ADAPTIVE LEARNING : PERSONAL RECOMMENDATION MODELS REVIEW Текст научной статьи по специальности «Науки об образовании»

ТЕХНИЧЕСКИЕ НАУКИ

ADAPTIVE LEARNING : PERSONAL RECOMMENDATION MODELS _REVIEW_

Ammar Wisam Altaher

PhD Student in the Department of programming technologies

Kazan Federal University /Russia Address

Russia / Kazan, Respublika Tatarstan Postal Code: 420010 ORCID ID /0000-0002-1373-0665

ABSTRACT

Higher education is notorious for adjusting slowly to change. Despite its status quo position on higher learning, the pedagogical view that instruction must be taught in the traditional classroom has long been defeated. Adaptive learning technologies, a hot topic among postsecondary pundits at the moment, seem to have risen in its wake.

The aim of this paper is to clear the concept of adaptive learning in the delivery of education or training with utilises technology and data to provide an individually customised learning program to students, intelligently adapting to their learning needs.

Keywords : Adaptive Learning, Personal Recommendation, E-Learning , M-Learning .

© Ammar Wisam Altaher - Iraq - ministry of higher education and scientific research / Al-Furat AL-Awsat Technical University

1. Introduction (Heading 1)

Adaptive Learning Defined

The basic premise of adaptive learning is using technology to improve education and training by providing individualised learning programs to students based on data that is gathered both before and throughout the learning process. The best adaptive learning platforms use some form of data mining to put together learning content for students that's optimised for their learning needs. The platform uses data that it continually gathers when a student interacts with any learning content.

Using adaptive learning, each student goes through a highly individualised learning experience, which should provide much better learning results.

Introduction

Learning Management Systems (LMS) have spread among teachers' communities in virtue of their friendly interfaces for organizing existing contents, authoring new ones, and finally deploying them for learners on the Web. Despite their improvements, the functionalities of all of the LMSs on the market do not permit the teacher to take advantage of user's information, that would allow to overcome the "one size fits all" problem of web-based courses. Moreover, studies from the Instructional Design field [1] show how much the adaptive instruction paradigm has been, and still is, a common trait in every day instructional situation: a teacher in a classroom naturally adapts his/her learning goal, presentation style, instructional strategy, language to match the needs of the class, thus why this can not happen online?

Learning design (LD) is about identifying necessary learning activities and assigning Learning Objects (LOs) to those activities in order to achieve a specified

learning objective. IMS Learning Design (IMS-LD) Specification

[1] provides a common set of concepts for representing LDs, enabling one to specify LDs targeted for different learning situations, based on different pedagogical theories, comprising different learning activities where students and teachers can play many roles, and carried out in diverse learning environments. However, neither IMS LD nor other learning specifications (e.g. IEEE LOM) capture enough information that can be used to provide advanced levels of learning process personalization, such as personalization in accordance with the students' objectives, learning styles, and knowledge levels. As a result, the annotations of the developed LOs and LDs do not contain explicitly represented information that is important for personalization. Effective personalization requires

[2]: 1) direct access to low-granularity content units comprising the structure of a LO;

2) recognition of the pedagogical role played by each content unit in a specific context (i.e. learning activity in terms of IMS LD);

3) awareness of learner's evaluations about usefulness of a specific content unit within a specific LD;

4) characteristics of learners that best fit a specific

LD.

The issue:

Adaptive technologies in the field of education have proven so far their effectiveness only in small lab experiments, thus they are still waiting for being presented to the large community of educators. There are several reasons for this low degree of diffusion in the practice of teacher communities. First of all, as pointed out by some recent studies [2], adaptive educational hypermedia systems are difficult to design, set-up, and

implement, due to the high technical competencies required. In particular, all of the (few) existing generalpurpose adaptive educational systems (we will call them Adaptive Educational Platforms - AEP) have a steep learning curve, which forbids a non technical teacher to autonomously create his course. In our opinion a viable direction to spread the seed of adaptivity among the, possibly large, communities of instructors consists in extending current LMS approaches with a set of simple and ready to use well established adaptive techniques (i.e. course sequencing, link annotations and conditional fragments), packaging them with straightforward authoring interfaces, and couple them with tutorials, and patterns for the most common instructional strategies.

