This study proposes and evaluates a novel Bayesian network classifier which can asymptotically estimate the true probability distribution of the class variable with the fewest class variable parameters among all structures for which the class variable has no parent. Moreover, to search for an optimal structure of the proposed classifier, we propose (1) a depth-first search based method and (2) an integer programming based method. The proposed methods are guaranteed to obtain the true probability distribution asymptotically while minimizing the number of class variable parameters. Comparative experiments using benchmark datasets demonstrate the effectiveness of the proposed method.

This study proposes and evaluates a new Bayesian network classifier (BNC) having an I-map structure with the fewest class variable parameters among all structures for which the class variable has no parent. Moreover, a new learning algorithm to learn our proposed model is presented. The proposed method is guaranteed to obtain the true classification probability asymptotically. Moreover, the method has lower computational costs than those of exact learning BNC using marginal likelihood. Comparison experiments have demonstrated the superior performance of the proposed method.

Automated essay scoring (AES) is the task of automatically assigning scores to essays as an alternative to grading by human raters.Conventional AES typically relies on handcrafted features, whereas recent studies have proposed AES models based on deep neural networks (DNNs) to obviate the need for feature engineering.Furthermore, hybrid methods that integrate handcrafted features in a DNN-AES model have been recently developed and have achieved state-of-the-art accuracy.One of the most popular hybrid methods is formulated as a DNN-AES model with an additional recurrent neural network (RNN) that processes a sequence of handcrafted sentencelevel features.However, this method has the following problems: 1) It cannot incorporate effective essay-level features developed in previous AES research.2) It greatly increases the numbers of model parameters and tuning parameters, increasing the difficulty of model training.3) It has an additional RNN to process sentence-level features, enabling extension to various DNN-AES models complex.To resolve these problems, we propose a new hybrid method that integrates handcrafted essay-level features into a DNN-AES model.Specifically, our method concatenates handcrafted essay-level features to a distributed essay representation vector, which is obtained from an intermediate layer of a DNN-AES model.Our method is a simple DNN-AES extension, but significantly improves scoring accuracy.

For academic writing, elaborating an argument particularly addressing an argument strength is important to establish causal relations between sentences. However, when an argument becomes large or complex, elaborating an argument considering the argument strength is difficult. To solve this problem, this article presents a proposal for an argument elaboration support system using a Bayesian network representation of the Toulmin model. Using that Bayesian network representation, the proposed system can estimate argument strength, sentence validity, and sentence influence. Moreover, it can generate optimal advice for revising the argument.

Observing various learning goals from peers allows learners to specify new objectives and sub-goals to improve their personal experience. Setting goals for learning enhances motivation and performance. However an unrelated goal might lead to poor outcome. Hence learners have divergent objectives for a same learning experience. Latent Dirichlet Allocation (LDA) is a model considering documents as a mixture of topics. This study then proposed a recommendation model based on LDA, able to determine distinct categories of goals within a single dataset. Results focused on a dataset of 10 learning subjects and over 16,000 goal-based Twitter messages. It showed (1) different goal categories and (2) the correlation between the LDA parameter for the number of topics and the type of subject. Evaluations of goal attributes also showed an increase of goal specificity, commitment and self-confidence after observing different types of goals from peers.

We propose an automatic construction method of multiple equivalent test forms indicated by test information function, and the method reduces the probabilities of selecting redundant items to the same test form. In previous studies, although their methods minimized the different between the test information functions of the constructed test forms, they neglected the content similarities of selected items in the same test form. Therefore, the content similar items have probabilities to be selected into the same test form. This affects the test reliability. The main idea of this paper is to reduce the probabilities by applying a latent Dirichlet allocation in the test construction method to detect the content similarities between the selected items and the remaining items in the item banks.

