Estimating posterior inference quality of the relational infinite latent feature model for overlapping community detection