Intellectual Technology of Making Quality Posters

Abstract

Poster design is the process of reciprocally assembling layout elements, taking into account the balance, shape, scale, proportionality, color, and other factors that determine perception, to achieve the purpose of the publication. In this process, there are both creative side and a technical side. In this work, we will separately consider the task of poster layout, that is, publications that contain the basic graphical and textual information about an event or event. The characteristics of the layouts of posters from scientific materials, documentation for graphic packages are studied, and the practical experience of workers in this area is systematized. The characteristics on the basis of which it is proposed to build an intelligent technology for the layout of quality posters are determined. Using professional design Internet resources Behance, DeviantArt and Dribbble, which aggregate the best solutions in the field of poster layout, including posters, a dataset was created using the modern online service Google Sheets. According to the target characteristic, the ratio of the number of likes to the number of views of the poster is accepted. A new intelligent design technology of quality posters is proposed, which at the setup stage builds a model based on decision tree with hyperparameters optimization using the dataset of high-quality and low-quality posters as an example; classes and conditions of parameters that correspond to them are determined from this decision tree. The user sets a number of parameters that cannot be changed (restriction), then the technology filters out all classes that definitely do not have such parameters, and among those that remain, they select the best class for the poster layout that will simultaneously satisfy these restrictions and will maximize the feedback of the professional community in as the ratio of the number of likes to the number of views of the poster. Based on the selected class, a list of recommendations for designers is formed. Recommendations are formed using conditional nodes leading from the root node to the leaves necessary for the task. Based on these recommendations, the layout of the poster is synthesized and the estimated ratio of the number of likes to the number of views is immediately reported to the user. The proposed technology has been successfully tested on a number of posters in Ukrainian and English. It will simultaneously improve the quality of poster layouts and the time for their design.