This paper relates to the effect of tablet PCs in e-learning. We performed analysis based on the “dual channel model”, which models the information processing capabilities of humans. More specifically, we provided paper media, keyboards, pen tablets, and tablet PCs as input devices used for annotations during e-learning, measured the gaze point of each learner by an eye-mark recorder, and evaluated each device by setting memory and comprehension tests, giving questionnaires, and evaluating the note-taking. As a result, we have shown that the use of tablet PCs in e-learning 1) enables concentration on the content, 2) reduces the extraneous cognitive load imposed by making annotations, 3) increases learners' comprehension and memory retention, and 4) enables efficient note-taking, thus increasing the accuracy of notes as learning aids.

This paper proposes a new computerized adaptive testing employing a decision tree model, instead of test theories. The attribute variable of the model is examinees' responses to each item and the output variable is examinees' test total scores. Some simulation experiments show better performances of the proposed method compared to the traditional methods and solve the problems.

In e-learning area, the content development method is one of the most important research topics. This paper proposes a presentation method of visual contents (a text, a still image) in synchronization with narration (sound content) and a pointer to make the efficiency of the resource allocation of cognitive memory capacity increase, and make the transmitted amount of information maximize, based on a human's cognitive-information-processing model "dual channel model." Under various contents presentation environment {(1)narration, (2) text (with narration / without narration), (3) still images, (4) still images + text (with narration / without narration), (5) video, and (6) video + text} in e-learning, we performed some control experiments (measure the point of fixation for e-learning students by an eye mark recorder, memorization and a contents understanding test and the questionnaire) with or without pointer. The results showed the validity of this model and the effectiveness of the proposed content development method in e-learning.

Item-response theory is applied to peer assessment by a new method for estimating assessment-criterion parameters. The proposed model is a modified graded item response model with an assessorpsilas evaluation-criterion parameters. The advantages of this method are as follows: (i) the capability of evaluating learnerspsila abilities on the same scale even if the assessors have different assessment criteria; (ii) evaluating the learnerspsila abilities by considering each assessorpsilas characteristics, thus giving reliable evaluated results; and (iii) easy estimation of the parameters from missing data. As a result of these advantages, the proposed method improves the accuracy of peer assessment. The proposed model was compared with conventional methods, and the proposed method showed the best performances.

We have developed a method for online detection of examinees' aberrant responses. This method uses response time data in e-testing. Unique features of this method are: 1. It includes an outlier detection method using Bayesian predictive distribution. 2. It can be used with small-sample sets. 3. It provides a unified statistical test method of various statistical tests by changing hyper-parameters and provides more accurate test results than commonly used methods. 4. Outlier statistics are estimated by considering both examinee abilities and the difficulty level of items. We evaluated this system, and results of our evaluation show that it is effective.

Maximum likelihood (ML) method for estimating parameters of Bayesian networks (BNs) is efficient and accurate for large samples. However, ML suffers from overfitting when the sample size is small. Bayesian methods, which are effective to avoid overfitting, have difficulties for determining optimal hyperparameters of prior distributions with good balance between theoretical and practical points of view when no prior knowledge is available. In this paper, we propose an alternative estimation method of the parameters on BNs. The method uses a principle, with roots in statistical thermal physics, of minimizing free energy. We propose an explicit model of the temperature, which should be properly estimated. We designate the model "data temperature". In assessments of classification accuracy, we show that our method yields higher accuracy than that of the Bayesian method with normally recommended hyperparameters. Moreover, our method exhibits robustness for the choice of introduced hyperparameters..

In the context of e-learning contents, if we take an approach based on the "Dual Channel" model, which is well known as a processing model for human perceptual judgments, cognitive resources can be used most effectively by methods that synchronize narrations with images and video contents. In general, however, the learner must search for a visual fixation point in the image or video while simultaneously listening to the narration, and if it is difficult to locate that fixation point, then the cognitive burden is increased, and the efficiency of understanding the learning content decreases. In cases such as these, it is considered effective to introduce a presentation method that uses a pointer to visualize the fixation point. In this research, we used an eye mark recorder (a device for measuring the subject's point of visual focus and pupil diameter) to measure the point of fixation for e-learning students, in order to measure the effects of leading the subject's fixation point with a pointer. The results indicated that in more than 80% of presentation material displays, the subjects followed the lead of the pointer. Tests following the experiment also showed that the subjects demonstrated higher percentages of correct answers for contents that displayed the pointer.

