Recommendation systems have become a cornerstone of modern life, spanning sectors that include online retail, music and video streaming, and even content publishing. These systems help us navigate the sheer volume of content on the internet, allowing us to discover what’s interesting or important to us. The classic modeling approaches to recommendation systems can be broadly categorized as content-based, as collaborative filtering-based, or as hybrid approaches that combine aspects of the two. These approaches generally tend to utilize historical user-item interactions (i.e., the items that a user has clicked on in the past) to learn a user’s long-term preferences. The underlying assumption in both of these systems is that all of the historical interactions are equally important to the user’s current preference—but in reality, this may not be true.
A user’s choice of items not only depends on long-term historical preference, but also on short-term and more recent preferences. Choices almost always have time-sensitive context; for instance, “recently viewed” or “recently purchased” items may actually be more relevant than others. These short-term preferences are embedded in the user’s most recent interactions, but may account for only a small proportion of historical interactions. In addition, a user’s preference towards certain items can tend to be dynamic rather than static; it often evolves over time.
These considerations have prompted the exploration and development of a new class of recommendation algorithms: known as session-based recommendation algorithms, these rely heavily on the user’s most recent interactions, rather than on the user’s historical preferences. In addition, this approach is especially advantageous because a user could appear anonymously—that is, a user may not be logged in or may be browsing incognito.
The latest report and experiments from Cloudera Fast Forward Labs explores session-based recommendation algorithms that provide recommendations solely based on a user’s interactions in an ongoing session, and which do not require the existence of user-profiles or their entire historical preferences. It explores a simple, yet powerful, NLP-based approach (word2vec). While NLP-based approaches are generally employed for linguistic tasks, here we exploit them to recommend the next item to a user.
Want to learn more? Join us on Wednesday, June 2nd at 9 AM PST for a live stream with Nisha Muktewar and Melanie Beck of Cloudera Fast Forward Labs. You don’t want to miss the demo and live Q&A.