Identifying impact of intrinsic factors on topic preferences in online social media: A nonparametric hierarchical Bayesian approach

Social media offers a new communication channel for users and affords an interactive opportunity between users and the firms about the products and the brands. Understanding what topics are important to users and the corresponding internal motivation is crucial for managers to successfully engage customers and promote business through social media. Assuming topic preference is the outcome of intrinsic factors such as gender, age and personality traits, this paper proposes an improved nonparametric hierarchical Bayesian topic (NHBT) model to investigate the multiple-to-multiple generative relationships from intrinsic factors to topic preferences. The proposed NHBT model employs a three-level generation framework based on Dirichlet process to study the impact of intrinsic factors on users topic preference. Our study of Facebook data shows that NHBT model is able to draw valuable latent topics (e.g. music band, chemical biology, cosplay) from the open social media environment, and reveal the internal motivation for users topic selection behaviors (e.g. users with low conscientiousness and high extraversion personality prefer topics about campus party). Our experiments also show that NHBT model can identify the intrinsic factors dominating topic preferences for individual users, and provide foundations to predict the intrinsic factors for new user generated contents.