In this research, we aim to evaluate the readability of a document on the basis of eye movements. Since it is not realistic to ask participants to read documents with eye tracking devices every time, we develop a system which synthesizes artificial eye movements on the document and utilizes them for the readability measurement. The main purposes in FY 2019 were improving the performance of the gaze generation (which was proposed in 2018) and implementing the method as a working demo.
For the performance improvement, we received feedback regarding experimental design and conducted a new study with a precise setup. The data samples were collected from nine participants reading 15 documents by using an eye tracker. Our new approach, a random forest regression with lexical and syntactical features, could predict the fixation duration per word with a mean regression score (R2) of 0.757 for all the documents.
By using the experimental results, we have implemented a system that creates a fixation duration heatmap on an unknown document to help the document more readable. The system and the generation results were demonstrated at IUI 2020. An additional study was conducted to evaluate how the gaze based features are effective in assessing the readability of documents.