Table 3 - uploaded by Gautam Biswas
Content may be subject to copyright.
Source publication
This paper discusses our approach to building models and analyzing student behaviors in different versions of our learning by teaching environment where students learn by teaching a computer agent named Betty using a visual concept map representation. We have run studies in fifth grade classrooms to compare the different versions of the system. Stu...
Context in source publication
Context 1
... preliminary analysis consisted of examining the prevalence of each behavior in the resultant stationary probabilities. The derived stationary probability values are listed in Table 3. In a sense, this analysis is equivalent to the frequency count analysis that we have performed in other studies [3], and indicates an estimate of the relative time spent in each state. ...
Citations
... In this study, different learning strategies corresponded to different behavioral sequences, and learners employing different strategies had significantly different learning ROI. This finding is consistent with the conclusions of some scholars, such as Jeong and Biswas (2008), who found a strong correlation between the learning performance of middle school science students and the identified behavior patterns, and Kovanović et al. (2015), who identified effective learning strategies and established a significant relationship with high-level cognition from the perspectives of self-regulation, goal orientation, and cognitive presence. Similarly, Ahmad Uzir et al. (2020) used clustering methods to discover that students who employed time management strategies in a flipped classroom achieved better learning outcomes. ...
Began with Computer-Assisted Language Learning (CALL) in the 1960s and extended to the widespread use of various Mobile-Assisted Language Learning (MALL) tools in education, language learning has embraced technology early on, achieved noticeable results, and found extensive practical use. However, due to the challenges in accessing user data from various language learning platforms, the measurement and assessment of language-related variables continue to rely on self-reporting and peer evaluations. This reliance hampers researchers to observe language learning from alternative perspectives, especially when it comes to analyzing raw behavioral data. To explore potential correlations between different learning modes, this study analyzed 2 million samples from Chinese students using an English language learning application. The study quantified the effectiveness of English vocabulary learning using the economic concept of return on investment (ROI) as an evaluation metric and identified four distinct learning strategies. It observed significant differences in learning ROI among learners who adopted different strategies. Based on this analysis, we recommend the following suggestions for improving language learning ROI: when memorizing new vocabulary, investing excessive amounts of time may be counterproductive; a more effective approach is to "eat less but more often," which means arranging review sessions at a reasonable pace and shortening the interval between each review.
... It uses the essence of data mining in education to extract useful information about student behaviour in the learning process. EDM has been used in a wide range of fields, such as predicting student performance and detecting student behaviours [2], grouping similar materials or students based on their learning and interaction patterns [3], identifying relationships in learner behaviour patterns, and diagnosing student difficulties [4], identification of relationships among student behaviours and characteristics or contextual variables [5], interpretation of the structure and relations in collaborative activities and interactions with communication tools [6], and so on. EDM has become the basis for applying big data and machine learning techniques to improve education quality. ...
... Zimmermann et al. [30] used a model built with regression methods and various variable selection methods to predict Master's course performance using Bachelor's performance parameters. Jeong and Biswas [62] created a learning-byteaching model in which students teach a computer agent. ...
... SNA is used to examine educational processes involving collaborative learning [10]. Many other methods and procedures, such as optimization techniques [16], genetic algorithms [21], [22], [26], semisupervised learning methods [63], active learning methods [64], HMM [62], NMF [60], and so on, have been used in EDM. ...
Education and computer science are both involved in the burgeoning inter-disciplinary research field known as Educational Data Mining (EDM). EDM uses data mining software and ways to extract meaningful and practical data from big educational databases. EDM introduces better and more efficient learning techniques in an effort to enhance educational processes. The term "EDM methods" refers to a group of techniques for creating models and applications. This page provides a thorough literature review on EDM techniques. The essay also covers EDM research problems and trends.This EDM insight aims to provide researchers interested in furthering the field of EDM with useful and valuable information.
... Romero and Ventura's taxonomy [7] provides insight into numerous research in this area. For example, the problems of determining significant contributors that affect learner performance [8], [9]; the development of student learning profiles based on learner behavior data [6], [10], [11]; and the prediction of miscellaneous academic outcomes (student dropout, learner performance, learner behavior) [12]- [15]. ...
