DeepMind researchers provide theoretical analysis on recommender system, ‘echo chamber’ and ‘filter bubble effect’

3 min read

DeepMind researchers published a paper last week, titled ‘Degenerate Feedback Loops in Recommender Systems’. In the paper, researchers provide a new theoretical analysis examining the user dynamics role and the behavior of recommender systems, that can help remove the echo chamber from the filter bubble effect.

Recommender systems are aimed to provide users with personalized product and information offerings. These systems take into consideration the user’s personal characteristics and past behaviors to generate a list of items that have been personalized as per the user’s tastes.

Although very successful, there are certain concerns related to the systems that it might lead to a self-reinforcing pattern of narrowing exposure and a shift in user’s interest. These problems are often called the “echo chamber” and “filter bubble”.

In the paper, researchers define echo chamber as user’s interest being positively or negatively reinforced due to the repeated exposure to a certain category of items. For “filter bubble”, researchers use the definition introduced by Pariser (2011) that states that the recommender systems select limited content to serve the users online.

Researchers have considered a recommender system that is capable of interacting with a user over time. At every time step, the recommender system serves a different number of items (or categories of items such as news articles, videos, or consumer products) to a user from a set of finite or countably infinite items. The goal of this recommender system is to provide those items to a user that she/he might be interested in.

                    The interaction between the recommender system and user

The paper also considers the fact that the user’s interaction with the recommender system can change depending on her interest in different items for the next interaction. Additionally, to further analyze the echo chamber or filter bubble effect in recommender systems, researchers track when the user’s interest changes extremely.

Futhermore, researchers used the dynamical system framework to model the user’s interest. They treated the interest extremes of the users as the degeneracy points within the system. For the recommender system, researchers discussed the influence on the degeneracy speed of these three independent factors in system design including model accuracy, amount of exploration, and the growth rate of the candidate pool.

As per the researchers, continuous random exploration along with linearly growing the candidate pool is the best methods against system degeneracy. Although this research is quite effective, it still has two main limitations. The first limitation is that user interests are hidden variables and are not observed directly which is why a good measure for user interests is needed for practice to reliably study the degeneration process.

Secondly, since the researchers have assumed the items and users being independent of each other, the theoretical analysis has been extended to study possibly mutually dependent items and users in the future.

For more information, check out the official research paper.

Read Next

Google DeepMind’s AI AlphaStar beats StarCraft II pros TLO and MaNa; wins 10-1 against the gamers

Blizzard set to demo Google’s DeepMind AI in StarCraft 2

Deepmind’s AlphaZero shows unprecedented growth in AI, masters 3 different games