Jianwen Sun

Sentiment recognition of online course reviews using multi-swarm optimization-based selected features

Sentiment recognition of online course reviews is valuable to understand emotions and feelings of learners. Nowadays, an increasing number of course reviews are being generated with the emergence of Massive Open Online Courses (MOOCs), which offers teachers a chance to analyze the opinions of learners and improve teaching strategies. However, the unstructured data contain large amounts of redundant features, which will significantly impact the performance of machine learning. To select effective emotional features, we adopt a multi-swarm particle swarm optimization (MSPSO) method, which generates multi diverse particle swarms on several cross training subsets. These swarms are utilized to find the best features by the F-Measure fitness function. The experimental results on the real-life dataset show that MSPSO can effectively reduce redundancy of text features and capture discriminative features. Compared with conventional feature selection methods, MSPSO can gain the better performance when selecting the same dimensions. Besides, the result of a user survey indicates that 72.19% of subjects approve of the usability of the recognition results and effectiveness of the feature selection. MSPSO is effective to pick discriminative features in course reviews.Discriminability of features is used to iteratively optimize particle swarms.MSPSO exploits the differences among sample subsets to form diverse swarms.

Variable Length Character N-Gram Approach for Online Writeprint Identification

The Internet’s numerous benefits have always beencoupled with shortcomings due to the abuses of onlineanonymity. Writeprint identification is a technique to identifyindividuals based on textual identity cues people leave behindonline messages. Character n-gram is one of the most effectiveapproaches to identify writeprint according to previousresearch. In this study, we propose a variable length charactern-gram based writeprint identification framework to addressthe identity tracing problem, integrating a genetic algorithm(GA) based feature selection component to solve the definitionproblem of n. To examine the approach, experiments areconducted on a test bed encompassing hundreds of reviewsposted by 20 Amazon customers. The experimental resultsshow the proposed approach is effective, obtaining aconsiderable improvement in identification accuracy and aheavy reduction of feature dimensionality.

Applying Stylometric Analysis Techniques to Counter Anonymity in Cyberspace

Due to the ubiquitous nature and anonymity abuses in cyberspace, it’s difficult to make criminal identity tracing in cybercrime investigation. Writeprint identification offers a valuable tool to counter anonymity by applying stylometric analysis technique to help identify individuals based on textual traces. In this study, a framework for online writeprint identification is proposed. Variable length character n-gram is used to represent the author’s writing style. The technique of IG seeded GA based feature selection for Ensemble (IGAE) is also developed to build an identification model based on individual author level features. Several specific components for dealing with the individual feature set are integrated to improve the performance. The proposed feature and technique are evaluated on a real world data set encompassing reviews posted by 50 Amazon customers. The experimental results show the effectiveness of the proposed framework, with accuracy over 94% for 20 authors and over 80% for 50 ones. Compared with the baseline technique (Support Vector Machine), a higher performance is achieved by using IGAE, resulting in a 2% and 8% improvement over SVM for 20 and 50 authors respectively. Moreover, it has been shown that IGAE is more scalable in terms of the number of authors, than author group level based methods.

Unfolding online learning behavioral patterns and their temporal changes of college students in SPOCs

ABSTRACT The emergence of massive open online courses not only changes the ecology of higher education, but also facilitates a blending learning paradigm, also known as small private online courses (SPOCs). In order to understand how college students interact with an SPOC platform, this study collects their online behaviors for a semester and adopts a lag sequential analysis approach to identify significant transitions between interactions with content, peers, and instructors. Regarding content, after entering courses, the students tend to access learning resources. Besides, the transitions between learning resources and personal performance are significantly interconnected to each other. Regarding peers, the interaction with classmates was mainly connected to the access of assignments and performance. Regarding instructors, the interaction with teachers was minor but connected to all other behaviors. In addition, the results also show that students’ online behavioral patterns in SPOCs may change over time. The implications of the findings for SPOCs research are discussed in this paper.

Feature Selection for Online Writeprint Identification Using Hybrid Genetic Algorithm

One major task of online writeprint identification is to select the key features for representing the writeprint and facilitating the classifier built by using only the selected feature subset. In this study, we develop a hybrid genetic algorithm: RelieF Fed Genetic Algorithm (RFGA) which incorporates feature weight information produced by using RelieF as the heuristic to identity the key features and improve the identification performance. Experiments are conducted on a test bed encompassing hundreds of reviews posted by 20 Amazon customers to examine the method. The experimental results using RFGA show the proposed approach is effective, obtaining a significant improvement in performance, with satisfactory classification accuracy of 96.67%, and having a heavy reduction in feature dimensionality that is only 3% of the no feature selection baseline.

