A key improvement of the new rating mechanism is to replicate a more correct preference pertinent to popularity, pricing policy and slot effect based on exponential decay model for online users. This paper research how the web music distributor ought to set its ranking coverage to maximise the worth of on-line music rating service. However, previous approaches typically ignore constraints between slot value illustration and associated slot description representation in the latent area and lack sufficient model robustness. Extensive experiments and analyses on the lightweight fashions present that our proposed strategies achieve considerably larger scores and considerably enhance the robustness of each intent detection and slot filling. Unlike typical dialog fashions that rely on big, complicated neural network architectures and huge-scale pre-educated Transformers to attain state-of-the-art results, our technique achieves comparable results to BERT and even outperforms its smaller variant DistilBERT on conversational slot extraction tasks. Still, even a slight improvement may be price the price.