As the Internet develops, the number of Internet of Things (IoT) devices increases. Simultaneously, the risk of IoT devices being infected with malware also increases. Thus, malware detection has become an important issue. Dynamic analysis logs are effective at detecting malware, but it takes time to collect a large amount of data because the malware must be executed at least once before the logs can be collected. Moreover, dynamic analysis logs are affected by external factors such as the execution environment. A malware detection method that uses a static property analysis log could solve these problems. In this study, deep learning (DL) was used as a machine learning method because DL is effective for large-scale data and can automatically extract features. Research has been conducted on malware detection using static properties of portable executable (PE) files, establishing that such detection is possible. However, research on malware detection using hash functions such as Fuzzy hash and peHash is lacking. Therefore, we investigated the characteristics of hash values in malware classification. Moreover, when the surface analysis log is viewed in chronological order, that the data are considered have concept drift characteristics. Therefore, we compared malware detection performance using data with the concept drift property. We found that the hash function could be used to prevent performance degradation even with concept drift data. In an experiment combining PE surface information and hash values, concept drift showed the highest performance for certain data.

Mirai is an active malware that targets and poses constant threats to IoT devices. IoT malware penetrates IoT devices illegally, makes them download other malware such as bots, and infects them. Therefore, to improve the security of IoT devices, it is important to analyze the behaviors of IoT malware and take countermeasures. In this study, to analyze the behaviors of IoT malware after entering IoT devices and propose new security functions for operating systems to prevent activities such as IoT malware infection, we analyze Telnet logs collected by a honeypot of IoT devices. Thereafter, we report the analysis results regarding IoT malware input commands. The results show that many commands related to shell execution, file download, changing file permissions, and file transfer, are often executed by IoT malware.

Mental health illness has become a serious public problem. Finding changes in everyday behavior is a demand. This paper tries to detect persons who have mental health illness using their short comments posted to social network systems. The novelty of this study is using comments in a system for communication between users with mental health illness, in order to prepare a sufficient amount of supervised data for machine learning. The authors used approximately 120,000 comments in the system as positive samples and 120,000 comments in Twitter as negative samples for detecting mental health illness. Both data are posted short comments on a daily basis. The authors conducted a straightforward classification of the comments using a support vector machine and surface-level features of the comments. The accuracy of the classification is 0.92 and the characteristic phrases used for the classification are related to troubles in mental health. The ability to classify everyday statements can be expected to lead to the early detection of mental disorders.

Researchers are expected to find previous literature that is related to their research and potentially has a scientific impact from among a large number of publications. This paper addresses the problem of predicting the citation count of each research paper, that is, the number of citations from other papers to that paper. Previous literature related to the problem claims that the textual data of papers do not deeply affect the prediction compared with data about the authors and venues of publication. In contrast, the authors of this paper detected the citation counts of papers using only the paper abstracts. Additionally, they investigated the effect of technical terms used in the abstracts on the detection. They classified abstracts of papers with high and low citation counts and applied the classification to the abstracts modified by hiding the technical terms used in them. The results of their experiments indicate that the high and low of citation counts of research papers can be detected using their abstracts, and the effective features used in the prediction are related to the trend of research topics.

Predicting author’s attributes is useful for understanding implicit meanings of documents. The target problem of this paper is predicting author’s native language for each document. The authors of this paper used surface-level features of documents for the problem and tried to clarify the practical tendencies of the writing style as word occurrences. They conducted a classification of the abstracts written in English of approximately 85,000 scholarly papers written in English or in Japanese. As a result of the experiment, the accuracy of the binary classification was 0.97, and they found that a number of distinctive phrases used in the classification were related to typical writing styles of Japanese.

Researchers are required to find previous literature which is related to their research and has a scientific impact efficiently from a large number of publications. The target problem of this paper is predicting the citation count of each scholarly paper, that is, the number of citations from other scholarly papers, as the scientific impact. The authors tried to detect the high and low of the citation count of scholarly papers using only their abstracts, especially, non-technical terms used in them. They conducted a classification of abstracts of scholarly papers with high and low citation counts, and applied the classification also to the abstracts modified by deleting technical terms from them. The results of their experiments indicate that the scientific impact of a scholarly paper can be detected from information which is written in its abstract and is not related to the trend of research topics. The classification accuracy for detecting scholarly papers with the top or bottom 1% citation counts was 0.93, and that using the abstracts without technical terms was 0.90.

A lot of information concerning the status of companies are available on the Web. However, a simple search of documents does not explain the meaning or the cause the status. Semantical interpretation and hypotheses generation are necessary for further analysis. This paper proposes a method to analyse the cause and the situation of bankruptcy with respect to particular condition that a user can specify as a query. The method is based on the theory of formal concept analysis. The novelty of the method is in (a) that sentences are considered as objects and words are considered as attributes and (b) that a concise subgraph of the concept lattice is introduced and used to guess the cause. Two cases of interactive and iterative process are shown where a user proceeds from a simple query to a new hypothesis, which would not be able to found by a naive cross tabulation or keyword extraction.

A simple document search is insufficient when we analyse corporate information. Not only a list of search results, but also a justification why the results match the query condition is important. This paper proposes a method to extract cause of bankruptcy from news articles applying the co-occurrence analysis of words.

It is known that a lot of incidents has happened ahead of a serious accident. Such experiences have been collected in medical sites as incident reports. The text mining is expected as a method that discovers the factors of incidents and the improvement of the situation. This paper proposes a method to analyse the co-occurrence relation of the words that appear in the medical incident reports using concept lattice.