DATA612- Research Discussion One

Instructions

Now that we have covered basic techniques for recommender systems, 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?

Response

The two main commercial recommender systems that i normally interact with are: NETFLIX and AMAZON. In the introduction of myself last week, i mentioned that NETFLIX is the recommender system i least favour. Hence, i choose to explore this system in response to this first instruction

A synopsis of how NETFLIX’s Recommender System works?

I would say that NETFLIX uses a hybrid filltering system. This means that it combines two or more recommender system techniques. Based on readings, the Netflix recommendation engine uses your past activity and returns movies and shows it thinks you will enjoy (Codecademy, Netflix Recommendation Engine, 2019). The engine uses a combination of machine learning, data from users, and algorithms in order to rank order its content and decide what to show users on their account. The machine learning and algorithms used by NETFLIX helps break viewers’ preconceived notions and find shows that they might not have initially chosen. To do this, it looks at nuanced threads within the content, rather than relying on broad genres to make its predictions (Plummer, 2017). This is known as the personalized method where movies are suggested to the users who are most likely to enjoy them based on a metric like major actors or genre.

Machine learning is able to create “smart” platforms because it uses ‘probability’ to discover the likelihood of a user liking a product (Codecademy, Netflix Recommendation Engine, 2019). Below is an example which shows how a probability model, in this case a utility matrix, can be used to predict user preferences.

In the image above, we have three Netflix users: Fatima, Maya, and Leslie. Each user has watched a few movies on Netflix and rated them and each user has movies in the catalog they haven’t watched or rated yet. The user ratings are utility scores which represent the relationship between the movie and the user. In this example, utility scores are represented by the checkmark and X symbols. The checkmarks represent movies that they’ve seen and liked and the Xs represent movies they’ve seen and not liked. The question marks represent movies that they haven’t seen yet. What kind of movies then will the recommender suggest to each user?

It may suggest Fatima watch a who-done-it since she enjoyed the thrilling plot of a horror movie she watched last week. The engine may recommend more light-hearted films to Maya, like a comedy or romance, because she didn’t enjoy the horror film. Because Leslie equally enjoyed a horror movie and a romance movie, the engine will give her a more niche, curated experience, like romantic thrillers. Hence, each user will receive a suggestion that’s personalized for them!

Additionally, NETFLIX seems to go beyond recommending movie titles. They also use recommendation engines to curate the preview images you’re seeing on your feed. If you rated Johnny Depp movies highly, you’re likely to see an image of Captain Jack Sparrow when Netflix recommends Pirates of the Caribbean to you. If you tend to like romance movies, your feed will likely have preview images of scenes with actors embracing one another (Codecademy, Netflix Recommendation Engine, 2019).

Attacks on Recommender Systems

Instructions

Read the article below and consider how to handle attacks on recommender systems. Can you think of a similar example where a collective effort to alter the workings of content recommendations have been successful? How would you design a system to prevent this kind of abuse? Travis M. Andrews, The Washington Post (2017): Wisdom of the crowd? IMDb users gang up on Christian Bale’s new movie before it even opens.

Response

Based on readings, user-adaptive systems such as collaborative recommender systems seem to be most vulnerable to attacks and security issues. The extract below shows an example of how content recommendation can be altered in collaborative recommender systems by what is called a ‘nuke attack’.

" Suppose the system is using a simplified userbased collaborative-filtering approach that obtains the predicted ratings for Alice on item 6 by finding Alice’s closest neighbor. Without the attack profiles, the most similar user to Alice, using correlationbased similarity, is user 6. The prediction associated with item 6 is 2,essentially stating that Alice will dislike item 6. After the attack, however, the attack 1 profile (a push attack) is most similar to Alice, yielding a predicted rating of 5 for item 6. So, in this example, attack 1 is successful, and the system recommends item 6 to Alice. Alternatively, Eve might seek to prevent the system from recommending an item—a competitor’s product, for example. An attack with this aim is called a nuke attack. Such an attack looks similar to the push attack, but the target item instead receives a low score" (Mobasher et al., 2007).

In order to design a system to prevent this kind of abuse, it is first important to understand how they occur and be cognizant of triggers that may influence their occurrence. For example, one way that a user attacks a system is by creating multiple profiles under false identities . Two approaches to addressing this problem can be: (1) profile classification which could involve identifying suspicious profiles and discounting their contribution toward predictions and (2) make creation of a profile very challenge or expensive. Both approaches have strengths and weaknesses which will have to be deliberated upon according to circumstances to discern tradeoffs.

References

Codecademy, Netflix Recommendation Engine. (2019). Netflix Recommendation Engine | Codecademy. Codecademy. https://www.codecademy.com/articles/how-netflix-recommendation-works-data-science#:~:text=Companies%20like%20Netflix%20collect%20thousands,known%20as%20a%20recommender%20engine.&text=This%20suggestion%20is%20the%20Netflix,it%20thinks%20you%20will%20enjoy.

Mobasher, B., Burke, R., Bhaumik, R., & Sandvig, J. J. (2007). Attacks and Remedies in Collaborative Recommendation. IEEE Intelligent Systems, 22(3), 56–63. doi:10.1109/mis.2007.45 

Plummer, L. (2017, August 22). This is how Netflix’s top-secret recommendation system works. Wired.Co.Uk; WIRED UK. https://www.wired.co.uk/article/how-do-netflixs-algorithms-work-machine-learning-helps-to-predict-what-viewers-will-like