We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.IR

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Information Retrieval

Title: Designing a Sequential Recommendation System for Heterogeneous Interactions Using Transformers

Abstract: While many production-ready and robust algorithms are available for the task of recommendation systems, many of these systems do not take the order of user's consumption into account. The order of consumption can be very useful and matters in many scenarios. One such scenario is an educational content recommendation, where users generally follow a progressive path towards more advanced courses. Researchers have used RNNs to build sequential recommendation systems and other models that deal with sequences. Sequential Recommendation systems try to predict the next event for the user by reading their history. With the massive success of Transformers in Natural Language Processing and their usage of Attention Mechanism to better deal with sequences, there have been attempts to use this family of models as a base for a new generation of sequential recommendation systems. In this work, by converting each user's interactions with items into a series of events and basing our architecture on Transformers, we try to enable the use of such a model that takes different types of events into account. Furthermore, by recognizing that some events have to occur before some other types of events take place, we try to modify the architecture to reflect this dependency relationship and enhance the model's performance.
Subjects: Information Retrieval (cs.IR)
Cite as: arXiv:2205.00265 [cs.IR]
  (or arXiv:2205.00265v1 [cs.IR] for this version)

Submission history

From: Mehdi Soleiman Nejad [view email]
[v1] Sat, 30 Apr 2022 13:13:09 GMT (705kb,D)

Link back to: arXiv, form interface, contact.