The battle of the algorithms, Xbox vs PlayStation

The similarities in Xbox and PlayStation recommendation algorithm

The console recommendation algorithm is a algorithm that is designed to recommend games and other content to users based on their individual interests and preferences. The algorithm uses a variety of factors to make its recommendations, such as the user’s past behavior on the console, their purchase history, and the ratings and reviews of other users. The goal of the algorithm is to provide personalized recommendations that will help users discover new content that they will enjoy.

This screenshot is off of the Xbox store. Xbox is showing the user games that he would like as the text in the screenshot says, “Based off your resent activities.”

The dataset in Xbox and its recommendations

The dataset contains information about how people use Xbox consoles. This includes data on how long they interact with different games and movies in the Xbox catalog. The dataset has millions of active users, most of whom play games on their Xbox consoles. Some users also watch movies using their Xbox consoles. The number of Xbox users increases on a daily basis, and the catalog of games and movies is regularly updated with new products. An article from authors Noam Koenigstein, Ulrich Paquet, Nir Nice, and Nir Schleyen, “The Xbox Recommender,” says Because of the large number of users, the dataset rarely experiences “cold start” problems, where there is not enough data on a new item or user to make accurate recommendations. However, in the movies domain, there can be “cold users” who have not watched any movies on their Xbox consoles. There are two identical recommendation engines to provide personalized recommendations for both games and movies. Unlike other recommendation problems, where user preferences are explicitly given in a rating scale, the data in this dataset largely consists of implicit signals, such as purchasing a game or watching a movie on the Xbox console. This poses a challenge for conventional recommendation methods that rely on factorizing a ratings matrix into user and item features.

The Xbox prediction model

In the bilinear model, each user is represented by a vector of numbers, and each item (such as a movie or song) is also represented by a vector(Koenigstein 3). The inner product of these two vectors represents the user’s affinity for the item, and can be used to personalize recommendations. In order to account for the fact that some items are more popular than others, a bias term is added for each item. This probability that a user will like or dislike an item is given by the formula:

p(lmn|um, vn, bn) = Φ(lmn * (uTmvn + bn))

where lmn is 1 if the user likes the item and -1 if the user dislikes the item, Φ is the Gaussian cumulative density function, and acts as a link function that maps its argument to a value in the (0, 1) interval. (Koenigstein 2) The parameters of the model (the vectors for each user and item and the bias terms) are inferred from the usage data. Alternatively, additional meta-data about the users or items can be used to improve the accuracy of the model. There is plenty more information on the dataset in Xbox, here is a PDF: (PDF) The Xbox recommender system (researchgate.net)

Image taken from the article “The Xbox Recommender.”

PlayStation Recommendation

The recommendation for PlayStation was very unclear to me as I’ve done much research and found minimal data and I’ll explain why. Searching for more information on PlayStations Recommendations, I found a reddit group of many people complaining about how inaccurate the recommendation of the PlayStation store is. A reddit user has posted a screenshot of the PlayStation store algorithm in progress as the recommendation seemed a little off.

Image taken from reddit post


According to this screenshot PlayStation has recommended this person Minecraft, Rocket League, and Battlefield 1, because he previously purchased The Evil Within. As I pointed out in the beginning of this blog the algorithm recommends video games off previous purchases and actions but, if you’re familiar with the games that were recommended, you would know that there is no correlation with the game that was purchased. Another screenshot below shows the humorous comments to people reacting to this screenshot.

It is obvious that players find that the recommendation system in the PlayStation store is a joke and that there isn’t real algorithms at work. Looking at screenshots this only brings me to the question, is there really a recommendation algorithm?

Is there a PlayStation recommendation algorithm?

Doing tons of research on the recommendation algorithm it is safe to say that there is no specific recommendation algorithm like Xbox for the PlayStation. However, the PlayStation store does have a section for recommended games that are tailored to each user’s individual interests and past purchases.

Which is better, Xbox or PlayStation?

In conclusion, the Xbox recommendation algorithm is superior to the PlayStation recommendation algorithm because Xbox has detailed information available on how their algorithm works, whereas PlayStation does not have a discernible recommendation algorithm in place. This transparency and understanding of the inner workings of the algorithm allows Xbox to make more accurate and personalized recommendations to its users, ultimately enhancing their gaming experience. On the other hand, the lack of information and clarity surrounding the PlayStation recommendation algorithm leaves users in the dark and potentially hinders their ability to discover new and enjoyable content.

Worksited

-“(PDF) The Xbox Recommender System – Researchgate.net.” The Xbox Recommender System, Noam Koenigstein, Ulrich Paquet, Nir Nice, and Nir Schleyen., 12 Sept. 2012, https://www.researchgate.net/publication/254464376_The_Xbox_recommender_system.

-“R/Playstation – PS4’s Game Recommendation System.” Reddit, 2017, https://www.reddit.com/r/playstation/comments/7mk4ag/ps4s_game_recommendation_system/.