HMI Talk: Maya Ramani and Rob Schwartz
Can Patterns in Mystery Novels Be Predicted?
This research involves the analysis of mystery novels and stories using Natural Language Processing (NLP) and network analysis to accurately determine the identity of the guilty suspect. We are creating a “detective algorithm” that combines various NLP computing techniques to parse and process the narrative, and output the predicted identity of the guilty suspect along with a prediction score. First, the sentiment analysis of the text surrounding each character mention was collected to see how that character is discussed by the author and other characters and to ascertain if the context surrounding the mention of their name may have a unique pattern or connotation. Next, network analysis was used to provide a representation of the interconnectedness of the characters in the story arc, based on how many contexts they were involved together in, whether that be a mention or real time conversation. Network patterns across narratives are compared to identify patterns in how the guilty suspect is interconnected with the other characters in the story. We are currently investigating our algorithm on Sir Arthur Conan Doyle’s Return of Sherlock Holmes collection and hope to apply it to deduce the guilty party in other stories. Our project helps strengthen the bridge between the fields of computing and literature, as it explores the capabilities of an algorithm to be fine-tuned to predict human behavior in stories. Creating a “detective algorithm” also provides more insight into how predictable the mystery genre is, creating a new lens through which literature can be studied.