依据教育理念；解决教学问题；完成教学目标和学习目标。

学习过程的数据：是学生在学习过程中积累的大量数据。这类数据有线上与线下之分。线上的学习数据有线上登录、浏览、收藏、播放、暂停、快进、停止，有交互中的提问、回答、质疑、检讨，还有在测试环节中的答题时长、提交次数、正确率等；线下的课堂表现数据，包括通过摄像头、捕捉监测学生的情绪动作，分析他们的表情、动作，并基于此判断学生的情绪，再综合判断学生的学习态度、效果。但我们要注意，问卷调查数据不太准确，因为学生不一定认真填写，主观性比较强，包括学习动机、学习风格，都设计了量表，并把电子版发给学生填写。宏观环境数据：这类数据包括产业数据和社会热点数据等，我们在教学过程、教学设计过程中也能用到。管理数据：学校教务部掌握的诸如选课学生人数数据，专业及学科课程体系数据，以及学生各科课程成绩等数据等，这些也可用来作为混合教学的数据基础。此外，餐饮数据和社交数据也在管理数据范畴，餐饮数据涉及学生就餐的时间等。

Principle 6: Higher education cannot afford to not use data

Principle 5

Principle 4: S

Principle 3:

Principle 2:

Principle 1:

a) The location and interpretation of data b) Informed consent, privacy and the de-identification of data ETHICAL ISSUES AND DILEMMAS 4 c) The management, classification and storage of data

Stakeholders view lack of clarity around student data policies as a key barrier inhibiting both the interpretation of student data and ability to act on takeaways. Three key policy areas — Data Transparency, Fidelity & Responsible Use, and Consent & Privacy — are identified below that are highly important for administrators to consider and address in the near-term.

Technology & Infrastructure Ensure that technology and infrastructure eases the ability for users to leverage student data. Outline and communicate procedures for acquiring new education technology to create a seamless integration with existing campus infrastructure

Data Ethics, Privacy, & Policies every learner everywhere ' Establish and communicate institutional data policies surrounding the use of student data (beyond FERPA). Policies should include fidelity and responsible use, consent and privacy, and data transparency.

Strategies to execute • Define consistent course-level learning outcomes across general education and foundational courses with multiple sections to enable analysis at scale. • Working with key stakeholders, set and share quantifiable goals and anticipated outcomes at both the institution and course level. • Faculty review their de-identified learning outcome data by student subgroup. • I nclude key stakeholders in the interpretation and review. • Create cross-discipline communities to i nterpret data and share best practices. • Develop a continuous learning culture and make adjustments to close identified gaps. • Especially in early stages, start your learning analytics efforts in ways that integrate with existing activities (e.g., use of systems) and processes (e.g., end of term review cycles).

1 . Equity & Learning Outcomes Explicitly set and communicate institution-level goals to achieve equity in academic outcomes across student groups, including students of color and low income, through the use of learning analytics.

The results indicated that outdegree centrality of a student, being in a group with higher average grade, communicating with the tutors, are the factors most correlated with better learning.

SNA can be used to map the relationships among actors, groups (visual analysis) and calculate the mathematical metrics of interactions and actors. The visual analysis was used to explore its role in shedding lights on the individual actors and their groups and how could this help learners or teachers. Furthermore, the mathematical analysis was used to quantify the interactions on the same levels, to demonstrate the utility of quantifications of interactions and how it compares to visual analysis.

roup level For each group, we collected the following parameters: The total number of interactions and type (student-stu dent, student-tutor, and tutor-tutor), average in-degree centrality (average in-degree of all group members), average degree centrality (average degree of group members).

Statistics were performed using Paleontological Statistics Software Package for Education and Data Analysis version 3.2. Correlation among variables was performed using the non-parametric Spearman rank correlation test

Problem-based learning (PBL) uses problems as triggers to facilitate dis course and interaction among students, the discussions occur in small groups and are facilitated by teachers

7KHPRVWFRPPRQ1/3DSSURDFKWRDQDO\]LQJVWXGHQW ODQJXDJHLQ022&6KDVEHHQWKURXJKWRROVWKDWDQDO\]H HPRWLRQV

7KHPRVWUHSUHVHQWDWLYHDSSURDFKHVDUH'RQJ V GHVLJQRIWHDPFRPPXQLFDWLRQ3RO\SKRQ\ 7UDXVDQ0DWX5HEHGHD'UDJDQ $OH[DQGUX WKH.QRZOHGJH6SDFH9LVXDOL]HU7HSORYVDQG 5HDGHU%HQFK'DVFDOX6WDYDUDFKHHWDO'DVFDOX 7UDXVDQ0DWX0F1DPDUD 'HVVXV

TAALES focuses on SURYLGLQJH[WHQVLYHLQIRUPDWLRQDERXWWKHOHYHORI OH[LFDOVRSKLVWLFDWLRQSUHVHQWLQDWH[W

$Q alternative approach involves the calculation of the NATURAL LANGUAGE PROCESSING CHAPTER 8 NATURAL LANGUAGE PROCESSING & LEARNING ANALYTICS PG 95 IHDWXUHVRIWKHZRUGVDQGVHQWHQFHVLQDWH[W

6HYHUDO advantages of n-gram analyses include their simplicity and the potential for providing information about the VSHFLāFFRQWHQWRIDWH[WWKHOLQJXLVWLFDQGV\QWDFWLF IHDWXUHVRIDWH[WDQGUHODWLRQVKLSVEHWZHHQWKRVH features (Crossley & Louwerse, 2007).

