From Online Behaviors to Offline Retailing
Ping Luo
♯
, Su Yan
♯♭
, Zhiqiang Liu
§
, Zhiyong Shen
§
, Shengwen Yang
§
, Qing He
♯
♯
Key Lab of Intelligent Information Processing of Chinese Academy of Sciences
Institute of Computing Technology, CAS, Beijing, China.
♭
University of Chinese Academy of Sciences, Beijing, China.
§
Baidu, Inc., Beijing, China.
luop@ict.ac.cn, shenzhiyong@baidu.com
ABSTRACT
To combat the ease of online shopping in pajamas, offline
mall owners focus increasingly on driving satisfaction and
improving retention by identifying customers’ preferences.
However, most of these studies are based on customers’ of-
fline consuming history only. Benefiting from the internet,
we can also get customers’ online behaviors, such as the
search logs, web browsing logs, online shopping logs, and so
on. Might these seemingly irrelevant information from two
different modalities (i.e. online and offline) be somehow in-
terrelated? How can we make use of the online behaviors
and offline actions jointly to promote recommendation for
offline retailing?
In this study, we formulate this task as a cross-modality
recommendation problem, and present its solution via a pro-
posed probabilistic graphical model, called Online-to-Offline
Topic Modeling (O2OTM). Specifically, this method explic-
itly models the relationships between online and offline top-
ics so that the likelihood of both online and offline behav-
iors is maximized. Then, the recommendation is made only
based on the pairs of online and offline topics, denoted by
(t, l), with high values of lift, such that the existence of the
online topic t greatly increases the response on the corre-
sponding offline topic l compared with the average response
for the population without the online topic t. Furthermore,
we evaluate this solution in both live and retrospect exper-
iments. The real-world deployment of this model for the
anniversary promotion campaign of a famous shopping mall
in Beijing shows that our approach increases the occurred
customer purchases per promotion message by 29.75% com-
pared with the baseline. Also, our model finds some inter-
esting interpretable relationships between the online search
topics and offline brand topics.
Keywords
Brands recommendation, Topic modeling, Recommendation
explanation
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.
KDD ’16, August 13-17, 2016, San Francisco, CA, USA
c
2016 ACM. ISBN 978-1-4503-4232-2/16/08. . . $15.00
DOI: http://dx.doi.org/10.1145/2939672.2939683
1. INTRODUCTION
Recommendation systems (RS) [16] have been popular for
decades, improving the quality of our daily lives by facilitat-
ing the item selections from a large set of candidates. Here,
we focus on the recommendation systems for offline retail-
ing. Although online shopping has the tendency of booming
development, offline retailing still covers 92% and 90% of
the US and China retailing market in 2014 [1, 2], respec-
tively. Thus, small improvements on offline retailing might
draw great benefits. Meanwhile, with the penetration of
IT systems into this traditional sector, the customers’ of-
fline consumption logs can be recorded to better undersatnd
customers’ interests for driving satisfaction and improving
retention. However, previous studies mostly consider the
behaviors from offline and online separately [8, 12].
In this paper, we study the online and offline activities
from a joint perspective. Here, online behaviors consist of
all the activities on Internet, such as web browsing, web
searching, online video watching, online music listening, and
online shopping etc, while offline behaviors include all the
activities in real lives, such as shopping in a mall, traveling,
dinning out, watching films etc. Figure 1 shows an example
of the data we have from these two modalities. For a smal-
l fraction of users, we might know their activities on both
online and offline sides. On the online side, we know what
they searched in a search engine, what videos they watched,
and what items they bought online. On the offline side,
we know what brands they consumed in an offline shopping
mall. With these data, we might ask could the videos user-
s watch on Youtube interpret the tour sites they select for
travelling? Could the news users read on internet be related
with the books they borrow in a library? If we connect the
online and offline behaviors together, could these two modal-
ities of data fire the chemical reactions for more meaningful
recommendation?
This study pioneers towards this direction. As a case s-
tudy, here we consider the online search logs as the online
behaviors, and the offline shopping history as the offline be-
haviors (note that the proposed model can b e applied to
any type of online and offline activities). Originally, these
two sets of data are owned by Internet companies and local
malls, separately. We will show how profitable if Internet
companies with online data and local retailers with offline
data cooperate for a win-win situation (with the consent
from users).
Our basic assumption in this study is that customers’ on-
line behaviors could reflect their interests, which in turn in-
fluence their offline behaviors. From the real-world data, we