International Data Corporation (IDC) believes that data in the era of big data has four characteristics-large scale, great value, fast data flow and many data types. The mining and utilization of big data has a far-reaching impact on education, especially classroom teaching. Sawyer, a learning scientist, believes that more and more learning will be conducted through computer intermediaries and more and more data will be generated. It is necessary for us to use these data to analyze when effective learning occurs. Therefore, data mining can be used to explore the relationship between behavior and learning, such as the relationship between individual differences of learners and learning behavior, and what different learning results will be caused by different behaviors. In 20 12, the United States published "Strengthening Teaching and Learning through Educational Data Mining and Learning Analysis", which put forward the characteristics of educational data in the era of big data: hierarchical, temporal and situational, in which hierarchical data refers to the collection of data from both teachers and students, including classroom data and activity data, which provides multi-dimensional resources for the establishment of later models; The time series of data means the real-time and continuity of data, which guarantees the cutting-edge of materials; Contextuality of data means that data is based on real situational context, which ensures the reliability of the model.
Big data technology can promote student-centered learning. Data is not only to collect data and synthesize models under the guidance of scientific and technological rationality, but also to predict and judge students' group behavior. We can also modify the inherent model by diagnosing students' behavior in class on the basis of the inherent model, so as to make the course content more suitable for students' long tail needs and realize personalized teaching. The use of big data can support the modeling and prediction of educational activities, and may also support adaptive teaching in educational practice. The former is the foundation of the latter, and the latter is the deepening of the former.
Application of Big Data for Modeling and Forecasting
One of the ways that data promotes education reform in the era of big data is to collect, analyze, process data and make predictions. Nowadays, due to the convenience of data recording, storage and operation, it is convenient to collect massive and multi-level data and reduce the errors caused by random sampling. Modeling and forecasting can be based on full data and real data, so it is more accurate. In the era of big data, the most successful case is Amazon's marketing by exploring the relationship between massive data. Amazon collects the data of readers' online browsing behavior and purchasing behavior, establishes a reader's reading preference model, predicts the group behavior of readers' purchasing, and realizes book recommendation. In recent years, the object of educational research has gradually paid attention to students' learning behavior, behind which is the change of learning concept. Learning is regarded as a process of understanding, and understanding is an activity, rather than transferring knowledge as an object. Knowledge is always contextualized, not abstract and divorced from specific situations. Knowledge is constructed interactively in the interaction between individuals and the environment, which is not objective and accurate, nor is it created subjectively. Therefore, the data of students' behavior activities are considered to reflect the situation of students in the learning process, which is a dynamic process of situation change. Massive, multi-level and continuous behavior data are collected and fitted to the model, so as to realize prediction, such as the application of learning management system (LMS). However, because the basic principle of modeling and prediction dependence is mathematical statistics, its prediction object is mainly the group behavior of students.
1. Case study
Learning Management System (LMS) is a network-based management system platform, which is used to monitor students' learning activities, identify and predict students with learning difficulties, and provide them with corresponding help. Most LMS consists of five parts: learning materials related to courses, evaluation tools to ensure that students submit homework and complete tests, communication tools (such as emails and chat rooms), curriculum management tools to ensure that teachers record and store students' learning activities and announce the deadline of activities, and learning management tools to help students learn to review and track the learning process. BB(Blackboard) platform is a common learning management system, which is widely used in colleges and universities. The system records the types of online courses that students attend, the online duration, the number of articles read and browsed, and reflects learners' learning behavior. In 2008, Leah P.Macfadyen and Shane Dawson established a prediction model by analyzing the data of five undergraduate classes of the University of British Columbia studying biology on BB platform. The platform records the use of students' course materials, their participation in academic exchanges and the completion of homework submission and review. The hierarchy of educational data records in the era of big data is fully reflected here. The use of course materials includes recording online time, email reading time, email sending time, discussion information reading time and so on. Participate in academic exchanges and record the time of publishing new discussions, replying to discussions, using search tools, accessing personal information, browsing files, browsing who is online at the same time, browsing web links, etc. The evaluation module records the reading time of evaluation and the time of submitting evaluation. By using statistical tools to describe the scatter plot, it is found that the online time recorded by LMS is related to academic performance. In multiple regression, the researchers found that the last quarter of students' academic performance spent a little longer than the average online time, while the first quarter spent less than the average online learning time. Then, in order to make predictions, researchers use logistic regression to generate prediction models. By collecting students' new behavior data, they predict whether students really participate in learning activities, and draw the following conclusions: the model composed of three dimensions: the number of discussions, the amount of email information sent and the completion of evaluation can predict students' academic level.