In classrooms where a high degree of implementation is achieved, teachers tend to spend more time on instruction than on managing students and students tend to be highly task oriented. Steady and productive interaction between teachers and students, and among students, replaces the passive learning mode typically found in conventional classrooms. Interactions among students, for the most part, focus on sharing ideas and working together on learning tasks. Distracted behavior on the part of individual students is minimal and does not seem to interfere with the work of others.

Standardized achievement test scores in reading and math indicate that implementation of the model consistently leads to student achievement that meets or exceeds expected gains. Achievement results from various sites over the years have compared favorably with comparison sites in terms of national test norms, as well as district and population norms. Significant differences have been found with special education students who are integrated in regular Adaptive Learning Environments classes.

Adaptive Learning System Models:

Inference Models

Think of inference models as having a predetermined map of pathways students can follow when using adaptive learning technology. Although the paths may be predetermined, they are not limited - when content reaches upward of 500 or more topics, you can imagine how complex the map of pathways for students to follow can be.

As time moves on during a course and students complete more work and answer more questions using an inference based adaptive learning system, they continue to grow their personal map of pathways. Along the way, depending on how they answer, students can move both forward and backward in their progression along the path.

Within inference models, you can find three different models they are built upon: content, learner and instructional.

- Content Model

This model houses the computer's understanding of what needs to be learned, for example, the main topics that are covered in a course. The system understands

Национальная ассоциация ученых (НАУ) # 6(33), 2017 how the topics are linked and is thus able to help scaffold or support a student who is learning each topic. The I system also understands each possible progression, asi sisting in the creation of the complex map of topics and I pathways.

- Learner Model

This model is the computer's understanding of what the user knows about the subject matter being learned. When a student begins a new course, they will complete a task so the system can assess their grasp of basic concepts to get a clear picture of their level of proficiency. As the student continues through the course, the system updates the model of student understanding.

- Instructional Model

This model decides how to bring the previous two models together, determining which content to show i next based on the student's current level of understand-i ing. As not all students learn the same, they will not : start a course in the same place either. Some students ; will already have a firm grasp on basic concepts and can move on to more advanced material, while others ; will need to master the basics before progressing.

Biological Models

There are a fair few similarities between inference and biological models in terms of assessing student understanding and assigning content based on what they know. However, biological models are more flexible. Instead of having a predetermined map of pathways for students to navigate, the design of biological models allows for quick adaptation to a large expanse of possibilities that may occur in a student's learning path.

However, creating a unique map of any topic for any student is not necessary. The semantic organization along with the process of making mistakes are combined to help the material adapt, making a one-of-a-kind learning experience that is constantly changing.

How Is The Adaptive Model Implemented In A University?

The Adaptive Learning Environments Model is designed to provide instruction that is responsive to student needs and to provide school staff with ongoing professional development and school-based program implementation support to achieve student success. Implementation features the following design elements.

Individualized Progress Plans consist of two components. The first is a highly structured prescriptive component for basic skills mastery. In addition, an exploratory component provides learning opportunities that foster student self-direction and problem-solving ability while fostering social and personal development to enhance student learning success.

A Diagnostic-Prescriptive Monitoring System incorporates a standards-based curriculum and assessment system to ensure student mastery of subject-matter knowledge and learning skills.

A Classroom Instruction-Management System ; provides implementation support that focuses on student self-responsibility and teacher teaming in implementing a coordinated approach to instructional and re- 1 lated service delivery.

]

A Data-Based Professional Development Program provides ongoing training and technical assistance support that is targeted to meet the implementation support needs of the individual staff.

A School-Based Restructuring Process provides school and classroom organizational support and redeployment of school resources and staff expertise to achieve and sustain a high degree of program implementation.

An active Family Involvement Program is targeted to support student learning success.

When a high degree of implementation is achieved, a unique classroom scenario is created. Students can be found working in virtually every area of the classroom, engaging in a variety of learning activities, including participating in small-group instruction, receiving one-to-one tutoring, or engaging in peer-based collaborative activities. Teachers circulate among the students, instructing and providing corrective feedback.

Instruction is based on diagnostic test results and informal assessments by the teacher. Every student is expected to make steady progress in meeting the cur-ricular standards. Learning tasks are broken down into incremental steps, providing frequent opportunities for evaluation.