2QHDSSURDFKWR1/3LVWRDQDO\]HWKHZRUGVXVHGLQ WKHODQJXDJHGLUHFWO\

e models as those that in clude historical knowledge as to the effects of and intervention in the model itself. Thus a predictive model that used student interactions with content to determine drop out (for instance) would be an H[DPSOHRIāUVWRUGHUSUHGLFWLYHPRGHOOLQJZKLOH a model that also includes historical data as to the effect of an intervention (such as an email prompt or nudge) would be considered a second order predictive model. Moving towards the modelling of intervention effectiveness is important when multiple interventions are available a

Creating community-led educational data science challenge initiatives .

Supporting non-computer scientists in predictive modelling activities  7KHOHDUQLQJDQDO\WLFVāHOG i s highly interdisciplinary and educational re searchers, psychometricians, cognitive and social SV\FKRORJLVWVDQGSROLF\H[SHUWVWHQGWRKDYH VWURQJEDFNJURXQGVLQH[SODQDWRU\PRGHOOLQJ 3URYLGLQJVXSSRUWLQWKHDSSOLFDWLRQRISUHGLFWLYH modelling techniques, whether through the inno vation of user-friendly tools or the development of educational resources on predictive modelling, could further diversify the set of educational researchers using these techniques.

(e.g., Barber & Sharkey, 2012), each corresponding to a different time period and set of observed variables. For instance, one might gen erate predictive models for each week of the course, incorporating into each model the results of weekly TXL]]HVVWXGHQWGHPRJUDSKLFVDQGWKHDPRXQWRI engagement the students have h

3UHGLFWLYHDQDO\WLFVDUHEHLQJXVHGZLWKLQWKHāHOGRI teaching and learning for many purposes, with one VLJQLāFDQWERG\RIZRUNDLPHGDWLGHQWLI\LQJVWXGHQWV at risk in their academic programming. For instance, Aguiar et al. (2015) describe the use of predictive models to determine whether students will graduate from secondary school on time, demonstrating how the accuracy of predictions changes as students advance from primary school through into secondary school. 3UHGLFWHGRXWFRPHVYDU\ZLGHO\DQGPLJKWLQFOXGHD VSHFLāFVXPPDWLYHJUDGHRUJUDGHGLVWULEXWLRQIRUD student or class of achievement (Brooks et al., 2015) LQDFRXUVH%DNHU*RZGDDQG&RUEHWWGHVFULEH a method that predicts a formative achievement for a student based on their previous interactions with an intelligent tutoring system. In lower-risk and semi-formal settings such as massive open online courses (MOOCs), the chance that a learner might disengage from the learning activity mid-course is another heavily studied outcome (Xing, Chen, Stein, 0DUFLQNRZVNL7D\ORU9HHUDPDFKDQHQL  O’Reilly, 2014). Beyond performance measures, predictive models have been used in teaching and learning to detect learners who are engaging in off-task behaviour (Xing DQG*RJJLQV%DNHUVXFKDVÜJDPLQJWKH V\VWHPÝLQRUGHUWRDQVZHUTXHVWLRQVFRUUHFWO\ZLWK  out learning (Baker, Corbett, Koedinger, & Wagner, 3V\FKRORJLFDOFRQVWUXFWVVXFKDVDIIHFWLYHDQG emotional states have also been predictively modelled 'Ú0HOOR&UDLJ:LWKHUVSRRQ0F'DQLHO *UDHVVHU 2007; Wang, Heffernan, & Heffernan, 2015), using a YDULHW\RIXQGHUO\LQJGDWDDVIHDWXUHVVXFKDVWH[WXDO GLVFRXUVHRUIDFLDOFKDUDFWHULVWLFV0RUHH[DPSOHV of some of the ways predictive modelling has been XVHGLQ(GXFDWLRQDO'DWD0LQLQJLQSDUWLFXODUFDQ EHIRXQGLQ.RHGLQJHU'Ú0HOOR0F/DXJKOLQ3DUGRV and Rosé (2015).

Module level access - accessible from within a Blackboard module area. At a module level you will see predictions for your module cohort only. Global advisor access - accessible from the main Blackboard institutional page. Global advisor access allows you to view any student, any programme or any module at Ulster.