In the era of big data, exploring the correlation between students' behavior and academic level, establishing models and realizing predictions can have an important impact on classroom teaching. However, in the process of data modeling, in order to ensure the validity and reliability of the model, extreme individual data are processed, so that the model can only predict group behavior, but not customize and personalize individual learners.
2. Insufficient modeling and forecasting.
The ideas and methods of positivism are fully embodied behind data modeling and forecasting. /kloc-in the first half of the 9th century, sociologists represented by Comte put forward the basic creed of positivism: explore the relationship between them by observation and classification, and get scientific laws. In the 1960s, the philosophical trend of positivism evolved into a scientific and technological rationality, practical knowledge gradually became instrumental, and professional activities existed in instrumental problem solving. All professional activities are regarded as the process of setting goals and applying known methods to solve problems. During this period, a large number of disciplines have been systematically integrated and developed, even including "soft science" such as pedagogy and sociology. It is a trend to solve unknown problems with evidence and predict the future with data.
The modeling of student activity behavior data pays special attention to the idea of empirical positivism. When the model pays attention to the commonness of successful teaching behavior and ignores the unique needs of teachers and students, the leading of scientific and technological rationality may make classroom teaching be regarded as a module independent of the real context. As long as the teaching behavior is successful, it will be abstracted from the data to form a model to predict the students' behavior. The rationality of science and technology depends on people's recognized common goals, and the determination of teaching practice goals is extremely complicated, which contains great uncertainty and uniqueness, and even brings value conflicts due to different social roles. A stable goal recognized by everyone no longer exists, and the behavior pattern predicted by scientific and technological rationality and methods can not meet everyone's needs. Education reform has a new orientation in the era of big data.
From data model to supporting adaptive learning
Realizing the adaptability of teaching on the basis of data modeling is another achievement of promoting education reform in the era of big data. Data modeling and behavior prediction still belong to the behavior mode under the guidance of scientific and technological rationality, which may lead to the neglect of students' individual needs, which is an important feature of knowledge society. Personalized education has attracted more and more attention from educational researchers, policy makers and educational practitioners. Lagos, an expert in educational system design, believes that a very important reason why educational investment has not achieved results is that social transformation has been ignored. "The society has stepped into the information age from the industrial society. The demand for talents in the labor market is no longer workers operating on the assembly line in the industrial age, but intelligent talents with innovative thinking and strong determination. Teaching is facing the transformation from producing uniform labor to producing people with judgment and adaptability. In 20 10, the OECD report "the essence of learning" pointed out that adaptability is the core competitiveness of 2 1 century, including flexible and creative use of meaningful knowledge and skills in real situations. WU GANG put forward the necessity and inevitability of personalized education in "Personalized Education in the Age of Big Data: Strategy and Practice", pointing out that "only with the strong support of information technology can personalized learning be truly realized". The arrival of the era of big data is a good opportunity for the development of personalized education. In 20 12, the United States promulgated "promoting teaching and learning through educational data mining and learning analysis", which proposed that in the era of big data, online learning data should be collected, classified, and the correlation between data should be explored to form a data model for data mining. Through the interaction between students' behaviors and models, an adaptive learning system is formed. In a word, we can make full use of behavior data, change the content and progress of teaching, build an adaptive evaluation and teaching system, fully realize the customization of education and meet the long tail needs of students.
