Master thesis : Automatic Quality Assessment of Digital Comics Using Machine Learning

Abstract

The objective of this master's thesis is to develop an algorithm capable of assessing the quality of digital comics using machine learning. This was conducted to enable \textit{Stripik}, a mobile application that allows users to write comics with a generative tool and also read the content created, to recommend works that meet a certain quality threshold. The learning process must be based on comics that have been manually assessed beforehand.

The development of such an algorithm first involves selecting and managing the data to be used during the learning phase. Afterwards, the data were divided into three main categories : statistical data, textual data and graphical data. For each category, data were specifically processed, according to the corresponding category. A predictive model then had to be chosen to determine which one would be implemented in the final algorithm. This choice was made sequentially, beginning with the optimization of each model individually on a training set, followed by a comparison of the optimized models using a validation set. The resulting model, trained on the selected and processed data, ultimately constitutes the algorithm to be developed in this work.

This thesis concludes with encouraging results, although they are not yet fully satisfactory in terms of performance. Future perspectives for improving the assessment algorithm and its outcomes are proposed.