Topic-sensitive Influencer Mining in Interest-Based Social Media Networks via Hypergraph Learning


Quan Fang, Jitao Sang and Changsheng Xu





Social media is emerging as a new mainstream means of interacting around online media. Social influence mining in social networks is therefore of critical importance in real-world applications such as friend suggestion and photo recommendation. Social media is inherently multimodal, including rich types of user contributed content and social link information. Most of the existing research suffers from two limitations: 1) only utilizing the textual information, and/or 2) only analyzing the generic influence but ignoring the more important topic-level influence. To address these limitations, in this paper we develop a novel Topic-Sensitive Influencer Mining (TSIM) framework in interest-based social media networks. Specifically, we take Flickr as the study platform. People in Flickr interact with each other through images. TSIM aims to find topical influential users and images. The influence estimation is determined with a hypergraph learning approach. In the hypergraph, the vertices represent users and images, and the hyperedges are utilized to capture multi-type relations including visual-textual content relations among images, and social links between users and images. Algorithmwise, TSIM first learns the topic distribution by leveraging user-contributed images, and then infers the influence strength under different topics for each node in the hypergraph. Extensive experiments on a real-world dataset of more than 50K images and 70K comment/favorite links from Flickr have demonstrated the effectiveness of our proposed framework. In addition, we also report promising results of friend suggestion and photo recommendation via TSIM on the same dataset.






The above figure shows the framework of our proposed approach for TSIM, which primarily consists of three learning stages: hypergraph construction, topic distribution learning, and topic-sensitive influence ranking. First, a unified hypergraph is constructed to model users, images, and multi-type relations in the Flickr network. Visual-textual content relations among images are used to construct homogeneous hyperedges. Social link relations between users and images are used to generate heterogeneous hyperedges. Second, due to the sparse contextual links of images, informative images with rich tags are selected to learn the topics by employing the hypergraph regularized topic model with homogeneous hyperedges. We obtain the topic distribution for all images and users via the collaborative representation based similarity propagation. Finally, a hypergraph ranking algorithm based on affinity propagation is performed on the hypergraph with heterogeneous hyperedges to obtain the topic-specific social influence for users and images. We have conducted experiments on a real-world dataset from Flickr to validate the effectiveness of our proposed approach. In particular, we apply our model to the social tasks of friend suggestion and photo recommendation, in which it has shown superior performance.


Experimental Results


We conduct the experiments on the dataset collected from Flickr containing 2,314 users, 556,942 photos.

Topic-Sensitive Influencers

We present the representative nodes identification on the collected Flickr dataset. The above figures illustrates some representative nodes (users and photos) found on four topics of interest from dataset by our TSIM algorithm. The representative score depicted in red font is the social influence score of each user inferred by TSIM on specific topics; the score depicted in blue font is the social influence score of each photo on specific topics. The influence score indicates the strength of topic-level social influence of users and photos in the social circles. User and their photos are ranked by their corresponding influence scores. We generate a word cloud for each topic using Wordle. The topics are represented by their most relevant tags, which are presented with the font size proportional to $P(w|z)$. We present eight topics. Topic 1 relates to flowers, topic 7 relates to girls, topic 13 relates to city, topic 21 relates to beach and sunset scene, topic 16 relates to sky and field scene, topic 23 relates to decayed scene, topic 47 relates to portrait, topic 49 relates to photoshot. From the representative photos, we can clearly see that the photos are relevant with the topics respectively.




Topic-sensitive Influencer Mining in Interest-Based Social Media Networks via Hypergraph Learning [pdf]

Quan Fang, Jitao Sang and Changsheng Xu
In IEEE Transactions on Multimedia 16(3): 796-812 (2014).