Friday Hacks #154, April 13
Posted on by Julius
Hi hackers! Nowadays, one of the most important pieces of our modern web apps, be it in e-commerce, social media, or even search engines is a recommender system – and this is what our speakers this week from Nurture.AI and Shopee will be speaking about this week.
Date/Time: Friday, April 13 at 6:30pm
Venue: Seminar Room 1, Town Plaza, University Town (UT22-02-04A)
Free pizza is served before the talks.
A Deep Learning-based Recommender System for AI Papers
Nurture.AI is a platform for discovering, discussing and reviewing AI-related arXiv pre-prints, and crowdsourcing code implementations for these pre-prints. Although users discover new papers through their personalised Twitter newsfeed cleaned up on the platform to show only AI-related tweets, we wanted to build a recommender system for papers similar to any particular paper on the platform. We use GloVe word embeddings on a corpus of 10,000 AI papers taken from arXiv and used LDA to assign topics to each of the 10,000 abstracts, followed by a single layer LSTM on the 10,000 abstracts to obtain a feature representation for each abstract in terms of the LSTM hidden layer units. The recommendation is done by performing a k-nn search on the LSTM hidden layer units to find similar abstracts, and hence similar papers.
Yap Jia Qing – Jia Qing is the Founder and CEO of Nurture.AI. He was the first in Asia Pacific to be certified as an NVIDIA Deep Learning Institute instructor, and has taught AI at places like General Assembly, MAGIC, ADAX and SGInnovate. Through Nurture.AI, Jia Qing also started the AI Saturdays movement which is now in more than 100 cities around the world, as well as the Global NIPS Paper Implementation Challenge which saw more than a 1000 participants and 46 successful code implementations of NIPS 2017 papers
Il Shan Ng – Il Shan graduated from Carleton College with a degree in Mathematical Statistics and Economics, where he took the opportunity to learn about topics like the theory of computation, stochastic processes and artificial intelligence. Back in Singapore, he is a public servant who volunteers to teach reinforcement learning, deep learning and computer vision at AI Saturdays, and played a key role in building the natural language recommender system for Nurture.AI.
Towards Flexible Granularity Intention Recognition
Recommender system generally involves recognizing users’ intention based on their navigation path. In the context of e-commerce, intention is usually session-based and needs to be recognized in a real-time fashion. In this talk, we will share how Shopee generates the intention candidate set with flexible granularity by clustering techniques, and how we map users’ behaviors to possibly multiple candidates through a 2-layer classification system.
Zhenwei is Ph.D graduate from NUS. He has spent his last two years working on Shopee recommendation system, during which he practiced various model building techniques to solve relevance-related problems. He is keen on building intelligent systems and dealing with system dynamics.
The HANGAR by NUS Enterprise — the campus hub for entrepreneurs.