This study aims to investigate the potential of educational data mining (EDM) to address the issue of delayed completion in undergraduate student thesis courses. The problem of delayed completion of these courses is a common issue that impacts both students and higher education institutions. The study employed clustering analysis to create clusters of thesis topics. The research model was constructed by using expert labeling to assign each thesis title to a computer science ontology standard. Cross-referencing was employed to associate supporting courses with each thesis title, resulting in a labeled dataset with three supporting courses for each thesis title. This study analyzed five different clustering algorithms, including K-Means, DBScan, BIRCH, Gaussian Mixture, and Mean Shift, to identify the best approach for analyzing undergraduate thesis data. The results demonstrated that K-Means clustering was the most efficient method, generating five distinct clusters with unique characteristics. Furthermore, this research investigated the correlation between educational data, specifically GPA and the average grades of courses that support a thesis title and the duration of thesis completion. Our investigation revealed a moderate correlation between GPA, thesis-supporting course average grades, and the time to complete the thesis, with higher academic performance associated with shorter completion times. These moderate results indicate the need for further studies to explore additional factors beyond GPA and the average grades of thesis-supporting courses that contribute to thesis completion delays. This study contributes to understanding and evaluating the educational outcomes within study programs as defined in the curriculum, particularly concerning the design and implementation of thesis topics. Additionally, the clustering results serve as a foundation for future research and offer valuable insights into the potential of using EDM techniques to assist in selecting appropriate thesis topics, thereby reducing the risk of delayed completion.
... Furthermore, the aforesaid algorithms may be used to examine students' learning state and ability, as well as gather and anticipate their learning using their various data in the network. The following is a list of JEONG and BISWAS' primary research goals [8]: (1) creating student models (which contain learners' knowledge, motivation, metacognition, and attitude) to predict learners' future learning models; (2) exploring or improving domain models (which describe what has been learned and the ideal instructional sequence); (3) studying the effects of instructional support provided by various types of learning software; (4) increasing educators' scientific understanding of learning and learners by creating computational mode (including learner models, domain models, and software instructional models). ...
... To better understand the relationship between students' achievement and their interaction, as well as the ways in which knowledge and skills are building, temporal analysis of the learning process has been highlighted in recent years in learning analytics field (Knight et al., 2017;Lämsä et al., 2020). Various data, such as dialogue texts, videos, voices and log data, have been collected, and some mature typical techniques include process mining (Trcka et al., 2010), sequential pattern mining (Kinnebrew et al., 2013), Markov chains (Faucon et al., 2016;Hansen et al., 2017), hidden Markov models (HMM) (Geigle & Zhai, 2017;Jeong & Biswas, 2008), lag sequential analysis (LSA) (Bakeman & Gottman, 1997;Sackett, 1978) and epistemic network analysis (ENA) (Shaffer et al., 2009) ENA, a method advocated by Shaffer et al. (2009), 'is a set of techniques that identifies and measures connections among elements in coded data and represents them in dynamic network model' (Shaffer et al., 2016, pp. 9-10). ...
Background
As a non‐cognitive trait, grit plays an important role in human learning. Although students higher in grit are more likely to perform well on tests, how they learn in the process has been underexamined.
Objectives
This study attempted to explore how students with different levels of grit behave and learn in an exploratory learning environment.
Methods
In this study, 66 students participated in seven exploratory tasks in Snap! for approximately 60 min after a 30‐min lecture. Students were categorized into a high grit group and low grit group using a grit scale. The Mann–Whitney U test, epistemic network analysis and lag sequential analysis were used to explore the differences between groups in learning performance, technology acceptance and behavioural patterns.
Results
Students with different levels of grit engaged in explorative tasks in a short period of time might not present significantly different learning performance, perceptions of usefulness and ease of use, but students higher in grit, actively engaging more different types of activities, tended to put greater sustained effort to solve the challenging task.
Take Away
Although grit was not significantly correlated with learning performance when students engage in classroom‐based explorative activities, grit did predict whether students are more likely to explore and put greater sustained effort into solving the challenging task.
... A major area of educational data mining research [4,5,7,8] is done to analyze MOOC and learning management system log data to identify patterns of learning behavior that can provide insights into educational practice. Research in educational data mining [2,6] is done towards building student models and analyzing student behaviors in various interactive learning environments to predict student learning behaviors. For instance, hidden Markov models (HMM) were used to model school students' behavior based on the trace data generated from Betty's brain system which used the pedagogy of learning by teaching [2]. ...
... Research in educational data mining [2,6] is done towards building student models and analyzing student behaviors in various interactive learning environments to predict student learning behaviors. For instance, hidden Markov models (HMM) were used to model school students' behavior based on the trace data generated from Betty's brain system which used the pedagogy of learning by teaching [2]. In a later study, Jeong et al. (2010) applied the same HMM approach to study the learning behavior of adult professionals in an asynchronous online learning environment. ...
... We propose to use HMM similar to the work proposed by Jeong (2008) to investigate how engineering students interact with learning environments designed for complex problem solving and analyze student behaviors to get insights into how the learning environment facilitates learning of complex problem solving among high and low performers. ...
