Image Network and Interest Group - A Heterogeneous Network Embedding Approach to Analyze Social Curation on Pinterest
Abstract:
Social curation platforms help consumers navigate through vast digital content to find what fits their interests. However, little is known about this important phenomenon. Using the popular image curation site Pinterest.com as empirical context, this research aims to understand: (i) how digital content is organized at social curation platforms; (ii) what users' curation activities reveal about consumer preferences, content characteristics, and brand positioning; and (iii) how we can predict users' curation actions.
We propose a novel approach which has two key components. First, we represent social curation using a heterogeneous information network. Images, users, and curation words are represented as nodes, while edges represent users' collection and annotation actions. Second, we leverage heterogeneous network embedding, a recently developed machine learning method, to map network nodes to lower-dimensional vectors, while preserving the network's structural and semantic information. We then analyze the embedding vectors for prediction and interpretation.
Analyzing a large furniture-related dataset from Pinterest, our proposed approach significantly outperforms prevailing benchmarks on predicting users' curation actions. Furthermore, embedding results reveal various user interest groups and image clusters, each with distinct characteristics. The match between users and images is stable out-of-sample. The analysis also generates insights on brand positions.
Bio:
Kunpeng 'KZ' Zhang is currently an Assistant Professor of Information Systems at the University of Maryland, College Park. He received his PhD in computer science from Northwestern University. His research mainly focuses on large-scale unstructured social media analytics by developing and applying machine learning algorithms, including social network analysis, image analysis and text mining. He currently serves as Associate Editor of INFORMS Journal on Computing. For more details, please see his website: https://kpzhang.github.io.