In our paper we would like to make a cross-disciplinary leap and use the tools of network theory to understand and explore narrative structure in literary fiction, an approach that is still underestimated. However, the systems in fiction are sensitive to readers subjectivity and attention must to be paid to different methods of extracting networks. The project aims at investigating into different ways social interactions are read in texts by comparing networks produced by automated algorithms-natural language processing with those created by surveying more subjective human responses. Conversation networks from fiction have been already extracted by scientists, but the more general framework surrounding these interactions was missing. We propose several NLP methods for detecting interactions and test them against a range of human perceptions. In doing so, we have pointed to some limitations of using network analysis to test literary theory, e.g. interaction, which corresponds to the plot, does not form climax.