Smart Course Browser

PUBLISHED ON MAY 21, 2018 — MACHINE LEARNING

Abstract—Choosing a course at the university is a challenge to students that is overlooked by the system. The student to advisor ratio is big and the student needs to spend months researching a course without finding sufficient data. With so few resources, the student is left with the few lines of description from the course website and the word of mouth, which have such a big risk. Education shouldn’t be relying on risky actions. In this project, I address this issue and suggest a platform that can help in fixing it. I propose a recommender system that uses the new techniques that are used in different fields and applies them to education. The final product is a complete platform with a search engine, recommender algorithms, and a decision-making tool. The value in the methods presented here is that, unlike other educational recommendation systems, it tries to help the student decide with a step-by-step approach rather than doing all the prediction automatically.

Data Collection

The data used for this project was obtained from the Software Product Management Specialization Massive Open Online Course (MOOC) that is hosted on Coursera. Data include personal data, such as grade books, ratings, performance, and assessment. This tabular data comes in Comma Delimited Value (CSV) files. The data also included notes and transcripts from the lectures. User names were anonymized by Coursera to protect the privacy of the students. So each user has a specific encrypted user ID. Those user ID’s are not consistent across the sections, for example, students’ performance cannot be linked to their feedback. Five courses are required, plus a capstone project, to finish the specialization. Table I shows the names of the courses that were used for this project.

Courses
Introduction to Software Product Management
Software Processes and Agile Practices
Client Needs and Software Requirements
Agile Planning for Software Products
Reviews & Metrics for Software Improvements

Main parts

  • Apache Solr
  • Word2vec
  • Recommender Systems
    • Item-Similarity
    • Matrix Factorization
  • IBM Watson
    • TradeOff Analytics API

Dependencies

  • SQLAlchemy
  • Pandas
  • Scikit-Learn
  • Gensim
  • NLTK
  • SparseSVD

The system

Results

Sample results of the Word2Vec model
agile manager
methodologies owner
principles leader
Agile delivering
methods decision
Lean seagull
practices vision
methodology evolves
Scrum evolving
lean managers
linear responsibility

Word embeddings

TradeOff Analytics

Our UI