams. This algorithm, called QACC (Quick Adaptation to Changing Concepts), realizes quick adaptation to changing concepts using an ensemble of classifiers. For quick adaptation, QACC sensitively detects concept changes in noisy streaming data. Empirical studies show that the QACC algorithm is efficient for various concept changes.