This paper describes a system for detecting inappropriate images in manga books intended for reading by boys and girls. However, since a page of a manga book has multiple frames, each of which contains images of different scenes, it is necessary to determine inappropriate images for each frame of the page. In this study, we developed an inappropriate image detection system that combines a CNN for automatically extracting frames on a page as polygons and a CNN for recognizing the presence or absence of inappropriate objects in each extracted frame, specializing in detecting exposed female breasts, the most frequently appearing inappropriate image in manga books. This system has improved the efficiency and quality of inappropriate image detection, which was previously possible only by human visual inspection.

Automated negotiations have been studied very widely. Usually, in the field of automated negotiations, researchers focus on bilateral negotiations, in which two agents negotiate for each other. One-to-one bilateral negotiations are more natural in the real world among companies compared with a market-style (one-to-many). In the real world, companies are doing multiple bilateral negotiations every day. However, there is less research on multiple bilateral negotiations. Some mathematical analysis was conducted in the case of 3 agents in the classic economic literature, and also there has been a proposed protocol called SAOP for multiple bilateral negotiations in the international competition, ANAC. In multiple bilateral negotiations, the results may differ greatly depending on the order in which the bilateral negotiations are conducted. However, in previous research on the automatic negotiation protocol between multiple agents, which has been conventionally used in international competitions, almost no research considering the negotiation order was found. Therefore, in this study, we investigated whether better negotiation results could be obtained by using a proposed protocol that dynamically determined the negotiation order. The better negotiation result described here means that the value of the social surplus, which is the sum of the acquired utility values of all the negotiation agents, is high and has the shortest possible time. The method proposed here is social surplus-based pre-negotiation (hereinafter referred to as“ pre-negotiation”), in which the agent who evaluates the most important issues in a multi-agent negotiation starts the negotiation first. In this study, we compared the values of social surplus, the number of agreements, and negotiation time for 3, and 6-agent negotiations against the proposed and conventional methods. The results showed that the protocol proposed in this study achieved the highest social surplus, especially when pre-negotiation was used, and the protocol was more effective as the number of agents participating in the negotiation increased.

Visualization of discussions using diagrams is effective in facilitating efficient debates. However, because actual debates involve many conflicts, the diagrams that represent them tend to be complex in structure. Therefore, we use AF and BAF, which have simple structure and semantics to extract arguments that may be included in conclusions of debates, to support real-time debate progression and to verify the logical structure of debates. For this purpose, we propose a “BAF diagram summarization method” that recognizes and reduces subgraphs that represent local arguments about related sub-issues from these diagrams. The final output of this method is a diagram in the form of “Reliability based Argumentation Framework (RAF)”, which consists less nodes and less links between them. RAF is an extensional model of AF in which each argument is assigned an “argument class” based on its feasibility, and semantics for RAF is defined by argument classes of arguments and attack relations between them. The RAF semantics has the role of preserving the semantic information in the BAF diagram summarization method. This is because by computing the semantics of the resulting RAF, it is possible to capture some of the important information in the semantics of the original BAF. Furthermore, we developed an on-line discussion support tool that implements this method, and visualize the results of user input.