In this paper, we propose a novel rent estimation method of real estate properties for restaurants. Previously determining the rent was based on the tacit knowledge, that is, intuition and experience gained by experienced salespeople. However, this business custom has problems (e.g., no evidence to the determined rent). Therefore, the transference of the knowledge and experience from the experienced salespeople to fresh one, is not effective. We propose a new rent estimation model which solves the above problems, and construct a rent estimation system based on the model. In order to build a model, we focus on the overt information and potential information. The overt information can be classified as follows: static information and dynamic information. The static information and dynamic information can be indexed by the salespeople. However, some parts of the indexing include scoring data manually by the salespeople. On the other hand, potential information can not indexed by the sales people. In this thiesis, we build the rent estimation concept model based on above three factors. In addition, we discuss about the specific challenges in building the proposed system, that are (1) acquisition and indexing of the tacit knowledge and (2) construction of a rent estimation model. To tackle the challenge (1), we interviewed the experienced salespeople of ABC-tenpo Inc., and extracted some factors related to the rent estimation. Based on the result, we determined parameters (variables) representing static information, dynamic information and potential information. To tackle the challenge (2), we developed a rent estimation model by Random Forest algorithm of machine learning. Through experiments, we confirmed that the dynamic information was most effective in estimating basic rents. Also, we confirmed that the potential information was superior in adjusting rents. Finally, the case of using all factors achieved the highest accuracy that coefficient of determination was 0.738.

Selection of housing, one of the necessities of human life, has a great influence on life for a long time. However, since it requires a wide range of information gathering and consideration before decision, state-of-the-art recommendation algorithms such as collaborative filtering do not work well. In this presentation, after reviewing issues specific to the real estate field, I cited examples of "application of crowdsourcing to social media (Twitter timelines)" and "application of deep learning to property images" as an effort by our research group. Finally I discuss what kind of AI technology is applicable in the real estate field.

We investigate the conditions in which cooperation is dominant in social media using the models of evolutionary games of public goods games with diminishing marginal utilities and attempt to clarify the mechanisms of prosperity of social media and effect of diminishing marginal utilities. Cooperation dominant situation corresponds to that in which social media thrives, i.e., posting articles and taking reactions by comment lead to more benefits result than do nothing as free-riders. A number of studies examined what mechanism contributes to maintain cooperation dominant situation by using evolutionary game, but their models don't include the effect of diminishing marginal utility. Thus, we developed an abstract model of social media using utility function with diminishing marginal utility and we experimentally investigated the effect of diminishing marginal utilities in the meta-rewards and SNS-norms games. The result showed that the games with diminishing marginal utility could model real SNS activities more suitably than those without it.