A SCORM-compliant learning management system (LMS) has been developed that enhances learning by effectively and efficiently managing the learning itself. First, a SCORM-LST was developed by adding to SCORM (sharable content object reference model) a framework that describes facilitation corresponding to the learner's state of learning and describes the learning state transitions. This makes it possible to describe collaborative learning, assessment, and facilitation for multiple users. Next, a SCORM-compliant LMS (SALMS: SCORM-compliant adaptive LMS) based on the SCORM-LST was developed. SALMS interprets SCORM-LST code and changes the user interface to match the learner's state of learning. It also manages the learning so that it matches the designer's intentions.

We utilized e-Learning using example data from special support education stored in a web database as e-Learning contents. The e-Learning was conducted with teachers from various educational institutes: from primary schools to universities, in addition to special support schools. Through a collective problem solving process we were able to collect in a single web database disparate knowledge from a variety of teachers.

E-learners typically check their understanding by taking end-of-unit quizzes, usually as often as they like. However, the benefits of doing this are not well understood. In this research, a "consistency index", which was defined for a series of answers from repeated attempts at quizzes, was used to classify learners into groups. The difference in the structure of the acquired knowledge for each group was clarified using Bayesian networks. As a result, learners who require additional individual counseling can be objectively detected by the index. Using networks that teachers thought to be ideal, adequate individual counseling for each learner can be provided.

This paper proposes an e-testing construction support system (eTCSS) with the prediction tools for the constructed test. This paper performs some comparison experiments to find the best predictive performances models for the predictive response-time distribution and the predictive response-time distribution. Furthermore, the amount of test information based on the item response theory, which is important to improve the measurement efficiency of the constructed test, is applied to be the prediction tool of the eTCSS. Finally, this paper evaluates the system by using actual data. The results show the effectiveness of the system.

This paper proposes an agent that acquires the domain knowledge concerned with the content from a learning history log database and automatically generates motivational messages. The unique features of this system are as follows: The agent builds a learner model automatically by applying the Bayesian network. The agent predicts a learner's final status (1.Failed, 2. Abandon, 3. Successful, 4.Excellent) using the learner model and his/her current learning history log data. 3. The agent compares a learner's learning processes with excellent learners' learning processes in the database, diagnoses the learner's learning processes and generates adaptive messages to the learner. The comparisons between the proposed method and the agent using the decision tree show that the proposed method has better prediction performances and effective to degrease the number of students withdrew from classes.

. This paper proposes a method of automatically analyzing the characteristics of e-learning contents, using e-learning time data stored in learning history databases. Although many studies exist on mathematical models of the response time, they have the disadvantage that the parameters are difficult to interpret and estimate. The response curve of elearning time data proposed in this paper has the unique feature of deriving its two-parameter model by employing the Entropy maximization method with certain restrictions so as to make it easier to interpret the parameters. The two parameters α and s in the model are interpreted as follows: α represents the complexity of the content (i.e., the number of simple cognitive processes required to understand or solve the content) and s represents the expected time of a simple cognitive process in the content. This means that the average learning time for a content is divided into the two parameters α and s, so that the average learning time for a content is equivalent to the product of α and s. Based on these parametric properties, this paper proposes a new content evaluation method using the α-s plane, or α-s chart. Incorporating this evaluation method, the authors have developed a LMS (Learning Management System), the effectiveness of which is demonstrated in practical situations. The results show that the system let a teacher grasp contents characteristic easily and is effective for contents improvement