The importance of the study due to the follow- ] ing: 1

1 - It benefits the designers and developers of ed- 1 ucational programs in the Ministry of Education and Ministry of Higher Education; adoption of a draft mo- ] bile education, supports the teaching and learning of the subject, with the launch of the software and applications in e-shops to help learning and teaching lectures ] in the Russian Federation.

2 - Encourage students to enable mobile learning i devices (smartphones and tablets) they own; download applications that are available in electronic stores that support the teaching and learning of the subject, and to : use them at any time and in any place where the available wireless networks (Wi-Fi, 3G, 4G).

3 - Assist in the transformation of traditional teaching and learning in the local community to mobile teaching and learning that will allow teachers and leaders in the field of school education, to get a real education, and helps them to get access to mobile learning world, and evaluate the performance of their students light in the extent that they benefit from the advantages thereof.

4 - To help students to get what they need infor- i mation in the learning process and the acquisition of

another language (English) skills; at any time and in any place, as is the case in developed countries.

5 - Helping families in the provision of educational materials for their children through their mobile devices (smartphones, tablets), and therefore can support teaching and learning at any time and in any place.

RELATED WORK

We provide a brief overview of state-of-the-art approaches [3] for content recommendation systems as well as MTL for personalization.

Generalized Linear Models: One simple way to build a large scale content recommendation system is to use generalized linear models to predict responses (e.g., clicks, ratings and etc.) for each user-item feature vector. Recent work [34] demonstrated how to achieve it in an industrial scale. Similar models are used in online advertising as well [2]. Note that, as these models induce global optimization problems, special learning algorithms like ADMM, with sophisticated communication schemes are proposed to solve them.

Tree Boosting: GBDT [13] has been proven effective in many machine learning applications. Together with LambdaMART, tree boosting methods show state-of-the-art performance on many LtR and recommendation tasks [6]. Recently, methods [11] are developed to learn global tree models on large-scale datasets.

Matrix Factorization: MF-based models are widely used for recommendation [19] and assume that there exists a latent vector associated with each user and each item. Also, user and item bias terms are exploited in the optimization [15]. Regression-based LFM [1] and Factorization Machines [28] further extended MF to incorporate arbitrary user and item features. In most cases, gradient descent techniques or alternating least squares (ALS) can be applied to solve the optimization problems. Recent work [7] proposed a customized matrix factorization objective for improved recommendations.

Personalization/Multi-Task Learning: ffiim-portance of a personalized application extends to many □ ends such as social networks [20] and ranking system [29] while social network has also been incorporated in personalization [9]. It is necessary for suggesting relevant/interesting results for users and making distributed computation efficient in mobile side, such as Distributed Matrix Factorization [16]. MTL algorithms have been extensively studied to tackle the problem of personalization in both information retrieval (IR) and rec-ommender systems (RecSys). Here, we highlight a few representative papers. In IR, MTL-style linear ranking model [32] (including feature-hashing methods [33]) and tree-based boosting ranking models [10]) have been proposed to adapt global ranking models to user specific ones. In RecSys, early work [26] formulated a MTL based solution for neighborhood-based collaborative □ filtering methods. Recommending items to users based on expert opinions has also been explored [4]. However, most MTL algorithms need to solve global optimization problems. Our propose framework differs from MTL in that personal models can be optimized independently from a global objective function. Also, the

generic framework accepts different objective functions. Curriculum Learning [5] is proposed to solve progressively harder problems, supplying the training examples in a meaningful order may actually lead to improved performance and better convergence. In comparison to Curriculum Learning, our proposed framework focus on efficient learning for personal recommendation models without reordering.

Conclusion

Adaptive education approaches to improve student learning outcomes has been noted by researchers and practitioners as a promising alternative approach for accommodating the diverse learning needs of individual students, including those with exceptional talents and those with special needs. Implementing adaptive education strategies as an alternative approach to improving student outcomes can be traced as a part of the progressive education movement. Changes in the conceptualization of individual differences and the growing research base in developmental and cognitive psychology have resulted in increasing attention to individual differences in how learning takes place and what influences learning. Individual differences in learning are no longer considered static, but capable of modification either before the instructional process begins or as a part of the process.

Reference

[1] O. Park, and J. Lee. Adaptive Instructional Systems. In Jonassen, D.H. (ed.) Handbook of Research on Educational Communications and Technology (2nd ed), AECT, 2004.

