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: Disentangled Self-Attentive Neural Networks for Click-Through Rate Prediction

Abstract: Click-Through Rate (CTR) prediction, whose aim is to predict the probability of whether a user will click on an item, is an essential task for many online applications. Due to the nature of data sparsity and high dimensionality of CTR prediction, a key to making effective prediction is to model high-order feature interaction. An efficient way to do this is to perform inner product of feature embeddings with self-attentive neural networks. To better model complex feature interaction, in this paper we propose a novel DisentanglEd Self-atTentIve NEtwork (DESTINE) framework for CTR prediction that explicitly decouples the computation of unary feature importance from pairwise interaction. Specifically, the unary term models the general importance of one feature on all other features, whereas the pairwise interaction term contributes to learning the pure impact for each feature pair. We conduct extensive experiments using two real-world benchmark datasets. The results show that DESTINE not only maintains computational efficiency but achieves consistent improvements over state-of-the-art baselines.
Comments: Accepted to CIKM 2021; authors' version
Subjects: Information Retrieval (cs.IR)
DOI: 10.1145/3459637.3482088
Cite as: arXiv:2101.03654 [cs.IR]
  (or arXiv:2101.03654v3 [cs.IR] for this version)

Submission history

From: Yanqiao Zhu [view email]
[v1] Mon, 11 Jan 2021 01:14:27 GMT (10107kb,D)
[v2] Thu, 10 Jun 2021 12:14:13 GMT (6933kb,D)
[v3] Tue, 17 Aug 2021 03:26:08 GMT (7003kb,D)

Link back to: arXiv, form interface, contact.