An online visualization system of a discussion process in computer-supported collaborative learning (CSCL) has been developed to assist learners by enabling them to monitor the actual states of their discussion and allowing them to improve their learning community. The unique features of this system are 1) the learners have to select the most suitable category that represents his/her message content from the choices presented in the developed BBS and 2) the proposed system estimates the structure of the Markov from the stored categories' data sequences and visualizes the structure online. We demonstrate the effectiveness of this system using actual data and provide some evaluations. The results show that the proposed system motivates the learners and improves their methods of discussion

Ten years ago, technology standards in e-learning and learning support systems became the center of public attention. Various international standards have been proposed. However, there is no portfolio assessment standard that enables the support of assessment activities. Therefore, a portfolio assessment modeling language (PAML) for supporting a portfolio assessment standard was developed. It is very likely that the modeling language is not only able to describe needed portfolios in portfolio assessment but also facilitate assessment activities using a learning support system.

This paper proposes a LMS (learning management system) with intelligent agent to provide effective adaptive messages to a learner. The unique features of this paper are shown as follows: The agent system proposed in this paper has a learner model, which is automatically and continually constructed by applying the decision tree model constructed from the learning histories data stored in the data-base. The constructed leaner model predicts a learner's future final status (1. Failed, 2. Abandon, 3. Successful, 4.Excellent) using his/her current learning history data. The constructed leaner model becomes more exact as the amount of data accumulated in the database increases. The agent system presents the optimal instructional message based on the learner's predicted future state. The agent provides some attention cues according to Ueno (2004) at the timing when a learner begins to be bored with his/her learning. In addition, this paper demonstrates the effectiveness of this system through actual e-learning classes.

This paper examines accumulation methods for effective use of portfolios and develops a meta-language ("MelaPass") for describing the framework of portfolios for making a portfolio assessment. The purpose of MelaPass is to describe the structure of portfolios which have "soundness" and "exhaustiveness" features, and the language can be used in cooperation with existing metamodels (i.e., it has an "affinity" feature). The paper also develops a portfolio assessment support system called "passports" that is based on MelaPass. By using this system, users can design portfolio assessments based on MelaPass and use and manage portfolios.

This paper proposes a new contents analysis method for e-learning by using response time data. The unique features of this paper are as follows: 1) From information theoretic approach, the Gamma distribution is derived as a probability distribution of response time in the e-learning. 2) The two parameters in the Gamma distribution, /spl alpha/ and /spl beta/, are respectively interpreted as follows: The parameter /spl alpha/ means "Complexity of the content (which means the numbers of simple understanding processes to understand or solve the content )" and the parameter /spl beta/ means "Easiness of the simple understanding process in the content". As a contents analysis method, this paper proposes "/spl alpha/-/spl beta/ chart". 3) An online system by using the proposed method is introduced. Furthermore, this paper demonstrates some efficient points of the developed system.

Recently, “collaborative learning” using internet has become popular in educational technology societies. One of unique advantages of the collaborative learning using internet is that much amount of learning process data concerned with their discussions can be stored. However, there are few studies about how to analyze the data and how to efficiently utilize it. It is an urgent task to consider how to utilize efficiently this much data concerned with collaborative learning. This paper introduces some new Data mining technologies and text mining technologies for collaborative learning through a ILMS (Intelligent Learning Management System) “Samurai” which the author has developed.

Proposes a method of constructing student models for intelligent tutoring systems (ITSs) by using information criteria. This proposal provides a method to automatically construct the optimum student model from data. The main problem when traditional information criteria are employed to construct a model is that a large amount of data, which is difficult to obtain in actual school situations, needs to be obtained. This paper proposes a new criterion for using a smaller amount of data by utilizing a teacher's expert knowledge. Concretely, (1) the general predictive distribution is derived, and (2) a method of determining the hyper-parameters by using a teacher's expert knowledge is proposed. Finally, some Monte Carlo experiments comparing some information criteria [BIC (Bayesian information criterion), ABIC (Akaike's extension of BIC), MDL (minimum description length), and the exact predictive distribution] are performed. The results show that the proposed method provides the best performance.