Integrating utility and interest to recommend healthcare interventions for online users

There is a critical need for online healthcare communities to recommend interventions that satisfy users' interests and improve their health. This paper proposes a recommendation algorithm based on representation learning of user interests and healthcare needs using a multi-task learning architecture. The algorithm learns user interest representations through an attention mechanism and assesses the expected treatment effects of recommended interventions via a deep neural network. An auxiliary loss function evaluates the intervention's utility for the target user, combining this with the user's interest representation to generate recommendations. Offline experiments using authentic data from online health communities substantiate the model's efficacy, exploring hyper-parameter influences and conducting ablation experiments to assess model components. The results demonstrate that the proposed algorithm outperforms baseline methods, effectively recommending interventions that help users improve their healthcare conditions and encourage their continued participation in online communities.