1. Case study:
Adaptive teaching system, also known as adaptive learning support system (ALSS system for short), emphasizes active learning based on resources, and holds that learning is not the transfer of knowledge, but the self-construction of learners. Since 1990s, researchers have developed many adaptive learning systems, such as AHA system developed by De Bra in 1998, MLtutor system for task-based learning developed by Brandsford and Smith in 2003, and the flipped classroom model, referred to as FCM system, which has attracted much attention in recent years.
Content delivery module: delivering relevant knowledge and information to support students' learning.
Learner database: it stores the relevant behaviors of students participating in teaching activities.
Prediction module: including student information and student behavior data, tracking students' learning situation and making predictions.
Display module: generate behavior report for students.
Adaptive module: feedback the report generated according to students' behavior to the preset model, and change the model accordingly to make it more in line with students.
Intervention module: Teachers, system managers and leaders can implement human intervention when the system is running.
When learners study related subjects, their learning behavior is recorded and tracked, and the data of students' learning behavior is sent to the background, recorded in the learner database, and acted on the prediction module. The prediction module acts on learners again by changing the content delivery module. In the whole process, teachers and teaching managers play an interfering role.
Adaptive learning system is an interactive dynamic system, which often provides some suggestions for students' learning behavior. Austria designed an adaptive learning system for students' problem solving process. The first step of adaptive learning system is educational data mining. The process of data mining includes data collection, data preprocessing, application data mining and interpretation and evaluation of development results. Moodle proposed CMS (Curriculum Management System). Researchers first use raw data to model. The first step is to collect raw data, including about 28,000 active cases generated by 73 users in 2007, 265,000 problem-solving cases generated by 97 users in 2008 and 1 15000 active cases generated by 45 users in 2009. The original data not only records the data generated when students answer questions, but also collects students' information, questions' information and the steps to solve them. After classifying the data, the types of problem solving are summarized, and DMMs, a sub-model of Markvo model (MMS), which is good at fitting continuous data, is used to fit the above continuous data. By adding a result model to judge students' learning behavior and a series of monitoring and adjusting modules, an adaptive system for solving the whole problem is formed. When students use this model, the model will provide students with their preferred process and methods to solve problems according to their behavior data.
Besides the adaptive teaching system, there is also an adaptive evaluation system. LON-CAPA (Computer Aided Personalized Learning Online Network) is a computer aided personalized online learning evaluation platform. The platform does not provide curriculum design and curriculum objectives, but is only a teaching tool. CAPA records students' basic information, students' interaction and academic situation in the background, and provides personalized examination resources for the difficulties in academic courses.
2. The significance of adaptive steering
In the era of big data, the model prediction under the guidance of scientific and technological rationality is unable to cope with the problem of poor structure. Data modeling under the guidance of scientific and technological rationality ignores the real context of learning and can only support the prediction of group behavior. The popularization of the model may make people ignore the individual experience and specific situation behind its practical success, thus leading to the confrontation between scientific and technological rationality and philosophical speculation. However, relying entirely on philosophical speculation and experience in teaching is not only detrimental to the theoretical development of educational disciplines, but also to the management of classroom practice and the cultivation of teachers. Donald A. Schon put forward the adaptive thinking mode. He pointed out: "If the rational mode of science and technology is incompetent, incomplete or even worse in the face of" diverse "situations, then let us look for an alternative, more practical, artistic and intuitive practical understanding. Adaptive learning is a kind of learning method that makes learning content and activities highly personalized according to individual differences under the guidance of system theory knowledge.
Adaptability balances the dilemma between rationality and experience British scholar Hargreaves (1996) first proposed that evidence-based education research should be closer to medical diagnostics. The similarity between clinical diagnostics and pedagogy lies in that both of them have to face a changeable and extremely complicated environment. In such a poorly structured system, they are fully aware of the uniqueness and commonness of the object (patient or student) and use the professional knowledge of the system to solve the problem.
