Jason Zeng is a rising senior at Detroit Country Day School. He has self-studied Python, Java, and MatLab out of interest for machine learning. In school, he most enjoys math, physics, and literature. He actively participates in math competitions, having qualified for AIME three times. Additionally, he is a captain of his Science Olympiad team. As an athlete, he competes in tennis as well as track and field. He's also an accomplished pianist and has become a three-time Grand Prix winner of Rising Talents Music Fest and Golden Key Festival competitions. He most of all loves conversation, puzzles, anime, and music.
Project: Automatic Evaluation of the Relationships Between Characters in Plays
Given any play, use text analysis to evaluate the relationships between characters. Analyze the sentiment of sentences that connect two characters—as well as the frequency of these sentences—to determine the health of their relationship. Then, create a graph with vertices to represent the characters and edges to represent their relationships. The health of these relationships can be visualized with color and width. Further operations with graphing functions can provide information on local groups within the play. In addition, accounting for who speaks, visualized with directional edges, allows for nuanced relationships in which feelings are not mutual.
Summary of Results
Through experimentation, I created a formula to calculate relationship health. After visualizing the graphs for multiple plays, I added custom parameters, including a bias for negativity, to best represent each play. Also, I created a graph to visualize the entire social network within a play.
Setting the values for health parameters can be automated. The size of the text used for each sentiment calculation could be a function of sentence length. Additionally, one could handle the cases of having multiple recipients for a sentiment. The most difficult avenue, though perhaps the most fruitful, would be catching pronouns, aliases and titles and assigning them to the proper recipient. Once completed, a novel can be analyzed using a similar process to determine the speaker.