UPCARS is kicked-off

Recommender Systems have become more and more critical for businesses not only to maximize customer satisfaction through personalized experiences and recommendations, but also as a more subtle way of communicating with the customer. A typical online platform has a multitude of products to offer to the customer and identifying what a client needs in a specific moment is crucial. 

There are two paths one can follow to navigate through the clutter of products: information retrieval and information filtering. The latter includes the Recommender Systems (RS), which are intelligent systems designed to recommend to a specific user, in a personalized way, the items (a movie, a book, a job, a restaurant, news articles, etc) that the user would be interested in.

Since the 90s when RS were introduced, several approaches (content-based filtering, item-based filtering, collaborative filtering) have been implemented, however the state-of-the-art acknowledges few issues and challenges not fully fixed by the current solutions. 

For example, it may be difficult to recommend new items to users or to propose recommendations to the new users (known as the cold start problem); also the same user has different expectations and taste depending on his context (e.g. a business lunch or a romantic diner, a movie watched with the kids or alone) and the temporal recommendation, the ability to recommend items in a specific timeframe, is still an open question. In addition, users are more and more concerned about their privacy, the personal data collected and how it is used and processed.

In the same time, there is a recent worldwide enthusiasm towards the intersection of complex mathematical theories such as Algebraic Topology and Differential Geometry with Data Science and Machine Learning. For example, the University of Lausanne have recently released giotto-tda, a python framework for topological data analysis. These domains open new gates to open problems in information filtering and retrieval. 

These reflections led us to think about the next generation of information RS, able to overcome the current challenges and to offer contextualized recommendations while limiting the data collected from the user to the minimum: this was UPCARS genesis: a User Privacy Context Aware Recommender System.

Our research direction consists in combining collaborative filtering with embeddings in non-euclidean space enhanced by topological data science approaches.