SUMMARY In Web‐based social media (blogs, Internet forums, etc.), readers express their opinions by attaching reply comments to articles and understand others' perspectives by reading these comments. On Web forums, users discuss and argue with others via reply comments. Thus, a series of comments may provide new information, particularly about disputes among participants. Unfortunately, current systems for viewing reply comments are primitive; it is hard to follow disputes and their relationships by using such systems. We examine the characteristics of disputes by using reply comments and propose a novel system for automatically detecting disputes and their relationships among a large set of reply comments. This paper defines a dispute section as a series of comments where the dispute index is sufficiently high. The dispute index is determined using an exponential model with three control parameters. Next, the dispute relationship can be defined according to a pair of arguing commenters. The basic concept is to exploit the order of comment posting times and comment writer information when detecting dispute sections because the comments tend to appear in ordered pairs. The proposed system detects dispute sections and visualizes them according to dispute relationships among commenters, enabling users to appreciate the controversial structure of comments. To test the performance of our system, we constructed a dataset containing comment sequences associated with 40 articles collected from an Web forum. We constructed the set of dispute comments manually. We found that the detection accuracy of our system for disputes and their relationships were 83% and 78% on average. Copyright © 2012 John Wiley & Sons, Ltd.

In this paper, we propose a new paper ranking algorithm that gives a high rank to papers which is credited by other authoritative papers or published in premier conferences or journals. Also, the proposed algorithm solves a problem that recent papers are rated poorly due to few citations.

It is very important to detect plagiarized programs in the field of computer science education. Therefore, many tools and algorithms have been developed for this purpose. Generally, these tools are operated in two phases. In phase 1, a program plagiarism detecting tool generates an intermediate representation from a given program set. The intermediate representation should reflect the structural characterization of the program. Most tools use the parse tree or token sequence by intermediate representation. In phase 2, the program looks for plagiarized material and evaluates the similarity of two programs. It is helpful to announce the plagiarized metarials between two programs to the instructor. In this paper, we present the static tracing method in order to improve program plagiarism detection accuracy. The static tracing method statically executes a program at the syntax-level and then extracts predefined keywords according to the order of the executed functions. The result of experiment proves this method can detect plagiarism more effectively than the previously released plagiarism detecting method.

Program similarity checking is an important application of programming education fields. Local alignment is one of the typical algorithms for comparing two strings. However local alignment based comparison does not reflect the weights of program keywords. This paper introduces an adaptive local alignment which reflects the frequencies of keywords to the similarity matrix. We experimented this method using a set of programs submitted to more than 10 real programming contests. The experimental result shows that the adaptive local alignment is more robust than greedy-string-tiling adopted in JPlag.

The protocol FAST (FIX adapted for streaming) is superior than FIX in that it performs data compression according to the protocol template prescribed. Since the FAST protocol template captures the sequence characteristic of the transmitted data, it is written manually by human experts. This paper describes an algorithm for generating FAST protocol templates automatically by analyzing the patterns of message sequence. The experimental result shows that the proposed algorithm is slightly more efficient than the human expert in respects of the compression ratio.