As AI develops and becomes a norm in our every day lives, users are more accepting of features that were previously considered 'spooky' such as product recommendations and online chatbots. Whilst this is a step forwards with regards to users becoming more trusting of artificial intelligence, on the flip side, their expectation is much higher. People are aware that companies are able to use AI to personalise their customer experience, so people will start to become frustrated when their Google Assistant doesn't quite understand them, or their Facebook adverts suggest a product they've already purchased.
When you open your Netflix and you've been through 'who's watching', your home screen is going to be vastly different to your four year old daughters. Thank goodness Netflix has this handy feature to keep Paw Patrol and Peppa Pig separate from Narcos and The Crown, but that's not all the system does. On a basic level, the recommender system learns from your account which type of series or movie you're likely to be interested in based on your previous history, and suggests the most relevant titles. Going into more detail about the shallow and deep latent models Netflix user for their recommender systems at the Deep Learning Summit in Boston this May 24 - 25 is Anoop Deoras. The Lead Researcher leads the algorithmic innovation and productisation of deep learning based recommender system models. He will discuss techniques for embedding discrete user action events into continuous but latent space for building a context aware collaborative filtering model for personalisation and recommendations.
We caught up with Anoop in advance of his presentation and asked a few questions we were keen to learn more about. Anoop explained that as a lead researcher at Netflix, he's particularly interested in doing research on both shallow and deep latent models to build highly personalized collaborative filtering models -- the workhorse of any recommendation they do at Netflix. 'Collaborative filtering is a ML technique where you try to group similar users together and then extrapolate from their consumption patterns to recommend relevant and highly personalized movie and TV shows to members with similar taste. Finding similar users is a hard problem and I spend most of my day trying to understand how to do that best.'
Anoop began his work in deep learning shortly after his undergrad in India where he work working on speech compression. He explained that he became 'interested in speech in general and decided to pursue a PhD in statistical speech recognition from Johns Hopkins University. My advisor was (late) Prof. Fred Jelinek, who was called the father of statistical speech recognition and statistical machine translation. He paved way for data driven ML models for doing automatic speech recognition and machine translation at scale. Working with him, I got even more passionate about machine learning and its application towards making AI complementary to human intelligence and abilities. After finishing up my PhD, I joined Microsoft's speech science division working on Cortana, a virtual personal assistant. While working on Cortana and getting a taste of user focused products and the scope of impact good AI applications can have, I started getting interested in the proactive recommendations and the personalization aspects of it. Netflix being the pioneer in recommendation system, where the main aim there is to maximize joy of Netflix member and entertain them by way of providing highly personalized recommendations, it became my natural choice for my next job. That is how I got deeply involved with recommender systems.'
Anoop answered a few more questions for us:
We aim to maximize joy of our Netflix member while the member is engaged with our service. We believe that by providing relevant and personalized recommendations to the user, we can aim towards maximizing this joy. A recommender system which is unaware of the user state (user's context, user's intent etc), can do personalization only to a certain degree. Thus, the main challenges in personalization is to get the user's state correct and modeled well.
NLP and RecSys are applications of machine learning and more often than not, the underlying machine learning methodologies turn out to be similar. When I was at Microsoft working on virtual personal assistants, our main priority was to get the user intent correct and be good at proactive recommendations (showing you traffic conditions before your flight, re-arranging the tickers in your stock app based on which stocks you like to see most when you wake up etc..). At Netflix too, we strive to get our member's intent correct so that the time it takes to play something the member truly likes, is minimized. Intent detection then becomes the common machine learning problem, applicable to both NLP and RecSys. There are several other commonalities between NLP and RecSys and we tend to borrow ideas from one field and apply it to other. Matrix Factorization, which is a very standard ML model for RecSys, was recently used to learn word embeddings in an NLP application. Deep Learning (DL), on the other hand, is not an application of ML, unlike NLP and RecSys, but it is a ML tool in itself. You can apply DL to do intent detection, for instance.
Recommendation systems are means to an end. We try to build models for recommendations so that we can maximize Netflix member's enjoyment of the selected item while minimizing the time it takes to find them. Enjoyment integrated over time i.e. goodness of the item and the length of view, interaction cost integrated over time i.e. time it takes the member to find something to play, are some of the factors we consider while building our ML/AI models for a positive impact on our 100M+ members.
I am most excited for Reinforcement Learning (RL), a sub field of AI. Applications of RL are abound, but I feel the biggest impact RL will have will not just be in Robotics but also in human computer interaction (dialogue modeling). Human computer interaction will be everywhere, from talking to phones to get suggestions on restaurants or local interests, to talking to machines to get assistance on some medical question. If you have not already, I would recommend 'Alpha Go' movie, which is available to stream on Netflix worldwide, in which RL experts from Google DeepMind train a ML model using RL to win the most complex game of Go against the world champion -- Lee Seedol.
AI and ML are function of the data that we feed into training them. A lot of times, data used to train ML models has inherent biases and those biases are reflected in the model predictions and AI behavior. It is therefore very important for a ML practitioner to be cognizant of the fact that such biases exists in the data and that it is necessary to ensure such biases are addressed, through either explicit de-biasing, data stratification, model adaptations or a combination of them. At Netflix, bias manifests in the form of 'feedback loops' i.e. impression biases inflating popularity of certain item (movie / TV show). Production biases in ML models can cause these feedback loops to be reinforced by the recommendation system. We do research and development in the causal recommendation space specifically to get our recommender models out of this feedback loop caused due to heavy production bias.
Keen to hear more about Netflix's recommender system? Join us in Boston to learn more. Additional confirmed speakers include Daniel Smilkov, Google Brain, Nitin Sharma, PayPal, Bjarke Felbo, MIT Media Lab and many more.