In this paper, we propose two models to weight each term in the document for document retrieval. Our idea of the models come from traditional Term Frequencies (TFs) and Term Weights (TWs) proposed in 2013. TF is based on the number of term occurrences in a document and used as de facto standard. On the other hand, TW is based on variation of term co-occurrences in a document and outperforms to TF. Our proposed models give much weight to terms which cooccur with terms frequently occur. We show experimental results comparing to the conventional models using a very large text corpus.

In the text mining, it must be conducted a preprocessing in order to obtain a structured data representation from the raw text unstructured. Preprocessing of the raw data greatly affect the results and accuracy of the mining process. In this paper, we extract text feature amount by using the Deep Auto-Encoder which is a kind of Deep Learning. (It's also called ``Semantic Hashing'') The data used in this article is the actual event information that is delivered by ``びもーる''. It was found that the features extracted by Deep Auto-Encoder from the real event information articles is useful for classification of event information.

Recently, DTN(Dley/Disruption-Tolerant Networking) Routings that achieve reliable end-to-end communication in environment not to be able to connect to whole area such as the disaster circumstance are worked with keen interests.The method spreading mobility plans assumes the case that immobile base-station exists in this DTN environment, and it creates a delivery channel of messages based on the trajectory.The problem of this method is that it consumes excessive amounts of buffer because all of node store trajectory to the buffer by all contact.In this research, I suggest new method that reduces consumption amounts of buffer while keeping delivery average of this method.

This article reports a case study of application of multi-agent evacuation simulations for disasters. We developped a system to assist understanding importance of trainings and thinking evacuation in each area based on exhaustive simulation of all combinations of conditions. Our system is displayed as an exhibition in a museum under a collaboration with ``Nigechizu'' project, in which local residents join to develop a map for evacuation from disasters. We analyzed log data of the usage of the simulation system for future improvements.