英文摘要 |
From an educational perspective, it is important to provide students of different grades with reading material of appropriate difficulty for better learning retention. To deal with this problem, it is common practice to use a set of handcrafted features, for example, hard word rate or word count, to distinguish articles into different readability levels. However, these traditional readability features are often too shallow to represent deeper semantic or syntactic structures of the articles. In view of this, we present a modeling approach that leverages a recurrent neural network to hierarchically encode both the semantic or syntactic structures of a given article for better readability classification. Furthermore, we also seek to make extra use of traditional handcrafted feature as side information to further boost the performance. |