Archive for the ‘Computational Intelligence’ Category

Implementing Recommendations at Feeds 2.0

Feeds 2.0 users are presented with recommendations for posts and feeds that they might want to read or subscribe to. It works by taking into account what the entire community of Feeds 2.0 readers with similar interests are finding interesting and it is based on two main ideas: Collaborative filtering and Content-based filtering.

Collaborative filtering (CF) filters information for a user based on a collection of user profiles. Users having similar profiles may share similar interests. Hence, CF exploits correlations between ratings across a population of users by first finding users most similar to some active user and by then forming a weighted vote over these neighbors to predict unobserved ratings. For a user, information can be filtered in/out regarding to the behaviors of his or her similar users. At Feeds 2.0, user profiles are collected implicitly which means that for each user his/her profile is based on passive observation and contains users historic interaction data.

Content-based filtering is an alternative paradigm that has been used mainly in the context of recommending items such as books, web pages, news, etc. for which informative content descriptors exist. Standard machine learning methods like those described in our previous post can be used in this context (e.g. neural networks, bayesian classifiers etc).

Feeds 2.0 implements a novel, unified approach for recommendations by combining collaborative and content-based filtering techniques. This combination systematically integrates all available training information such as past user-item ratings as well as attributes of items or users to learn predictions. The key ingredient is the examination of the similarity between user-item pairs that allows simultaneous generalization across the user and item attributes.

The user and item attributes are very useful especially in the, so-called, cold-start situations, that is for situations when new users or items enter the system for which little or no rating information is available. This is exactly the reason why users at Feeds 2.0 are immediately presented with recommendations when they have read (or marked as interesting) only just a small number of posts.

In our next post we will elaborate more on the Feeds 2.0 real-time Clustering feature for news items.