Online news recommendations based on topic modeling and online interest adjustment
| Date | 09 September 2019 |
| Published date | 09 September 2019 |
| Pages | 1802-1818 |
| DOI | https://doi.org/10.1108/IMDS-04-2019-0251 |
| Author | Duen-Ren Liu,Yu-Shan Liao,Jun-Yi Lu |
| Subject Matter | Information & knowledge management,Information systems,Data management systems,Knowledge management,Knowledge sharing,Management science & operations,Supply chain management,Supply chain information systems,Logistics,Quality management/systems |
Online news recommendations
based on topic modeling and
online interest adjustment
Duen-Ren Liu, Yu-Shan Liao and Jun-Yi Lu
National Chiao Tung University, Hsinchu, Taiwan
Abstract
Purpose –Providing online news recommendations to users has become an important trend for online media
platforms, enabling them to attract more users. The purpose of this paper is to propose an online news
recommendation system for recommending news articles to users when browsing news on online media platforms.
Design/methodology/approach –A Collaborative Semantic Topic Modeling (CSTM) method and an
ensemble model (EM) are proposed to predict user preferences based on the combination of matrix
factorization with articles’semantic latent topics derived from word embedding and latent topic modeling.
The proposed EM further integrates an online interest adjustment (OIA) mechanism to adjust users’online
recommendation lists based on their current news browsing.
Findings –This study evaluated the proposed approach using offline experiments, as well as an online
evaluation on an existing online media platform. The evaluation shows that the proposed method can
improve the recommendation quality and achieve better performance than other recommendation methods
can. The online evaluation also shows that integrating the proposed method with OIA can improve the
click-through rate for online news recommendation.
Originality/value –The novel CSTM and EM combined with OIA are proposed for news recommendation.
The proposed novel recommendation system can improve the click-through rate of online news
recommendations, thus increasing online media platforms’commercial value.
Keywords Recommendation system, Collaborative topic modelling, Online recommendation,
Recommendation adjustment, Semantic latent topic analysis
Paper type Research paper
1. Introduction
With the ubiquitous internet, growing numbers of people are obtaining news articles on
various topics, such as lifestyle, fashion and entertainment from new types of online media
platforms. As these new media platforms have become the primary news sources for users,
information overloading can make it difficult to meet individual user demands. In order to
address this problem, recommendation systems have been developed to enable users to
access news on desired topics, thereby increasing user loyalty and browsing aspirations for
online media platforms.
New articles are published on the websites every day. Thus, online news recommendation
systems need to recommend cold start news articles based on very few browsing records.
Research on news recommendation systems has focused on analyzing users’historical records,
including their ratings or click status, to achieve good recommendation results (Kompan and
Bieliková, 2010; Li et al., 2014; Liu et al.,2010;Zhuet al., 2018). Collaborative filtering (CF)
(Balabanovićand Shoham, 1997; Koren and Bell, 2015), like matrix factorization (MF ) (Koren
et al., 2009), is commonly used to extract latent user and item factors in order to predict
preference ratings. However, MF still faces the cold start and data sparsity issues, which may
influence the recommendation results. Therefore, collaborative topic modeling (CTM) (Li et al.,
2013; Wang and Blei, 2011) was proposed to address the above issues by combining
content-based filtering (CBF), such as latent Dirichlet allocation (LDA) (Blei et al.,2003),the
Industrial Management & Data
Systems
Vol. 119 No. 8, 2019
pp. 1802-1818
© Emerald PublishingLimited
0263-5577
DOI 10.1108/IMDS-04-2019-0251
Received 26 April 2019
Revised 2 July 2019
Accepted 29 July 2019
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
This research was supported by the Ministry of Science and Technology of Taiwan under
Grant No. MOST 105-2410-H-009-033-MY3. This research was conducted in collaboration with
NIUSNEWS (www.niusnews.com/).
1802
IMDS
119,8
latent topic modeling approach, with MF to improve the insufficiency of MF. However, news
articles not only comprise content topics, but word embedding semantic vectors also play the
important role of linking the elements of the text together (Das et al., 2015). Combining latent
topic distribution with word semantic vectors is more effective for discovering latent factors of
news articles. In addition, when users select news online, they are affected by various factors,
including their reading preferences, the recommendation layout, popularity and news content
(Kompan and Bieliková, 2010; Lommatzsch, 2014). Most studies emphasize offline analysis or
click behavior to analyze user preferences; very few have integrated offline and online analyses
to comprehensively fulfill users’online information needs. Most studies therefore had difficulty
providingeffectiveonlinenewsrecommendations when facing complex online user behavior.
Online user preferences may dynamically change according to the news that users are
currently browsing. Some studies focused on how to accelerate the offline learning models
or algorithms to enable more efficient online learning of user preferences (He et al., 2016;
Pálovics et al., 2014). However, they did not adopt the users’behavior and news correlation
to dynamically adjust the recommendation scores in online recommendation based on users’
current news browsing and the frequency of online news recommendations. Moreover,
online websites have a limited display layout for online recommendations, especially on
mobile devices.
This study proposes a novel recommendation system that considers multiple aspects of
user behavior in both offline preference analysis and online interest adjustment (OIA). In
offline preference analysis, the semantic LDA model (SLDA) is adopted to extract the
semantic latent news topic articles by integrating the semantic vectors of word embedding
with LDA. Moreover, Collaborative Semantic Topic Modeling (CSTM) is proposed to predict
user preferences by combining the SLDA and MF. An ensemble model (EM) SLDA–CSTM,
which combines SLDA-based and CSTM-based prediction models is further proposed to
predict user preferences.
A novel online news recommendation system integrating the SLDA–CSTM EM with
the OIA mechanism is proposed herein. Two OIA mechanisms are proposed to adjust
online interest scores of target news based on their similarity measures and association
with the news articles currently being browsed by the target user. Through the dynamic
adjustment of online interest scores, online recommendation can be better adapted to the
target user’s current browsing interest. Moreover, the proposed OIA approach adopts the
recommendation adjustment (RA) mechanism (Liu, Chen, Chou and Lee, 2017) to
dynamically adjust the recommendation list. Accordingly, the proposed news
recommendation system can adjust the recommendation list to recommend news that is
of potentially gr eater interest to use rs without restricting the number of recommended
news items due to the limited display layout.
The proposed approach is evaluated using both offline and online experiments in a real
website. The proposed recommendation system is implemented on an online media website,
NIUSNEWS (www.niusnews. com/). The offline experiments and online evaluation show
that the proposed CSTM and EM methods can improve the recommendation quality and
achieve better performance than the CTM method can. The online evaluation also shows
that the proposed EM method integrated with online interest analysis and RA can improve
the click-through rate for online news recommendation.
The remainder of this paper is arranged as follows. Section 2 illustrates related works.
The proposed recommendation approach is presented in Section 3. Section 4 describes and
analyzes the experiment and evaluation results. The final section concludes the research
and discusses possible future work.
2. Related work
This section introduces related work and the methods adopted in the proposed system.
1803
Online news
recommendations
To continue reading
Request your trial