Grant: Discovery Grants Program - Individual

Recommender systems have become useful tools for helping users deal with information overload problems. However, most of these systems today are built by the content or product providers (e.g., Amazon, Netflix) to recommend their own content or product to users. In this proposed research, we will design an information harvesting and recommendation system to facilitate individual users' online information-seeking and decision-making process. We plan to implement it as a cloud-based service. Compared to traditional systems, developing a user responsive recommender system is more complex and presents unique challenges, including: 1) finding data that is useful for building recommender systems, for example, social media data is rich in users' opinions on various items, nonetheless, identifying the most suitable subset for a user's need may be challenging; 2) integrating multiple data sources and transferring learned knowledge from these sources for effective recommendation; 3) identifying user decision criteria (beyond commonly considered predicted ratings) when choosing an item, and building a multi-criteria recommender system; 4) continuously monitoring user intent, need and emotion, and recommending items according to these psychological factors. Our objective is to build a personal recommendation service that can identify the right data sources for the target user and the item to be recommended, and make recommendations based on data collected from these sources as well as user decision criteria and intent. To begin with, we will select one or two use cases, such as a movie recommender service, an API (Application Programming Interface) recommender service, or a service recommending research papers to PhD students. We will collect data from multiple sources containing ratings, reviews and item features, including social media sites. We will then propose an effective integration approach. From sources with user interaction data, we will identify user decision criteria as well as intent and motive behind the information need. A multi-criteria recommender service will then be implemented. To the best of our knowledge, once successfully implemented, our personal recommender service will be the first such service encompassing all the mentioned characteristics. It could help users complete their daily tasks more efficiently. The consideration of psychological factors could help recommending items that are beneficial to a user's well-being. Our research agenda aligns with Canada's national strategy on AI (artificial intelligence) since we are exploring the use of different machine learning (a type of AI algorithms) models in building our recommender service. We will also train a number of HQPs to gain a deep understanding on topics such as recommender systems, social network analysis, machine learning and multi-criteria decision making. In addition, they will gain hands-on experiences in building cloud-based applications and personal assistant service.