Reconfigurable intelligent surfaces (RISs) are software-controlled passive devices to reflect incoming signals from the source (𝑆) to destination (𝐷), just like a relay (𝑅) with optimum signal strength, improving the performance of wireless communication networks. The configurable nature of the RIS can provide network designers the flexibility to use in a stand-alone or cooperative configuration with many advantages over conventional networks. In this paper, two new deep neural networks (DNN) assisted cooperative RIS models, namely DNN 𝑅 -CRIS and DNN 𝑅,𝐷 -CRIS, are proposed for cooperative communications. In these two models, the potential of RIS deployment as a relaying element in a next-generation cooperative network is investigated using deep learning (DL) techniques as a tool for optimizing the RIS. To reduce maximum likelihood (ML) complexity at the 𝐷, unlike the DNN 𝑅 -CRIS, in the DNN 𝑅,𝐷 -CRIS model, a new DNN based symbol detection method is presented for the same network model. For a different number of relays and receiver configurations, bit error rate (BER) performance results of the proposed DNN 𝑅 -CRIS, DNN 𝑅,𝐷 -CRIS models and traditional cooperative RIS (CRIS) scheme (without DNN) are presented for a multi-relay cooperative communication scenario with path loss effects. It's revealed that the proposed DNN based two models show promising results in terms of BER even in high noise environments with low system complexity.