Resnick, Paul, et al. “GroupLens: an open architecture for collaborative filtering of netnews.” Proceedings of the 1994 ACM conference on Computer supported cooperative work. ACM, 1994.
Summary:
This 1994 paper discusses the architecture, “ongoing experimentation”, and social implications of GroupLens, a distributed system using collaborative filtering for gathering, disseminating, and using ratings from some users to predict other users’ interest in articles. The architecture of GroupLens follows 5 design goals: openness, ease of use, compatibility, scalability, and privacy. Openness was highly discussed in the paper regarding how the GroupLens architecture can be used for many other netnews applications. Additionally, GroupLen’s architecture is described using a prediction algorithm that helps recommend articles through user generated ratings. Based on the ratings you are recommend articles that similar thinking peers read. The paper eds by discussing social implications and future work regarding ongoing experimentation (that has since finished years ago).
Reflection:
In 1994, GroupLens pushed the boundaries of predicting what users want to be recommended. My first thought when reading this paper was about recommender systems. A lot fo research and money has gone into recommender systems and how they can be used to improve “software as a service” services. The idea of recommending an article, movie or tv show, music, or products to buy allow for giving the user useful recommendations but also learning about the user themselves. Of course, this 1994 paper on GroupLens only presents its architecture and social implications the authors were correct that the implications may become far and wide. This work has led to huge tech companies creating world changing services. Some examples of these companies are Netflix, YouTube (Google by extension), and Amazon. Of course, this is not an exhaustive list, but these companies provide some services that utilize what GroupLens presented.
Netflix provides an entertainment services by hosting hundreds of movies and tv shows to users at a low monthly cost. When a user signs up, they are allowed to watch any movie or tv show available. At first, the selection of movies shown seem random and are more of the popular variety until the user starts watching a few movies and tv shows. Eventually, the “front page” of the user that shows movies and tv shows starts to recommend content similar to what you have watched prior. The recommended content is generated on search history, watch history, and user selected genres. This follow the template of GroupLens that used user ratings for new articles. User generated content is used to give context of what a system show recommend to the user. In a similar vein, YouTube follows a very similar approach to recommending videos to users. There is additional information from users like pressing the like button, comment history, and user generated playlists. What this leads to is that the more information one has about the user to the finer grain recommendations can be.
Lastly, Amazon may be the quintessential example here. Amazon’s history sits in selling products starting with books. Eventually, they began selling other products to customers online. This provided an opportunity to recommend products to users based on their purchase history and search history and it provides and experience that shoppers like. It is fast, convenient, “it knows what I need”, provides other similar products, and make shopping quick and easy. Of course, I don’t consider other information sources like cookies or trackers nor other additional services built onto a main service (i.e. amazon music when shopping on amazon) because they aren’t relevant to my thoughts here. What I wanted to reflect on was how tech giants built their success of providing niche services by using GroupLens insight of knowing a user’s preferences and providing a friendlier experience with their services because of this knowledge of users. To me, GroupLens began what we have for online services today.