IAd-learning: A new e-learning platform by José Antonio Omedes at Big Data Spain 2015

Purpose of the talk: Describing the use of Machine Learning and Big Data Techniques to improve the performance of elearning students. Presenting an existing case of an elearning platform (iAdLearn¡ng) and the technology used behind the scenes, to make adaptive/high performance elearning a reality.

Session presented at Big Data Spain 2015 Conference
15th Oct 2015
Kinépolis Madrid
http://www.bigdataspain.org
Event promoted by: http://www.paradigmatecnologico.com
Abstract: http://www.bigdataspain.org/program/thu/slot-17.html

Transcript

2. iAdLearning Next Generation E-Learning Oct 2015. Version 1.0 jose.omedes@itoptraining.com
3. Copyright iTopTraining. All rights reserved. Topics Page  2 1. Introduction and iTopTraining 2. Where are we coming from? What are we trying to achieve? 3. iAdLearning architecture
4. Copyright iTopTraining. All rights reserved. Page  3 1. Introduction and iTopTraining
5. Copyright iTopTraining. All rights reserved. iTopTraining  Privately held company  Based in Madrid (we have presence in Guadalajara – México – )  Founded in July 2013  Continuation from a previous company  Two main business lines:  eLearning platforms on the cloud (SaaS)  eLearning innovation  Research and Development of innovative e-Learning software solutions  ~50% people works on Research and Development Page  4
6. Copyright iTopTraining. All rights reserved. iTopTraining Technology Landscape Page  5 Day to Day Operations (SaaS eLearning Platforms) Cloud Technologies (AWS, Azure) PHP Javascript Apache MySQL (Aurora) MemCache Moodle Innovative eLearning Products (includes iAdLearning) Cloud Technologies (AWS, Azure) AngularJS NodeJS, D3js Java, Scala NGINX Cassandra, MongoDB Neo4j, Redis, ElasticSearch Apache Spark Machine Learning, NLP
7. Copyright iTopTraining. All rights reserved. Page  6 2. Where are we coming from and what are we trying to achieve?
8. Copyright iTopTraining. All rights reserved. Change of Learning Paradigms (*) Page  7 Linear Learning Static Training Contents Free Learning Content Adapts to Users (*) We are focusing on those factors relevant for our discussion but there are other changes such us the important irruption of mobile devices on eLearning “Mimics” Internet Navigation Learn “by interest” Every user is different and has different learning needs even in the context of the same course Instructor Lead Learning eLearning Users access to training courses whenever they want to and wherever they are
9. Copyright iTopTraining. All rights reserved. Change of paradigm brings new challenges … Page  8 e-Learning Volume Content Transformation for eLearning 1 2 • eLearning market CAGR ~20% (depends on studies) • +38% of company training is delivered online • MooCs are consolidating DELIVERY ANALYSIS of INFORMATION (is this a challenge?) • Very man intensive task • Very time consuming task COST End User Learning Experience 3 • Users are “left alone” … QUALITY INDIVIDUAL NEEDS
10. Copyright iTopTraining. All rights reserved. Change of paradigm brings new opportunities (1) … Page  9 Improve eLearning Experience Using all information gathered by eLearning platforms MACHINE LEARNING BIG DATA • Improving the way contents are visualized and navigated • Guiding users to discover the content which is relevant for them • Providing automatic feedback to trainers about the quality of the content materials and about the user experience when using those contents
11. Copyright iTopTraining. All rights reserved. Change of paradigm brings new opportunities (2) … Page  10 Facilitate migration of contents to eLearning • Generating eLearning contents in a format that facilitates free navigation and “learning by interest” from non-native eLearning formats (Word, PDF, Powerpoint) • Providing a cost effective solution for content transformation NATURAL LANGUAGE PROCESSING MACHINE LEARNING Processing non native eLearning contents
