We constructed a basic order-book model which is based on the Maslov model and includes some empirical results such as distributions of order volume. Although this basic model succussfuly reproduces power law distributions of price changes, the market price greatly oscillates and the Hurst exponent of this model is much smaller than that of real data. In order to resolve this problem, we revised the basic ordre-book model by adding the effect that the selection probability of order types (i.e. market order, limit order and cancel) depens on the bid-ask spread. Our revised model still reproduces power law distributions of price changes and the Hurst exponet is improved.

In this study, we calculated the maximum gross pro t in a retail store in case of using quantitative order and investigated the performance of some demand estimation methods and order methods, by using agent-based simulations. We developed a store and a customer model, which re ects the consumer reactions to stock-outs such as brand switches, based on some Point-of-Sales data. As a result, we obtained an order quantity to maximize gross pro t of a store. Moreover, we found that the demand prediction and order methods which use only past sales data showed as same performance as quantitative order at best.

During the 2011 East Japan Great Earthquake Disaster, some people used social media such as Twitter to get information important to their lives. Therefore, social media users pay attention to prevent wrong information from diffusing. In this paper, we propose a novel information diffussion model, the Agent-based Information Diffusion Model (AIDM). We have proposed information diffusion model which is based on SIR model until now. This model is represented by the stochastic state transition model for whether to propagate the information, and its transition probability is de ned as the same value for all agents. People's thinking or actions are not the same. To solve this problem, we adopted three elements in our model: A new internal state switching model, user diversity and multiplexing of information paths. Furthermore, we try reappearance of multi-burst type information diffusion, and evaluate our proposal model by comparing real data.

Golf is a sport played all over the world and many people practice to improve their score. There is an index called Strokes Gained Status that is used in the PGA to analyze golfer's data. This index uses massive detailed golf data and compares them with an individual golfer's data. One of the advantages of this index is that it is possible to find detailled points of improvement of a golfer. However, the problem with this index is that there is a need to collect large number of detailed golf data at specific golf course. In our research, we applied Q-learning to a golf simulation model to obtain expected score of any golfer at any golf course. We showed that using this method, it is possible to analyze golf data with the Strokes Gained Status with any skill or any course.

Many studies have been made on single social networks, however, actual networks are not single: it is multi-layer and contain more information rather than single networks. In this paper, we focus on the link prediction problem of multiplex networks, and clarify how feature vector matrix acts when predicting hidden links from the observed network.

This paper discusses architecture for artificial intelligence (AI) that can debate problems with humans. The proposed architecture is composed of (1) offline system which creates text annotations over large amount of text data in advance, and (2) online system which runs various algorithms asynchronously in real time. We built a prototype of debating AI system based on the architecture and evaluated it over 50 discussion topics extracted from a popular debate website. The AI system outputs three opinions per a discussion topic; each opinion is composed of 7 sentences. Thus, it outputs 150 opinions for 50 topics in total. We found that in 61 opinions out of 150 the argumentations are understandable. We plan to develop a feature of combinatorial optimization of algorithms onto the architecture as a future work.

This paper presents an issue selection method to collect sentences for construction of an argument in debate using AI. The sentences that can be used to construct an argument are collected from a database using the query words extracted from a given motion. One problem in collecting sentences is that the collected sentences contain sentences concerning the motion from various issues; for example, some sentence support/oppose to the motion from viewpoint of health, and an another from environment. This paper presents an issue selection method aiming to collect sentences consistently support/oppose to the motion from some fixed viewpoint.

We proposed a sentence ordering method which is designed for debate opinion gener- ation. The opinions in debate are required to be coherent and logically structured. Our proposed method employs a machine learning approach with two kinds of features which examine the co- herency and the structure of opinions. The rst feature measures similarities between sentences, and second feature represents roles of sentences. In order to measure sentence similarities exibly, we used Paragraph Vector (PV) which is an implementation of distributed representations. We con- rmed effectiveness of PV based ordering method by comparing with Bag-of-Words based method. The experimental results show that PV based method improves accuracy of major c