Persistent Homology of Topic Networks for the Prediction of Reader Curiosity