In an attempt to build bridges, Preferred.AI crossed the bridge and conducted a data science meetup at Kuala Lumpur (KL), Malaysia on 16 March 2019. We’re grateful to data science communities Big Data Malaysia and DataViz My for organizing the event, as well as to SEEK Asia for hosting the venue and LEAD for sponsoring the catering. Utmost thanks go to the nearly hundred keen attendees whom we learnt much from through the questions and interactions.
For us, it was a great chance to share an overview of our activities in extracting preference signals and modeling them in recommender systems. The meetup featured four specific talks on several projects being developed at Preferred.AI. Beginning from how data could be collected, Ween showcased the power of Venom in enabling focused crawling of Web sites. Such data might contain preference signals of various forms. Max spoke about how the subjectivity of text corpora might affect the word embeddings derived from them, and SentiVec that factored in a sentiment lexicon would perform more effectively on preference-related text analysis such as sentiment detection. These preference signals could be incorporated into a learning model for recommendation. For this, Aghiles introduced Cornac, a library of recommendation algorithms capable of leveraging auxiliary information. Finally, Hady touched on how IBPR, a recommendation algorithm compatible with several indexing schemes, would retrieve personalized recommendations more efficiently.
In the ensuing panel session, the audience brought up many good points about learning data science (how to make the math easier to learn) as well as data science in practice (the scale of data could be much bigger than benchmark datasets). In all, it was an eye-opening experience and a good learning opportunity for us as well.
To share the memories, we compiled a slideshow of the pictures of the trip. It was the very definition of a worthwhile trip: seen much, learnt more, made friends, would return.
The materials for all the talks can be found here.