Week 0 - before the course starts 0-20% (ex: weeks 1-3) 20-40% (ex: weeks 4-6) 40-60% 60-80% 80-100%

The “Random Forest” machine learning algorithm is used to make predictions. The result is a set of decision trees created using a random subset of features. This results in each tree having its own path for predicting the outcome.   For any given prediction, the data is passed through each tree in the forest and the results are averaged to determine the final prediction for the case.

Predict ingests data from Banner, Ulster’s Student Records System, and combines it with interactions in Blackboard Learn. The model does not rely solely on Blackboard interactions, but those modules and programmes which make use of assessment and interactivity in the VLE will see more detail in the dashboards.   

OUA employs three machine learning methods: (1) Naïve Bayes classifer (NB), (2) Classifcation and regression tree (CART), (3) k-Nearest Neighbours (k-NN). Those are used to develop four predictive models: (1) NB, (2) CART, (3) k-NN with demographic data, and (4) k-NN with VLE data.

Academic resistance model

CALL FOR COMMUNICATION AND COLLABORATION: EDM and LAK There is a positive value to having different communities engaged in how to exploit “big data” to improve education. In particular, different standards and values for “good research” and “important research” exist in each community, allowing creativity and advancement that might not otherwise occur in a single, monolithic research culture. For example, EDM researchers have placed greater focus on issues of model generalizability (e.g. multi-level cross-validation, replication across data sets). By contrast, LAK researchers have placed greater focus on addressing needs of multiple stakeholders with information drawn from data. Each of these issues are important for the long-term success of both fields, a key opportunity for the two communities to learn from one another. Friendly competition between the two communities will keep both communities vigorous, and is generally beneficial for science. This type of competition has occurred in the past, such as in the split between the International Conference on the Learning Sciences and the International Conference on Artificial Intelligence in Education in 1992. Research networks are increasingly global, as reflected by the multi-national executive committees of IEDMS/EDM and SoLAR/LAK, but reflect different nations to a significant degree. Hence, the existence of both communities broadens the number of researchers working and collaborating in the broader area of data-driven discovery in education. At the same time, it is very important to keep competition healthy. Healthy competition requires that both communities disseminate their research to each other through their respective conferences and journals to ensure awareness of important ideas and advances occurring in the other community. The two communities must communicate, in order to bring the greatest possible benefits to educational practice and the science of learning.

between what work appears in the two communities. One key distinction is found in the type of discovery that is prioritized. In both communities, research can be found that uses automated discovery and research can be found that leverages human judgment through visualization and other methods. However, EDM has a considerably greater focus on automated discovery, and LAK has a considerably greater focus on leveraging human judgment. Even in research which combines these two directions, this preference can be seen; EDM research which leverages human judgment in many cases does so to provide labels for classification, while LAK research which uses automated discovery often does so in the service of informing humans who make final decisions. This difference is associated with another difference between the two communities: the type of adaptation and personalization typically supported by the two communities. In line with the greater focus on automated discovery in EDM, EDM models are more often used as the basis of automated adaptation, conducted by a computer system such as an intelligent tutoring system. By contrast, LAK models are more often designed to inform and empower instructors and learners. A third difference, and an important one, is the distinction between holistic and reductionistic frameworks. It is muc

. Firstly, in order to allow for successful adoption, institutional strategies and goals must be underpinned by learning analytics principles. Secondly, to address the needs of various stakeholders, we need to plan (e.g., when designing a course) before undertaking specific learning analytics activities. Finally, (and similar to knowledge management) l earning analytics can be observed as an “evolutionary, iterative process directed by feedback loops and learning” (Rubenstein-Montano et al., 2001, p. 13).

Discourse analytics: Understanding student communications

Social network analysis (SNA) quickly emerged as one of the cornerstones of the l earning analytics research (Dawson et al., 2014).

From early predictions to multimodal learning analytics

The concept of learning analytics can be traced back to the work of Pressey ( 1 927) who developed the first automated teaching machine i n the 1920s. The work of Pressey (1927) can be argued as the start of intelligent tutoring systems (ITS), one of the key areas upon from which learning analytics draws. Similarly, another critical influence has been cognitive science, which originated in the work of Miller (1956) and new advances in computer science and artificial intelligence.

efining of learning analytics as the “measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimizing learning and the environments in which it occurs”

The goal is to better understand and support learning processes and outcomes through both short‐cycles improvements to educational practice and long‐cycle improvements to the underlying knowledge base.

Key systemic and societal issues that will determine the fate of learning analytics include deliberate consideration of the policy needs required to govern the ethical dimensions of analytics use and proactive planning for the required infrastructure.

An initial class of pedagogical use of learning analytics is for tailoring educational experiences to better meet the specific needs of one of more students. In this model of use,

Different from tailoring the materials that are given to students, another pedagogical use of analytics is to support students in conscious attention to and improvement of their own learning processes.

For instructors, pedagogical uses of learning analytics can be used to support refinement of both the overarching learning design and the decisions they make to orchestrate classroom activity within it. F