Abstract: To address the issue of data sparsity in Web API recommendation, this paper proposes a Web API recommendation model called CoNetLGN based on the Light Graph Convolution Network. CoNetLGN leverages the collaboration among Web APIs invoked within the same Mashup to construct an API collaboration graph as auxiliary information, thereby enhancing recommendation performance. In CoNetLGN, user and Web API representations are propagated in the user-API interaction graph through a light graphs convolution layer. Simultaneously, API representations are propagated in the API collaboration graph. Paper designs a graph fusion operation to fusion API representation from both graphs during propagation. Finally, model weight and combine the learned representations at each layer. Experimental results on the Programmable Web dataset demonstrates that proposed CoNetLGN outperforms three representative collaborative filtering methods in recommending Web APIs to users.