[2] P. Brusilovsky. Developing Adaptive Educational Hypermedia Systems: from Design Models to Authoring Tools. In Murray T., Blessing S., & Ainsworth S. (Eds.), Authoring Tools for Advanced Technology Learning Environment, Dordrecht: Kluwer Academic Publishers, 2003, p. 377-410.

[3] Xavier Amatriain and Deepak Agarwal. 2016. Tutorial: Lessons Learned from Building Real-life Recommender Systems. In Proceedings of the 10th ACM Conference on Recommender Systems (RecSys '16). ACM, New York, NY, USA, 433-433. DOI:hSp://dx.doi.org/10.1145/2959100.2959194

[4] XianXing Zhang, Yitong Zhou, Yiming Ma, Bee-Chung Chen, Liang Zhang, and Deepak Agarwal. GLMix: Generalized Linear Mixed Models For LargeScale Response Prediction. In Proceedings of KDD 2016. 363-372. DOI:hSp: //dx.doi.org/10.1145/2939672.2939684

[5] Deepak Agarwal, Bo Long, Jonathan Traupman, Doris Xin, and Liang Zhang. LASER: A Scalable Response Prediction Platform for Online Advertising. In Proceedings of WSDM 2014. 173-182. DOI:hSp://dx.doi.org/10.1145/2556195. 2556252

[6] Jerome H. Friedman. 2000. Greedy Function Approximation: A Gradient Boosting Machine. Annals of Statistics 29 (2000), 1189-1232.

[7] J. BenneS and S. Lanning. 2007. ffie Net/ix Prize. In Proceedings of the KDD Cup Workshop 2007. 3-6.

[8] Yehuda Koren. Factorization Meets the Neighborhood: A Multifaceted Collaborative Filtering

Model. In Proceedings of KDD 2008. 426-434. DOI:hSp: //dx.doi.org/10.1145/1401890.1401944

[9] Simon Funk. 2006. Net/ix Update: Try this at Home. (2006).

[10] Deepak Agarwal and Bee-Chung Chen. Regression-based Latent Factor Models. In Proceedings of KDD 2009. 19-28. D0I:hSp://dx.doi.org/10.1145/1557019. 1557029

[11] Ste,en Rendle. 2012. Factorization Machines with libFM. ACM Trans. Intell. Syst. Technol. 3, 3, Article 57 (May 2012), 22 pages.

[12] Alex Beutel, Ed H. Chi, Zhiyuan Cheng, Hubert Pham, and John Anderson. 2017. Beyond Globally Optimal: Focused Learning for Improved Recommendations. In Proceedings of the 26th International Conference on World Wide Web (WWW '17).

[13] Roy Levin, Hassan Abassi, and Uzi Cohen. 2016. Guided Walk: A Scalable Recommendation Algorithm for Complex Heterogeneous Social Networks. In Proceedings of the 10th ACM Conference on Recommender Systems (RecSys '16). ACM, New York, NY, USA, 293-300. D0I:hSp://dx.doi.org/10.1145/2959100. 2959143

[14] Ste,en Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt- ffiieme. 2009. BPR: Bayesian Personalized Ranking from Implicit Feedback. In Proceedings of the Twenty-Fi{h Conference on Uncertainty in ArtiDcial Intelligence (UAI '09). AUAI Press, Arlington, Virginia, United States, 452-461. hSp://dl.acm.org/citation.cfm?id=1795114.1795167

[15] Allison J.B. Chaney, David M. Blei, and Tina Eliassi-Rad. 2015. A Probabilistic Model for Using Social Networks in Personalized Item Recommendation. In Proceedings of the 9th ACM Conference on Recommender Systems (RecSys '15). ACM, New York, NY, USA, 43-50. D0I:hSp://dx.doi.org/10.1145/2792838.2800193

[16] Bikash Joshi, Franck Iutzeler, and Massih-Reza Amini. 2016. Asynchronous Distributed Matrix Factorization with Similar User and Item Based Regularization. In Proceedings of the 10th ACM Conference on Recommender Systems (RecSys '16). ACM, New York, NY, USA, 75-78. D0I:hSp://dx.doi.org/10.1145/2959100. 2959161

[17] Hongning Wang, Xiaodong He, Ming-Wei Chang, Yang Song, Ryen W. White, and Wei Chu. Personalized Ranking Model Adaptation for Web Search. In Proceedings of SIGIR 2013. 323-332. DOI :hSp://dx.doi.org/10.1145/2484028.2484068

[18] Kilian Weinberger, Anirban Dasgupta, John Langford, Alex Smola, and Josh ASenberg. Feature Hashing for Large Scale Multitask Learning. In Proceedings of ICML 2009. 1113-1120. D0I:hSp://dx.doi.org/10.1145/1553374.1553516

[19] Olivier Chapelle, PannagadaSa Shivaswamy, Srinivas Vadrevu, Kilian Weinberger, Ya Zhang, and Belle Tseng. 2011. Boosted multi-task learning. Machine Learning 85, 1 (2011), 149-173.

