You Don’t Have Time to Read This: An Exploration of Document Reading Time Prediction

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1789–1794 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 1789 You Don’t Have Time to Read This: An Exploration of Document Reading Time Prediction Orion Weller,1 Jordan Hildebrandt,1 Ilya Reznik,2 Chris Challis,2 E. Shannon Tass,1 Quinn O. Snell,1 Kevin Seppi1 1Brigham Young University 2Adobe orionw@byu.edu Abstract Predicting reading time has been a subject of much previous work, focusing on how differ- ent words affect human processing, measured by reading time. However, previous work has dealt with a limited number of partici- pants as well as word level only predictions (i.e. predicting the time to read a single word). We seek to extend these works by examining whether or not document level predictions are effective, given additional information such as subject matter, font characteristics, and read- ability metrics. We perform a novel experi- ment to examine how different features of text contribute to the time it takes to read, distribut- ing and collecting data from over a thousand participants. We then employ a large num- ber of machine learning methods to predict a user’s reading time. We find that despite ex- tensive research showing that word level read- ing time can be most effectively predicted by neural networks, larger scale text can be eas- ily and most accurately predicted by one factor, the number of words. 1 Introduction Understanding how we read and process text has proven a large area of both cognitive science and natural language processing (NLP) research (Graesser et al., 1980; Liversedge et al., 1998; Frank et al., 2013a; Busjahn et al., 2014; Weller and Seppi, 2019, 2020). Online content providers and consumers are also interested in this research; in the increasingly busy world of today, consumers lack the time to read long articles, prompting con- tent creators to aim for specific reading lengths. Many providers1 have even examined traffic pat- terns in order to determine the ideal content length, with the general consensus finding 3-7 minutes of Work done as part of a capstone course with Adobe 1Medium’s study can be found here. content optimal. Thus, having established the op- timal content length, article writers now face the next hurdle: when has their post reached the ideal length? A news article about last night’s football game may be easier to read than a technical post about NLP. Perhaps the font type or size influences the consumer’s comprehension, slowing down the reading process. There are many factors, both tex- tual and stylistic, that quickly come to mind when considering the potential reading time of an article. Although there has been an extensive body of work on reading time prediction applied to single words (Frank, 2017; Willems et al., 2015; Shain, 2019; van Schijndel and Linzen, 2018), to the best of our knowledge there has been no research into understanding these effects on document sized text. In this paper, we seek to address this area by build- ing models to predict, understand, and interpret …

Content truncated to 500 words