Users' review comments in shopping sites are useful for other users to decide whether or not buy the item. While users' comments or opinions are included in the reviews, descriptions about story contents are sometimes included in the reviews toward items with story like novels or movies. In some cases, these descriptions may spoil reader's or viewer's enjoyment and excitement. Hereinafter, we call these descriptions spoilers. Spoilers might be related to the position in the story line. In this study we use story documents that record all of the details of the given story. Using the story documents, we investigate the location to which the content of the spoilers correspond in the story documents. Based on the result of the investigation, we develop a method for detecting spoilers in users' review comments. We compared our proposed method with some baselines that detect spoilers using machine learning techniques with bag of words model. We found that our method performs as well as the baseline. This means that we can detect spoilers in the same level of precision even if we do not have labeled data on spoilers.
In cooperative maze exploration via multi-agent systems, deadlock resolution methods influence exploration efficiency. This paper evaluates the relation between communication protocols and exploration efficiency. This paper proposes three communication protocols for deadlock resolution on maze exploration via distributed collaborative control system considering communicable range of agents. The agents share different information depending on the communication protocols for deadlock resolution. The transmission route of the information is affected by the communication protocols for deadlock resolution. The first result is to show a relationship between communicable range and performance of deadlock detection and communication load as the characteristics of each proposed communication protocols. The second result is to derive the relation between flow quantity and density for some network structure. Using the two results we evaluate influence of communication protocols on exploration efficiency.
In recent ethological studies, the behaviors and interactions of animals have been recorded by digital video cameras and webcams, which provide high functionality at reasonable cost. However, extracting the behavioral data from these videos is a laborious and time-consuming manual task. We recently proposed a novel method for tracking unmarked multiple honeybees in a flat arena, and developed a prototype software named “K-Track”. The K-Track algorithm successfully resolved nearly 90% of cases involving overlapped or interacted insects, but failed when such events happened near an edge of a circular arena, which is commonly employed in experiments. In the present study, we improved our K-Track algorithm by comparing the interaction trajectories obtained from forward and backward playing of video episodes. If the tracking results differed between the forward and backward episodes, the trajectory with lower maximum moving distance per frame is chosen. Based on this concept, we developed a new software, “K-Track-kai”, and compared the performances of K-Track and K-Track-kai in honeybee tracking experiments. In the cases of 6 and 16 honeybees, K-Track-kai improved the tracking accuracy from 91.7% to 96.4% and from 94.4% to 96.7%, respectively.
Many localization methods have been proposed for an autonomous mobile robot and many studies focus on enhancement of capability of each means. Monte Carlo localization such as particle filter has good robustness and widely used for mobile robots but lacks precise estimation in a certain environment. ICP(Iterative Closest Points) matching has good performance about estimation precision when there are many good features such as many straight walls but lacks robustness if there are few distinguish feature points around the robot. In this paper we propose a fusion strategy using neural network which can be applied to many fields in a building such as an indoor public space. Simulation results for 3 characteristic environments show effectiveness of the proposed strategy.