[20] Xia Ning and George Karypis. 2010. Multitask learning for recommender systems. Journal of Machine Learning Research 13 (2010), 269-284.

[21] Xavier Amatriain, Neal Lathia, Josep M. Pujol, Haewoon Kwak, and Nuria Oliver. 2009. Œe Wisdom of the Few: A Collaborative Filtering Approach Based on Expert Opinions from the Web. In Proceedings of the 32Nd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '09). ACM, New York,

NY, USA, 532-539.

D0I:hSp://dx.doi.org/10.1145/1571941.1572033

[22] Yoshua Bengio, Jer' ome Louradour, Ronan Collobert, and Jason Weston. 2009. " Curriculum Learning. In Proceedings of the 26th Annual International Conference on Machine Learning (ICML '09). ACM, New York, NY, USA, 41-48. DOI:hSp: //dx.doi.org/10.1145/1553374.1553380

UDC 681 5 9 7558

_DEVELOPMENT OF SHIP COURSE STABILIZATION SYSTEM_

Satybaldina Dana Karimtayevna

(Candidate of Engineering Sciences, Associate Professor Eurasian National University named after L.N. Gumilyev,

Astana, Kazakhstan) Zekenova Gulsanat Ziyashovna (Master's Degree student, Eurasian National University named after L.N. Gumilyev,

Astana, Kazakhstan)

_РАЗРАБОТКА СИСТЕМЫ СТАБИЛИЗАЦИИ МОРСКОГО СУДНА ПО КУРСУ_

Сатыбалдина Дана Каримтаевна

(кандидат технических наук, доцент, Евразийский национальный университет им. Л.Н. Гумилева,

г. Астана, Казахстан) Зекенова Гульсанат Зияшовна

(магистрант,

Евразийский национальный университет им. Л.Н. Гумилева,

г. Астана, Казахстан)

ABSTRACT

The development of ship course stabilization (autopilot) is addressed below. A mathematical model of a control system and a description of its elements are obtained. The results of modeling the ship's stabilization system are presented by using PD and PID regulators.

АННОТАЦИЯ

Рассматривается разработка системы стабилизации морского судна по курсу (авторулевого). Получены математическая модель систему управления и описание входящих в нее элементов. Представлены результаты моделирования системы стабилизации морского судна с применением ПД- и ПИД-регуляторов.

Keywords: control systems, ship stabilization systems, regulators, transient characteristics

Ключевые слова: системы управления, системы стабилизации судна, регуляторы, переходные характеристики

Guiding the ship from the port of departure to the port of destination refers to complex management tasks. Its peculiarities are the complexity of the ship as an object of control and the diversity of the influence of the external environment on it; the need to process a large amount of data, both from internal and external sources of information; complexity of navigation equipment and power means; limited time for decisionmaking and a number of other circumstances.

At the present stage, through electronic means, the problem of guiding the vessel from the initial point to the final one in accordance with the planned plan is usually solved [8]. It determines the route and time of arrival at its intermediate and final points. The fulfillment of the plan is to maintain a correspondence between the ship's kinematic parameters and the time functions

specified by the plan. In this control process, the driving along the route and the speed regulation are distinguished.

The first task is automatically solved by the onboard systems of driving on route, which have been officially named Track Control Systems (TCS). The TCS control the course and lateral deviation from the specified track line, reducing the latter to a zero value. In order to change or maintain the constant of the first coordinate only, the Heading Control System (HCS) is used. Traditionally it is called autopilot (AP).

On ships of the world fleet, many types of AP are exploited [1, 2, 6]. Depending on the element base, they are divided into electromechanical, electronic analog, electronic digital. The drawbacks of the traditional electromechanical AP are: obsolete element base, low