12. Copyright iTopTraining. All rights reserved. Page  11 Wasn’t this presentation supposed to be technical?
13. Copyright iTopTraining. All rights reserved. Key Take-aways  The way people learn is changing  New formats (eLearning growth)  New ways of using existing formats such as e-Learning (free navigation, learn by interest, etc.)  Growth of e-Learning provides us with an important amount of information that properly analyzed may dramatically improve the user’s eLearning Experience  Requires technology  Requires proper algorithms  Content Transformation is key to help trainers and training companies move into e-Learning  Benefit from all the information analysis based on e-Learning data  Follow the market trends (continue in the market) Page  12
14. Copyright iTopTraining. All rights reserved. What is iAdLearning? Page  13 iAdLearning is a software that makes the e-Learning platforms adapt to the individual needs of the students -adaptive e-Learning- providing a unique personal e-learning experience throughout the course materials while maximizing the students achievements by making intelligent study suggestions based on previous learning experiences
15. Copyright iTopTraining. All rights reserved. iAdLearning Principles Page  14 ADAPTIVE Each student has a complete different experience throughout the course content based on his/her background Students can navigate through the entire training materials in a non lineal fashion based on their needs or interests NON LINEAR LEARNING MAKE USE OF EVERYBODY’s EXPERIENCE The system uses previous learning experiences to recommend customized learning paths throughout the content EXISTING CONTENT IS EASY TO IMPORT iAdLearning allows to easily import existing non e-learning native training materials (Word, PDF, PPT)
16. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  15 1 Importing documents, analyzing them and establishing semantic relationships among the document individual content components creating, as a result, a user navigable graph (knowledge network) 2 Allowing free navigation through the knowledge network not only based on the original content structure but also on the user preferences and interests 3 Discovering successful navigation patterns that maximize user performance through the course materials in a way those can be suggested to students during their learning experience IMPORT VISUALIZE RECOMMEND
17. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  16 Set of Documents (Word, PPT, PDF) Semantic Analysis Semantic Analysis ACE 1 1.2.2 2 Course Graph Knowledge Net 1.1 1.2 1.2.1 • A set of documents belonging to an e-Learning course is imported, analyzed and broken down into semantically relevant fragments called ACEs (Atomic Content Elements). ACEs represent course fragments that can be individually studied and understood • iAdLearning creates a graph representing the relationships between the different ACEs: • Structural Relationships (blue lines): represent connections related to the course structure as initially established by documents authors • Semantic Relationships (red lines): represent connections created due to the similarity of the contents being described by the connected ACEs IMPORT
18. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  17 IMPORT Section 1 Section 1 Section 1.1 Section 1.1 Section 1.2 Section 1.2 Section 2 Section 2 Section 2.1 Section 2.1 Section 2.2 Section 2.2 Section 2.2.1 Section 2.2.1 Section 2.2.2 Section 2.2.2 Section 7 Section 7 … Section 7.1 Section 7.1 Section 7.1.1 Section 7.1.1 Section 7.1.2 Section 7.1.2 ACE 1 ACE 1 ACE 2 ACE 2 ACE 3 ACE 3 ACE 4 ACE 4 ACE 5 ACE 5 ACE 6 ACE 6 ACE 7 ACE 7 0.95 0.72 ACE 1 ACE 1 ACE 2 ACE 2 ACE 3 ACE 3 ACE 4 ACE 4 ACE 5 ACE 5 ACE 6 ACE 6 ACE 7 ACE 7 Fragmentation Analysis of Relationships
19. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  18 IMPORT STRUCTURE OF THE DOCUMENT AS DISCOVERED BY iAdLearning
20. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  19 ACE 1 1.2.2 2 Course Graph Knowledge Net 1.1 1.2 1.2.1 VISUALIZE EXAM { 1, 1.1, 1.2.1, 2, 1.2.2, 1.2, EXAM} { 1, 1.2, 1.2.1, 1.1, 1.2.2, 2, EXAM} USER 1 USER 2 User Navigation Patterns • When accessing the e-Learning course contents, users navigate through the graph (*) either following the initial course structure (blue connections) or jumping into other related content elements according to their preferences/interests (red connections) • User navigation patterns (visited ACEs and their sequence) are recorded together with the user evaluation results (exam results), in a way they can be further analyzed (*) On top of the graph navigation, iAdLearning offers other navigation views such a tree-view (**) Contents created with iAdLearning can be visualized in Moodle through a plugin
21. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  20 VISUALIZE STRUCTURAL RELATIONSHIPS SEMANTIC RELATIONSHIPS
22. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  21 { 1, 1.1, 1.2.1, 2, 1.2.2, 1.2} { 1, 1.2, 1.2.1, 1.1, 1.2.2, 2} USER 1 USER n User Navigation Patterns … Behavioral Analysis Behavioral Analysis • Navigation patterns are analyzed using advanced machine learning techniques • As a result of the analysis, users are grouped according to their course navigation pattern and a set of relevant variables (age, studies, knowledge of the topics covered by the course, etc.). iAdLearning establishes recommended navigation paths for each group of users based on the recorded course performance. … Recommended Navigation Paths { 1, 1.1, 1.2, 1.2.1, 1.2.2, 2} { 1, 1.1, 1.2.1, 2, 1.2.2, 1.2} age, studies, previous knowledge, evaluation mark, etc. RECOMMEND
23. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  22 NEW USER Classification Classification Recommendation Recommendation 1 1.2.2 2 1.1 1.2 1.2.1 1 2 3 4 • When a new user comes into the e-Learning platform, it is classified according to the criteria established by the behavioral analysis already performed on the navigation patterns of preceding users • Based on the user classification, iAdLearning suggests an individual recommended navigation path targeted to maximize the user performance on the course RECOMMEND
24. Copyright iTopTraining. All rights reserved. How does iAdLearning work? Page  23 STRUCTURAL RELATIONSHIPS SEMANTIC RELATIONSHIPS RECOMMENDED NEXT NODE
25. Copyright iTopTraining. All rights reserved. Page  24 3. iAdLearning Architecture
26. Copyright iTopTraining. All rights reserved. Architecture Principles Page  25 All user actions are being logged for further analysis (logging intensive application) Separate Application and “Logging” Information Flows Individual Functions Scalability The different components of the application must be individually scalable according to the real needs The different parts of the application must be implemented in different logical entities that communicate through APIs Function Separation via APIs Redundancy / Resiliency All the system must be redundant Cloud Based / Cloud Provider Independent The system must reside on the cloud The system must be independent of the cloud provider
27. Copyright iTopTraining. All rights reserved. iAdLearning Architecture (version 1) Page  26
28. Copyright iTopTraining. All rights reserved. iAdLearning Architecture Page  27 IMPORT • Traffic coming from the Front End is redirected by NGINX into the Web Server • The Web Server sends the documents to import to a pending jobs queue handled by RabbitMQ. • Whenever there is processing power available, they are analyzed, fragmented and the “knowledge network” (graph) gets generated. • Results of the import process are persisted into MongoDB, Neo4j, Elastic Search and Cloud Storage.
29. Copyright iTopTraining. All rights reserved. Page  28 •Regular application flow is directed through NGINX into the Web Server •The Web Server gets information to visualize from MongoDB / Neo4j •When required, the front end component pulls content from the cloud storage •Information persisted in MongoDB / Neo4j corresponds to: • Training contents • Recommend paths through the content materials iAdLearning Architecture VISUALIZE
30. Copyright iTopTraining. All rights reserved. Page  29 •User “action events” come to NGNIX which redirects the traffic to the logging server. •The logging server persists the user action events into Cassandra. •Logs stored in Cassandra are periodically analyzed by a batch process which runs the analysis in Apache Spark. •Results of the Analysis (Recommended Paths through the content) are persisted into MongoDB. iAdLearning Architecture RECOMMEND
31. Copyright iTopTraining. All rights reserved. Page  30 iAdLearning Architecture - Technologies FRONT PROGRAMMING LANGUAJES DATABASES SEARCH ENGINES ANALYTICS CLOUD PROVIDERS QUEUEING
32. Copyright iTopTraining. All rights reserved. iAdLearning Architecture – Technologies Page  31 Members of DataStax Startup Program DataStax Enterprise 4.8 (DSE) Spark 1.4 Cassandra 2.1 OpsCenter DevCenter Spark Cassandra Drivers
33. Copyright iTopTraining. All rights reserved. iAdLearning Architecture – Thinking points Page  32 We are permanent newbies … We live in a world of complex technologies continuously evolving Solutions need to be simplified There is always a new technology/system that may play a role in your solution Management of solutions and cost need to be part of the equation We miss “relational” … Relational databases are great !!! Sometimes you cannot use relational databases but they are still great !!!
34. Copyright iTopTraining. All rights reserved. iAdLearning Architecture – Thinking points Page  33 Security is important from day 1 Don’t treat security as a marginal element in your system until you go into production Automate your deployments How many times are you going to install the same type of node? We use Chef Networking is important from day 1 Don’t treat networking as a marginal element in your system until you go into production
35. Copyright iTopTraining. All rights reserved. iAdLearning Architecture – Thinking points Page  34 Reuse and go to the source There are tons of libraries out there that may be used in your development Go to the source … minimize the library over library effect ...
36. Copyright iTopTraining. All rights reserved. iAdLearning Architecture (version 2) Page  35 1 Simplify the way the front end is delivered to the user by using a Content Delivery Network Content is static and the dynamic part (JavaScript) runs at the user browser 2 Reduce number of databases A graph oriented database is not needed for the time being 3 Reduce the number of software components Simplify the way importing of documents is performed (queuing part) 4 Introduce Deployment Tools and Continuous Integration Reduce the deployment of new software cycle
37. Copyright iTopTraining. All rights reserved. iAdLearning Architecture (version 2) Page  36
38. Copyright iTopTraining. All rights reserved. Page  37 Thank you very much iTopTraining www.itoptraining.com jose.omedes@itoptraining.com