Content-Based Bipartite User-Image Correlation for Image Recommendation
- PDF / 2,008,547 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 3 Downloads / 231 Views
Content-Based Bipartite User-Image Correlation for Image Recommendation Meng Jian1
· Ting Jia1 · Lifang Wu1 · Lei Zhang1 · Dong Wang1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The popularity of online social curation networks takes benefits from its convenience to retrieve, collect, sort and share multimedia contents among users. With increasing content and user intent gap, effective recommendation becomes highly desirable for its further development. In this paper, we propose a content-based bipartite graph for image recommendation in social curation networks. Bipartite graph employs given sparse user-image interactions to infer user-image correlation for recommendation. Beside given user-image interactions, the user interacted visual content also reveals valuable user preferences. Visual content is embedded into the bipartite graph to extend the correlation density and the recommendation scope simultaneously. Furthermore, the content similarity is employed for recommendation reranking to improve the visual quality of recommended images. Experimental results demonstrate that the proposed method enhances the recommendation ability of the bipartite graph effectively. Keywords Bipartite graph · Visual correlation · Personalized recommendation · Social multimedia network
1 Introduction With the rapid development of multimedia technology, online social networks, like Facebook, Twitter, Weibo, and Flickr, are emerging to change the way people communicate which successfully makes conversations more convenient with multimedia contents. Pinterest [1,2], a content curation social networks (CCSNs), was established in 2009. It became the fourth largest social network website in the following four years in America. The special “Photo wall” service of Pinterest attracts lots of users to share their preferred images. Besides, several similar social network sites such as Huaban, Meilishuo and so on have been built sequentially in China. Because of the fast development of CCSNs, many related research topics become hot to researchers. Liben-Nowell [3] indicates that users are more likely to know strangers by their own friends and family, which demonstrated the principle of social network. Cinar
B 1
Lifang Wu [email protected] Faculty of Information Technology, Beijing University of Technology, Beijing, China
123
M. Jian et al.
et al. [4] utilized text and images to infer users’ interests on social media. Geng et al. [5] analyzed user profiling by social curation. Yang et al. [6] recommended boards on Pinterest. To multimedia overload, the personalized recommendation becomes pivotal for users to seek satisfactory contents. A lot of graph-based models [7,8] are elaborately constructed for social recommendation, thanks to its capability to represent social relations with its intrinsic structure in capturing interaction attributes. The bipartite graph is a representative of collaboration filtering (CF) recommendation algorithms, which is constructed to reveal hidden correlations between user
Data Loading...
