As the variety of products and services continues to increase, recommender systems play a critical role in assisting customers by presenting products or services that are likely to be of interest to them. In the era of big data, there is an abundance of data available from various sources, encompassing different modalities. In addition to user rating information on products, other relevant data sources can include social networks, knowledge bases, product descriptions and reviews, as well as contextual and temporal information. Even cross-domain and cross-site information can prove useful. In this talk, our focus is on utilizing large multi-modal models through broad learning to fuse multiple information sources of diverse modalities and perform synergistic deep recommendation tasks across these fused sources in a unified manner. We examine the various heterogeneous information sources and explore ways to enhance the effectiveness of recommendation systems by leveraging large multimodal models to harness the power of deep and broad learning.