Kanayo Ogura

A Method of Extracting Topic Threads Towards Facilitating Knowledge Creation in Chat Conversations

This paper proposes a robust algorithm for extracting topic threads in chat conversations using some language devices and discusses how chat conversations can contribute to knowledge creation using the history of chat conversations and our proposed algorithm. There are two promised applications: 1) an automatically generated knowledge-base and 2) a supporting system for chat conversations by editing the history of chat conversations.

The Development of a Positive Community using Virtual Space for Cancer Patients

This research investigated the effect of using a three-dimensional online chat environment on community development for cancer patients¿ mental health. In the support group, from two to five patients and one facilitator chatted in virtual space for a one and half hour session every week for three years using avatars with emotional expressions. The messages in the support group were analyzed by text mining tool. The results showed that the frequency of positive emotion words increased, on the other hand, the negative words decreased after one year. The percentage of participants¿ messages gradually increased with the sessions. The informational support in messages for improving their QOL (quality of life) increased during the three years. These data showed that the virtual support group promotes patients¿ positive emotions, QOL, and autonomy of the self-help group.

Analysis of Community Development using Chat Logs: A Virtual Support Group of Cancer Patients

In this paper, we report the results of a support group of cancer patients who use a 3-dimensional chat system. We examined the sequences of their messages and intervals with chat logs to evaluate the community development by conversational data. The users tended to send serious messages during long intervals, and the roles they played in the conversations changed as a sense of community emerged.

A Defense Method against Distributed Slow HTTP DoS Attack

Threat of Distributed Denial of Service (DDoS) attack, that attempts to make a machine or network resource unavailable is getting serious. A service provided for load testing was abused to perform a DDoS attack targeting online game company in September, 2014. Such services are bringing serious threat for most services on the Internet. Slow HTTP DoS attack is one of the DoS attack methods that targets HTTP servers. This method obstructs the service by saturating the connection pool with slow and many requests. It is known that Slow HTTP DoS attack by just one attacker can be prevented effectively by limiting the number of connections for each IP address. On the other hand, it is also known that it is difficult to defend from Slow HTTP DoS attack from multiple attackers (i.e. Distributed Slow HTTP DoS attack). The threat of DDoS attack is getting serious, so we need a effective defense method against Distributed Slow HTTP DoS attack. In this paper, we propose and evaluate a defense method against Distributed Slow HTTP DoS attack by disconnecting the attack connections selectively by focusing on the number of connections for each IP address and the duration time. We show in our research that our proposed defense method is effective against Distributed Slow HTTP DoS attack.

A Study of Packet Sampling Methods for Protecting Sensors Deployed on Darknet

A darknet monitoring system is developed to grasp malicious activities on the Internet in an early stage and to copewith them. The darknet monitoring system consists of network sensors deployed widely on the Internet. The sensors capture incoming unsolicited packets. A goal of this system analyzes captured malicious packets and provides effective informationfor protecting good Internet users from malicious activities. To provide effective and reliable information, sensors mustbe deployed in secret and hidden from outside. On the otherhand, attackers intend to detect sensors for evading them. This attempt is known as localization attacks to darknet monitoring systems. If actual location of sensors is revealed to attackers, itis almost impossible to grasp the latest tactics used by attackers. Thus in our previous work, we proposed a packet sampling method, which samples incoming packets based on an attribute of packets sender, to increase a tolerance to a localization attack and to keep a high quality of information publicized by the system. As a result, we almost succeeded to counter from a localization attack, which generates spike on the publicized graph to detect a sensor. However in some cases, proposed sampling method works to attackers advantage and spikes appear clearly on the graph. Therefore, we propose advancedsampling methods, which sample incoming packets based on multiple attributes of packets sender. In this paper, we presentour improved methods and show a promising evaluation result obtained from the simulation.

A Proposal of QoS-aware Power Saving Scheme for SDN-based Networks

The amount of Internet traffic is increasing rapidly day by day due to increase in Internet users and communication devices such as smart phones. In order to handle network traffic, network devices such as routers and switches are being added steadily causing significant power consumption by the network. Internet traffic at a particular place/country fluctuates in different time of a day and different day of a week. However, network devices are configured to operate fully whether they are idle or fully utilized. Many researchers have proposed schemes to reduce power consumption by network devices by putting their links or devices themselves to sleep when their utilization is less than a certain threshold. However, the schemes do not consider the types and QoS of traffic when putting the devices or links to sleep. It has been noted that lower the rate of a link, the lesser it consumes energy. In this paper, instead of putting devices or links to sleep, we reduce the rate of links in order to save energy as well as satisfy the QoS of the traffic. We distinguish two types of traffic Real-Time and non-Real-Time. We maintain the shortest path delivery (QoS) for Real-Time traffic and reroute non-Real-Time traffic if needed to reduce the rate of a link. The simulation results have shown that we can reduce the power consumption by network devices and at the same time maintain the QoS of the traffic.

Evaluating a Dynamic Internet Threat Monitoring Method for Preventing PN Code-Based Localization Attack

The Internet threat monitoring systems are developed to grasp malicious activities on the Internet. Those systems consist of a data center and sensors deployed on the Internet. Sensors capture malicious packets and report to the data center. The data center investigates the latest trend of attacks by analyzing those packets and the result is open to the public. To publish precise monitored results, sensors are deployed in secret and hidden from outside. On the other hand, attackers intend to detect sensors for evading them. This attack is known as localization attacks to Internet threat monitoring systems. Recent localization attacks adopting PN code is sophisticated and effective countermeasure is not developed yet. Therefore, we propose a dynamic Internet threat monitoring method. This method switches sensors whose monitored results that reflect to published results in a data center as a countermeasure for PN code-based localization attack. We evaluated our method from the aspect of tolerance to the attack by applying raw captured packets provided by nicter. Meanwhile, the existing systems always publish monitored results reported by whole sensors. Therefore, the information that our method provides would decrease compared to that of the existing systems. However, we show that the decrease of information is sufficiently small.