Abstract: Recently, there has been much work on the use of Networked Point Processes (NPPs) to extract the latent network structure of timestamp data. Several models currently exist to capture implicit interactions in hospital visits, blog posts, e-mail messages, among others. The problem is that evaluating these solutions is not a trivial task. First, the methods have only been evaluated in a few datasets by a limited number of metrics. Second, and even worse, the evaluation metrics are often unsuitable for the typically sparse networks, which consequently lead to inconclusive results. To provide the community with a rigorous benchmark, in this paper we propose an empirical evaluation framework of NPP models in the task of network extraction. We reevaluate several models of the literature using our framework and compare the results to two null models designed for this task. In our discussion, we point out when some methods should be used depending on the expected efficacy, execution time, or dataset properties. Overall, we find that only three models show consistent significant results in real-world data.