Identifying the various cognitive processes that learners engage while solving an ill-structured problem online learning environment will help provide improved learning experiences and outcomes. This work aims to build a student model and analyze student behaviors in our technology-enhanced learning environment named Fathom used for teaching-learning of ill-structures problem-solving skills in the context of solving software design. Students’ interactions on the system, captured in log files represent their performance in applying the skills towards understanding the problem as a whole and formulating it into subproblems, generating alternative designs, and selecting the optimal solution. We discuss methods for analyzing student behaviors and linking them to student performance. The approach used is a hidden Markov model methodology that builds students’ behavior models from data collected in the log files.
... QRF listened for two types of events: affective sequences and behaviors related to self-regulated learning strategies (more details in section 3.2). Using previously integrated affect detectors (outlined below and in Jiang et al., 2018), we set the serverside process to listen for affect sequences that are aligned with theoretical models of affect dynamics in educational contexts (D'Mello & Graesser, 2012), and predefined action sequences relevant to SRL (Jeong & Biswas, 2008). ...
Artificial Intelligence in Education research for STEM domains has largely been quantitative in nature, but qualitative research offers several advantages as part of a mixed-methods approach. In particular, qualitative research enables researchers to develop deeper phenomenological understanding of how learners represent their activity to themselves. However, qualitative research can be challenging to apply in classrooms: it is resource-intensive, does not scale well, and the phenomena of the greatest interest to AIED researchers are often intermittent and occasional. For example, researchers may be interested in studying situations where a learning activity is known to be overly time-consuming or difficult, or in theoretical investigations of shifts in student affect such as transitions from confusion to frustration. However, given multiple potential learners to interview (e.g., a classroom of students), it can be difficult for a researcher embedded in the classroom to prioritize which learner to speak with next. Simple strategies, whether sequential or random, may miss (often fleeting) key moments in a participant's experience (e.g., affective transitions). We address this problem with a new app that leverages user modeling techniques (e.g., behavior and affect-sensing) to direct interviewers to learners at critical, theory-driven moments as they learn with AIED technologies in the classroom. This paper details the design and implementation of this research paradigm as an alternative method for studying learning and using existing STEM AIED technologies in research. We examine the potential of this paradigm through the lens of two case studies where 99 students interacted with a computer-based learning environment as part of their regular classroom instruction. Unscripted interviews were triggered at or immediately after critical moments (such as peak frustration or shifts from confusion to boredom). The app facilitated 594 interviews, each averaging 1-2 minutes in length. Our findings indicate that by using machine learned models to optimize researcher time, we can gain a deeper insight into students' behaviors and their motivations, thus furthering AIED research. We discuss the potential broader applications of this app and the research it affords.
... Different methods have been proposed to process sequencetype student action data dependent on the amount of data, the length of sequences, and the goal at hand. Several lines of research rely on Markov chains and hidden Markov models which lend themselves well to visualization of sequences, but can make quantification of group differences in outcomes challenging [14,13,19]. Another commonly used class of methods is clustering of activity sequences [12,22,16]. ...
Educational software data promises unique insights into students' study behaviors and drivers of success. While much work has been dedicated to performance prediction in massive open online courses, it is unclear if the same methods can be applied to blended courses and a deeper understanding of student strategies is often missing. We use pattern mining and models borrowed from Natural Language Processing (NLP) to understand student interactions and extract frequent strategies from a blended college course. Fine-grained clickstream data is collected through Diderot, a non-commercial educational support system that spans a wide range of functionalities. We find that interaction patterns differ considerably based on the assessment type students are preparing for, and many of the extracted features can be used for reliable performance prediction. Our results suggest that the proposed hybrid NLP methods can provide valuable insights even in the low-data setting of blended courses given enough data granularity.
... Zimmermann et al. [30] used a model built with regression methods employing various variable selection methods to predict the performance of the Master's course using parameters of Bachelor's performance. Jeong and Biswas [62] built a learning-by-teaching model that lets the students learn by teaching a computer agent. ...
... SNA is used to analyze the educational processes that involve collaborative learning [10]. Many other methods and procedures have been used in EDM, including optimization techniques [16], genetic algorithms [21], [22], [26], semi-supervised learning methods [63], active learning methods [64], HMM [62], NMF [60], etc. ...
Educational Data Mining (EDM) is an emerging
inter-disciplinary research area that involves education and
computer science. EDM employs data mining tools and
techniques, on large datasets related to education, to extract
meaningful and useful information. EDM works toward the
improvement of educational processes by introducing better
and effective learning practices. EDM methods refer to the set
of methods that are used for building models/applications. This
article presents an extensive literature survey of EDM methods.
The article also discusses research trends and challenges in
EDM. This insight into EDM attempts to provide useful and
valuable information to researchers interested in furthering
the field of EDM.
