Implicit Feedback Based Personalized Recommender System Using Collaborative Denoising Autoencoder

There are plethora of applications today that rely on hyper-curated content personalized according to user’s interests. It is usually difficult to get lots of users to explicitly specify their interests continuously and to ensure a rich user experience. To overcome this, Implicit feedback based on the actions performed by users tend to be very useful. In order to facilitate premium services to users, big chunk of applications today rely on Top-N recommendations so that users don’t have to spend lot of time searching for the needed contents or items. This project is done via Mitacs Accelerate Funding in association with Pintellect. Pintellect is Enterprise Social Software that gives employees access to the thoughts and ideas of the organization’s influencer set by encouraging them to share links to the internal files or external resources such as books, TED talks, podcasts, articles, etc. In this project, following are the high-level sub-objectives that are studied and implemented: 1. Requirement Identification 2. Feature Identification 3. Data Representation, Gathering and Cleaning 4. Design of Framework for Training and Evaluation 5. Model Architecture and various recommendation algorithms 6. Collaborative Denoising Autoencoder (CDAE) 7. Algorithm evaluation and training 8. Performance evaluation

Faculty Supervisor:

Craig Scratchley

Student:

Shams Narsinh

Partner:

HOP Operating Company Ltd

Discipline:

Engineering

Sector:

Information and communications technologies

University:

Simon Fraser University