Social Network Characteristics of Learners in a Course Forum and Their Relationship to Learning Outcomes.

Recently, learning analytics has become the focus in the interdisciplinary field of education technology. Among learning analytical approaches, social network analysis (SNA) plays a critical role in examining collective learning patterns. In this study, we collect the forum data in an undergraduate course from a university’s online learning system. On the one hand, SNA is adopted to investigate the learners’ social network characteristics including network structure and network positions. On the other hand, we adopt the Pearson correlation analysis to identify the relationship between social network positions (e.g., degree centrality, closeness centrality, betweenness centrality, prestige and influence) and learning outcomes of learners. The experimental results show that most high-performing learners are located in the core position of network. Moreover, there is a significantly positive correlation between learner’s social network centrality and learning outcomes, and high-performing learners have higher prestige and influence in the forum. The indepth analyses could help teachers establish effective interactive mechanism that meets knowledge skills of different individuals, as well as guide learners to help each other in collaborative learning.

Evaluating Children's Composition Based on Chinese Linguistic Features with Machine Learning

The traditional evaluation of composition is human evaluation which is time-consuming, laborious and easily affected by subjective. In recent years, the automatic essay scoring (AES) has become a hot issue in natural language processing, but few research focus on Chinese AES. Hence, this study designed a Chinese AES system and collected 4566 compositions from first grade to sixth grade students. We also extracted 43 linguistic features based on Chinese characteristic, and analysis these compositions based on three model by stepwise multiple regression technique and support vector machine. Results showed that the accuracy of classification is among 70~80%.

Integrating Clustering and Sequential Analysis to Explore Students' Behaviors in an Online Chinese Reading Assessment System

It is helpful for students and teachers to identify students reading abilities and testing strategies based on their behavioral records of reading tests in an online Chinese reading assessment system. In this study, a K-means clustering algorithm is used to divide students into three potential clusters, and the behavioral sequence diagram of each cluster is drawn by means of the lag sequential analysis. By comparing the characteristics and differences of clusters, this paper draws the following main conclusions: (1) For better reading performance, increasing the time of reading articles is more beneficial than directly searching for the answers in the articles according to questions and options; (2) Students with high reading abilities spend longer time on reading articles and inspecting items, but rarely alter options; (3) Students with low reading abilities, who spend longer testing time and have more behaviors of clicking on articles and items, are not focused enough on current questions; (4) Those students with low reading abilities, who spend shorter testing time, rarely have inspection behaviors. Finally, this paper puts forward some suggestions based on the reading ability and testing strategy of each cluster to improve students reading literacy and instruct teachers reading teaching activities.

Applying Item Response Theory to Analyzing and Improving the Item Quality of an Online Chinese Reading Assessment

In order to realize the individualized teaching of Chinese language in primary schools, this research has developed an online Chinese reading assessment for primary schools, which aims to record the students test process and analyze the development level of students Chinese reading ability. In this paper, the item response theory (IRT) is applied to the quality analysis of the assessment in terms of measurement attributes (difficulty, discrimination, guessing), item characteristic curve, item information function and test information function. Additionally, this paper further explores the relationship between the various parameters of the items. The results show that the IRT can effectively guide the construction of the reading assessment scale, improve the discrimination degree, reduce the guessing degree, and effectively improve the quality of the item. This paper also proposes a method to find out the unreasonable options and modify items locally through project parameters and option analysis. It is expected that researchers and educators can modify the item more efficiently by quality analysis.

Discovering the Two-Step Lag Behavioral Patterns of Learners in the College SPOC Platform

Due to high retention rates, small private online course (SPOC) has become increasingly popular among universities. However, existing analyses of learning behavioral patterns in SPOC remain extremely lacking. This present study conducts an empirical analysis on the behavioral patterns of 12,517 undergraduates engaging in a colleges SPOC platform, called StarC. In this study, the authors collected and summarized the learning behaviors generated from these learners during 348 days of observation. They further coded the behaviors and extracted the two-step lag sequences in learning processes of individuals. The frequency analysis and sequential analysis were subsequently adopted to discover the distributions and frequency transition patterns of the two-step behavioral sequence in StarC. Besides, grade similarities and differences were computed and analyzed in terms of behavioral patterns. With these results, the potential and inadequacies of the learning platform are discussed, and some suggestions are offered for future work on the study and development of SPOCs.