Joint Deep Modeling of Users and Items Using Reviews for
Recommendation
Lei Zheng
Department of Computer
Science
University of Illinois at Chicago
Chicago, U.S.
lzheng21@uic.edu
Vahid Noroozi
Department of Computer
Science
University of Illinois at Chicago
Chicago, U.S.
vnoroo2@uic.edu
Philip S. Yu
Department of Computer
Science
University of Illinois at Chicago
Chicago, U.S.
psyu@uic.edu
ABSTRACT
A large amount of information exists in reviews written by
users. This source of information has been ignored by most
of the current recommender systems while it can potentially
alleviate the sparsity problem and improve the quality of rec-
ommendations. In this paper, we present a deep model to
learn item properties and user behaviors jointly from review
text. The proposed model, named Deep Cooperative Neural
Networks (DeepCoNN), consists of two parallel neural net-
works coupled in the last layers. One of the networks focuses
on learning user behaviors exploiting reviews written by the
user, and the other one learns item properties from the re-
views written for the item. A shared layer is introduced on
the top to couple these two networks together. The shared
layer enables latent factors learned for users and items to
interact with each other in a manner similar to factoriza-
tion machine techniques. Experimental results demonstrate
that DeepCoNN significantly outperforms all baseline rec-
ommender systems on a variety of datasets.
CCS Concepts
•Information systems → Collaborative filtering; Rec-
ommender systems; •Computing methodologies →
Neural networks;
Keywords
Recommender Systems, Deep Learning, Convolutional Neu-
ral Networks, Rating Prediction
1. INTRODUCTION
The variety and number of products and services provided
by companies have increased dramatically during the last
decade. Companies produce a large number of products
to meet the needs of customers. Although this gives more
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full cita-
tion on the first page. Copyrights for components of this work owned by others than
ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-
publish, to post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from permissions@acm.org.
WSDM 2017, February 06-10, 2017, Cambridge, United Kingdom
c
2017 ACM. ISBN 978-1-4503-4675-7/17/02. . . $15.00
DOI: http://dx.doi.org/10.1145/3018661.3018665
options to customers, it makes it harder for them to pro-
cess the large amount of information provided by companies.
Recommender systems help customers by presenting prod-
ucts or services that are likely of interest to them based on
their preferences, needs, and past buying behaviors. Nowa-
days, many people use recommender systems in their daily
life such as online shopping, reading articles, and watching
movies.
Many of the prominent approaches employed in recom-
mender systems [13] are based on Collaborative Filtering
(CF) techniques. The basic idea of these techniques is that
people who share similar preferences in the past tend to have
similar choices in the future. Many of the most successful
CF techniques are based on matrix factorization [13]. They
find common factors that can be the underlying reasons for
the ratings given by users. For example, in a movie recom-
mender system, these factors can be genre, actors, or direc-
tor of movies that may affect the rating behavior of users.
Matrix factorization techniques not only find these hidden
factors, but also give the importance of them for each user
and how each item satisfies each factor.
Although CF techniques have shown good performance
for many applications, the sparsity problem is considered as
one of their significant challenges [13]. The sparsity problem
arises when the number of items rated by users is insignifi-
cant to the total number of items. It happens in many real
applications. It is not easy for CF techniques to recommend
items with few ratings or to give recommendations to users
with few ratings.
One of the approaches employed to address this lack of
data is using the information in review text [16, 17]. In
many recommender systems, other than the numeric rat-
ings, users can write reviews for the products. Users explain
the reasons behind their ratings in text reviews. The re-
views contain information which can be used to alleviate
sparsity problem. One of the drawbacks of most current CF
techniques is that they model users and items just based
on numeric ratings provided by users and ignore the abun-
dant information existed in the review text. Recently, some
studies [17] [16] have shown that using review text can im-
prove the prediction accuracy of recommender systems, in
particular for items and users with few ratings [34].
In this paper, we propose a Neural Network (NN) based
model, named Deep Cooperative Neural Networks (Deep-
CoNN), to model users and items jointly using review text
for rating prediction problems. The proposed model learns