Research Question

Choose one commercial recommender and describe how you think it works (content-based, collaborative filtering, etc). Does the technique deliver a good experience or are the recommendations off-target?

My Choice - YouTube

YouTube

Youtube is the world’s most popular streaming video service. Every minute, 400 new videos are uploaded to the site. Its recommendation system is desgined to recommend personalized content to over a billion users. In their research paper, “Deep Neural Networks for YouTube Recommendations” by Paul Covington, Jay Adams, Emre Sargin. YouTube had to solve problems with Scale, the massive size of their video corpus and their users. Freshness, given the number of videos loaded every minute, recommending new content can be a daunting challenge. And, Noise, the sparseneess and inaccuracy of user historical data makes it difficult for them to make recommendations without a lot of wrong choices.

As in the diagram below, Youtube uses a two neural networks, one for candidate generation and one for ranking of the videos. The candidate generation network takes user input, such as an activity log, IDs of watched videos, and user demos to output more precise recommendations of videos. The video ranking engine scores the features of the videos. “The ranking network accomplishes this task by assigning a score to each video according to a desired objective function using a rich set of features describing the video and user. The highest scoring videos are presented to the user, ranked by their score.”

Both of these systems are fed into a Deep Learning Neural Network.

Diagram

Youtube’s average visitors spend 40 minutes a day on the site. It has over 30 million active daily users. I would say that this is proof that Youtube’s recommendation system is effective at keeping users interested in the site.

Attacks on Recommender Systems

Similar Example

“Bot” influence on the movie review website, Rotten Tomatoes, has stirred controversy. In 2017. Star Wars: The Last Jedi received a 93% score from critics but only half of that from the fans on the site. It was suspected that bots were uses to give the film negative ratings in order to discredit its more diverse cast and stronger female roles.

My recommendation to a site like Rotten Tomatoes would be to requrire each reviewer of a film to answer a series of random questions about the film before they can post their review. The questions would be varied from user to user. The questions would ensure that the reviewer actually saw the movie in question before they could post a review. For example, if a reviewer lives in a country where the movie was not released, then that review would not be posted.

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