運用感知形容詞中文字型推薦之設計與分析

Abstract

人工智慧設計在傳播與相關設計領域已經逐漸受到關注，利用人工智慧、機器學習、自然語言處理技術所建構的設計代理人與自動化高速設計系統開始在一些設計與電商平台上扮演非常重要的角色。字型推薦方法的開發可為未來自動化設計提供技術基礎，有助於即時化、客製化、低成本、極大量的新形態設計趨勢需求。本研究設計一個運用感知形容詞中文字型推薦方法，以詞嵌入技術配合所收集23個特定感知形容詞的隱喻詞彙開發出短文字語句之情感分類器，使得設計內容的文字能自動運算出最符合該輸入語句的情感表達結果，之後再利用感知形容詞與字型的對應關係，最後得出該文字語句字型運用的推薦建議，並評估此字型推薦方法的有效性。 研究結果顯示，運用感知形容詞中文字型推薦方法設計中，感知形容抽取演算法輸出結果與受測者對文句語意的理解較為相符，大部分中文字型指派與文字語意的匹配呈度高，但系統輸出的第一名字型與隨機字型之間對語意的匹配度影響較小。綜上所述，本研究所設計之推薦方法具有很高的可行性，但是仍有一定程度的改進空間。

Artificial intelligence design has gradually attracted attention in the field of graphic communication and related design. Design agents and automated high-speed design systems constructed using artificial intelligence, machine learning, and natural language processing technologies have begun to play a very important role on some design and e-commerce platforms. Development of font recommendation method can design automation to provide technical basis for the future. It helps to meet the needs of immediacy, customized, reduce the cost, and extremely large new trend of form design requirements. This paper will design a content-based font recommendation method for graphic design, and develop an emotional classifier for short text sentences using the word embedding technology and the collected metaphors corresponding to 23 specific emotional adjectives. The most suitable emotional representation of an input sentence can be obtained resulting from this classifier, and finally get its font recommendation according to the relationship of emotion adjectives and fonts. A prototype application system test platform will also be constructed in this project to evaluate the effectiveness of this font recommendation method, and explore the possibility of its application in plane communication and design. The research results have shown that in the design of recommendation methods for Chinese fonts based on emotional adjectives, the output results by selection algorithm of emotional adjectives are relatively consistent with the subjects’ understanding of the sentences. Most Chinese font assignment is highly matched with the meanings of the sentences. But the NO.1 font output by the system and the random fonts have little impact on the matching degree of the meanings of the sentences. To sum up, the recommendation methods designed by this study are highly feasible, but they still need some improvement.