A Neural Network based Approach to Compute Text Readability Using Visual Features
Sangita Saha, Apurbalal Senapati, Ranjan Maity · 2023
Text elements are an important part of any interface. Readability of text elements present in an interface can be determined using language properties (text length, average word lengths etc.). Except these, the visual properties (color, font size etc.) of textual elements may have an effect on readability. In this work, we proposed a Neural Network based computational model for text readability using six visual features of text elements - background color (Red, Green and Blue), text color (Red, Green, Blue), font size, font style, line spacing, and word spacing. By varying these features, thirty-five samples were designed, and an empirical study was conducted by thirty-five participants. Finally, empirical data were used to develop our proposed model. In order to develop the proposed model, we considered a sequential neural network-based approach. Five dense layers with Relu activation functions were used to construct our model. Experimental results suggest a loss of only 0.0092 during the training process, while a mean error of 0.0796 was observed during testing.