Co-authorship networks have been used to study collaboration patterns in various fields, evaluate researchers and recommend policies. In their simplest form they are constructed by considering authors to be network nodes connected to each other if they published a paper together. We propose to further explore the same data by constructing a different network, in which nodes are articles linked to one another if they have a common author. For papers published in the fields of computer science and mathematics with affiliations to Romanian institutions, we show that this type of network reveals patterns of collaborative behavior and offers new insights about practices in the field. We find that the proposed networks are smaller and denser than the co-authorship networks, have a better defined community structure, and directly represent the results of collaborative endeavors by focusing on the actual outcome, i.e., published papers.