Analysis of interaction patterns is one of the most important indicators of quality of learning in educational web forums. Social network analysis (SNA) is gradually assuming importance in the study of interaction patterns as it focuses on the analysis of the interrelationships between individuals, thus providing a holistic perspective on group performance. However, most of the studies that use SNA in computer-supported collaborative learning scenarios derive their data from server log files, on the assumption that this data source reflects the way people really interacted online. This study, the purpose of which is to better understand the communication flows that really occur among users, challenges these assumptions through an experimental study that makes a comparison between the structural method normally used to detect the posting addressees and an approach enriched with semantic coding. Results show that this new coding schema, if compared with traditional structural coding, detected a greater number of addressees, thus allowing a greater number of postings to be included in an SNA adjacency matrix.