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

Download:

Current browse context:

cs.IR

Change to browse by:

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: Review Regularized Neural Collaborative Filtering

Abstract: In recent years, text-aware collaborative filtering methods have been proposed to address essential challenges in recommendations such as data sparsity, cold start problem, and long-tail distribution. However, many of these text-oriented methods rely heavily on the availability of text information for every user and item, which obviously does not hold in real-world scenarios. Furthermore, specially designed network structures for text processing are highly inefficient for on-line serving and are hard to integrate into current systems. In this paper, we propose a flexible neural recommendation framework, named Review Regularized Recommendation, short as R3. It consists of a neural collaborative filtering part that focuses on prediction output, and a text processing part that serves as a regularizer. This modular design incorporates text information as richer data sources in the training phase while being highly friendly for on-line serving as it needs no on-the-fly text processing in serving time. Our preliminary results show that by using a simple text processing approach, it could achieve better prediction performance than state-of-the-art text-aware methods.
Subjects: Information Retrieval (cs.IR); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2008.13527 [cs.IR]
  (or arXiv:2008.13527v1 [cs.IR] for this version)

Submission history

From: Zhimeng Pan [view email]
[v1] Thu, 20 Aug 2020 18:54:27 GMT (68kb,D)

Link back to: arXiv, form interface, contact.