The purpose of this research is to detect low usability web pages from the behavior of users, such as browsing time, mouse movement and eye movement. We experimented to investigate the relation between the quantitative data viewing behavior of users and web usability evaluation by subjects. We analyzed the data to detect low usability web pages using discriminant analysis. Low usability web pages, 94.4% (17pages / 18pages = detectable pages / low usability pages) were detectable from the moving speed of gazing points and the amount of wheel rolling of a mouse. Moreover, this detection reduced the number of web pages which should be evaluated by half (46% = 89 pages / 192 pages = detected pages / all pages).

Presents an experiment in understanding how learners of the Java programming language comprehend its concepts, such as classes, inheritance, interfaces, etc., in lectures and exercises. The authors used an empirical technique to test conjectures about how we learn the programming language. Usually, observations about how we learn a programming language are treated anecdotally. In this experiment, learners received lectures and did an exercise. The comprehension states of the learners were measured by tests in three learning phases. The first phase was before the lecture. In this phase, the learners had no knowledge of the programming language. The second phase was after the lecture and before the exercise. Here, the learners acquired some basic knowledge. The third phase was after the exercise. In this phase, the learners put the acquired knowledge into practice. Factor analysis was used to obtain factors affecting the test result of each learning phase. Changes in comprehension states are explained as a result of tracing the factors between the learning phases.

Empirical software engineering can be viewed as a series of actions to obtain knowledge and a better understanding about some aspects of software development, given a set of problem statements in the form of issues, questions or hypotheses. Experience has made us aware of the criticality of integrating the various types of data that are collected and analyzed as well as the criticality of integrating the various types of activities that take place, such as experiment design and the experiment itself. This has led us to develop a Computer-Aided Empirical Software Engineering (CAESE) framework to support the empirical software engineering lifecycle. The paper first presents the CAESE framework that consists of three elements: (1) a process model for the "lifecycle" of empirical software engineering studies, including needs analysis, experiment design, actual experimentation, and analyzing and packaging results; (2) a model that helps empirical software engineers decide how to look at the "world" to be studied in a coherent manner; (3) an architecture, based on which CAESE environments can be built, consisting of tool sets for each phase of the process model, a process management mechanism, and the two types of integration mechanism that are vital for handling multiple types of data: data integration and control integration. Next, the paper describes the Ginger2 environment as an instantiation of our framework. It concludes with reports on case studies using Ginger2, which dealt with a variety of empirical data types including mouse and keystrokes, eye traces, 3D movement, skin resistance level, and videotaped data.

We describe an experiment investigating the distribution of failure intensity in software reliability growth models.We found that the assumption of conventional models that the failure intensity follows a gamma distribution is not always true.Our new software reliability model does not make this assumption; rather, the failure intensity is calculated from failure data.We show that our new model predicts more accurately the number of detected faults for our study project than the conventional models.

The paper proposes a new method for increasing the reliability of multiversion software systems. The software using software breeding is more reliable than one using N version programming. But software breeding is not suitable for real time application because program versions are executed several times for detecting faulty modules. In the proposed method, the detection of faulty modules is performed in the background when program versions fail and the software continues the execution in the foreground. When the detection of faulty modules is finished, the combination of module versions in program versions are changed. Ten simulations, each of which executed program versions 10/sup 6/ times, were performed to analyse the effectiveness of the new method. This resulted in the reduction of the number of failures to range from 33% to 76% with an average of 56%.

This paper compares Software Breeding, a new approach for improving the reliability of multiversion software, with Community Error Recovery by mathematical models of the reliability of the multiversion software by these. The following results were obtained: (1) the reliability of the Software Breeding increases when the number of modules of the multiversion software increases, (2) the multiversion software by Software Breeding is more reliable than by the Community Error Recovery unless the reliability of a program version is low and (3) the Software Breeding can improve the reliability of multiversion software more easily than the Community Error Recovery since the Software Breeding does not need any additional routine which must have very high reliability except for the driver module.