Search papers, labs, and topics across Lattice.
This paper introduces RAPI, a framework for inferring user attributes from recommendation lists across four different scenarios. RAPI uses a surrogate recommendation model trained with content embeddings from BERT to simulate the original recommender, and then augments the recommendation lists using item similarity. An adaptive weight classification model is then used to infer user characteristics from the augmented lists, achieving improved accuracy compared to baselines.
Extracting user profiles from recommendation lists is now more accurate thanks to a novel framework that leverages BERT embeddings and adaptive weighting.
Recommender systems (RS) play a core role in various domains, including business analytics, helping users and companies make appropriate decisions. To optimize service quality, related technologies focus on constructing user profiles by analyzing users' historical behavior information. This paper considers four analytical scenarios to evaluate user profiling capabilities under different information conditions. A generic user attribute analysis framework named RAPI is proposed, which infers users' personal characteristics by exploiting easily accessible recommendation lists. Specifically, a surrogate recommendation model is established to simulate the original model, leveraging content embedding from a pre-trained BERT model to obtain item embeddings. A sample augmentation module generates extended recommendation lists by considering similarity between model outputs and item embeddings. Finally, an adaptive weight classification model assigns dynamic weights to facilitate user characteristic inference. Experiments on four collections show that RAPI achieves inference accuracy of 0.764 and 0.6477, respectively.