Professor Ralf St. Clair put forward three elements of evidence-based education research after referring to the three elements of medical clinical practice research: research evidence, educators' experience, learners' environment and characteristics. Among them, behavior prediction focuses on research evidence, while the construction of adaptive learning system focuses on the experience of educators and the environment and characteristics of learners.
The change from predicting behavior to supporting adaptive teaching is a humanistic change. The focus of educational research has shifted from focusing on the evidence of research to focusing on the experience of educators and the characteristics of learning environment, and focusing on the practical reform of supporting individualized learning with evidence. In the era of scientific and technological rationality, evidence is no longer its role in guiding decision-making, but is regarded as a resource. Educators find the most suitable way for their own characteristics and students' characteristics in a large number of evidence-based classroom teaching decisions, and promote the classroom teaching process. In other words, the more important value of big data is to support adaptive learning and meet the needs of personalized learning and personalized development. The prediction function of data depends on the comprehensiveness of big data collection and the convenience of data processing. Predicting group behavior according to statistical principles weakens individual characteristics and specific situations to some extent. Mainly points to behavior prediction. Adaptability is to change the model in the interaction between the model and the object. As shown in Figure 3, the adaptive operation model of data has a cycle until there are more systems than the predictive model, which makes it more suitable for personal needs, and it mainly points to practical improvement. Prediction is the basis of supporting personalized learning, which is the deepening and transformation of prediction function-from the whole population to individual learners, from theoretical models to practical strategies.
Analysis and enlightenment
In the era of big data, due to the large amount of data and the convenience of data collection and carrying, the massive data of students' behaviors are mined and collected, and the analysis of learners' behaviors through data modeling becomes more comprehensive and reliable than in the previous era of big data. Although the data age has great potential in data mining and prediction, the more value of the big data age is to meet the long tail needs of learners. On the basis of predicting behavior, the teaching mode is modified to make it personalized and customized. From data modeling to supporting adaptive teaching, the object of support shifts from groups to individuals, and the influence on educational activities shifts from understanding of behavior to practice of educational activities, from the context of scientific and technological rationality to teaching activities based on real situations.
Adaptability not only changes human behavior, but also changes cognitive style. In the pre-big data era, people were completely driven by data under the guidance of scientific and technological rationality. Teachers and students, educational decision-makers and schools have formed a traditional social contract relationship, in which the parties give themselves 100% to professionals, who abide by the contract and are wholeheartedly responsible for the parties, thus enabling professionals to enjoy the supreme monopoly position. In the era of big data, teachers are no longer the masters of knowledge. By participating in students' learning activities, the teaching steps, teaching progress and difficulty are constantly adjusted according to students' prior knowledge, cognitive characteristics and individual needs. Students don't have to completely entrust themselves to teachers as patients do to doctors. In the process of learning, through the interaction with teachers, with the help of teachers, become the main body of their own learning, control and be responsible for their own learning. Due to the limited energy of teachers, the network computer-aided learning system in the era of big data can provide teachers and students with opportunities for auxiliary guidance.
Nevertheless, on the one hand, we should embrace the convenient life and quality education brought by big data. On the other hand, it is necessary to guard against and prevent the misuse of causality and correlation to maintain data security.
In reasoning, educators need to be alert to the misuse of correlation and causality. Taking the case of Professor Leah P.Macfadyen as an example, the length of online time on BB platform is related to students' academic performance, but it is not causal. The online time of excellent students is lower than the average online time, but it can't be said that learning less than the average online time leads to excellent students' grades, and students are required to reduce online learning time.
In addition, in terms of information security, a lot of information of students and teachers is collected and used. In the process of use, relevant privacy protection laws must be formulated to ensure the security of information and prevent data abuse. Students' behavior data can not be used as the basis for teachers' teaching evaluation, so that big data can truly become a means to support teaching reform, improve teaching efficiency and promote students' development, rather than a tool to control teachers and students.