Task
Overview
Natural language processing (NLP) is commonly used to build recommendation engines. This assignment involves building reading recommendation engines for students in Australian higher education based on a sample of subject reading lists sourced from public sites.
The two main types of NLP reading recommendations models that are generally applicable to the reading list recommendation problem are:
Content-based filters — use item metadata (description, rating, products features, reviews, tags, genres) to find items like those the user has enjoyed in the past.
Collaborative filtering — Collaborative filtering systems analyse users’ interactions with the items (e.g. through ratings, likes or clicks) to create the recommendations.
Learning outcomes
Understand and apply new data science skills, knowledge, and techniques to solve problems in data science using natural language processing (NLP).
Work-based skills
The ability to automatically build labelled datasets and map text hierarchies using NLP is valuable back-office automation opportunity saving time and increasing accuracy for organisations.
Background
Currently University lectures rely on librarians to do background research on relevant, compliant, and current reading material suitable for course teaching units.
In Europe, it is possible to use existing and past reading list information from different University’s to inform academics, librarians and publishers about comparative reading that might be applicable to any selected topic. This is not possible yet in Australia as no up to date central repository exists for reading list material.
The ability to recommend reading material down to a book and page level and at a journal level would be a useful product for academics, publishers and agencies supporting higher education.
For education providers a recommendation engine will :
Be simpler and more intuitive to use
Reduce research time in material selection
Resolve current process deficiencies
Improve data availability.
For students it will:
Make material more relevant and contemporary
Improve engagement through usage monitoring
Help students complete their studies.
For publishers and software companies it will allow:
Insights based on which of their titles are used, where and how much.
More focused selling of course eBooks to institutions for their student
Better Integration with learning management systems
Tasks
1. Develop two NLP recommendation engines derived on the University’s reading list material (supplied data) and apply the NLP recommenders to one of the following options:
a. Recommend existing course material to similar subjects, or
b. Recommend new reading material to existing subjects, or
c. Provide a complete reading list of existing readings for a new subject.
2. Determine the quality of both NLP recommenders from Task 1 using test and training sets derived from the supplied data.
3. Compare the two NLP recommenders.
For each of the three tasks, a written report is to be provided. The report must show comprehensive thought of your decisions, clearly communicate your ideas, and linked to NLP theory/applications with appropriate references.
Data
The data to be used as the initial starting point for the Tasks is given in the Assessment 2 folder on Learn JCU. A summary of the variables is given in Table 1. The provided data is insufficient to develop NLP recommenders or provide assessments of NLP recommender quality.
The provided data will require:
Creation of a labelled dataset from the source data (ontology), and
Supplementation of the dataset from at least one external resource (API)
