<![CDATA[Novels are a well-known and frequently read form of fictional works. Novels typically have over 100 pages and are available in a wide variety of genres. However, most novels have simple covers and only containt a brief title and narrative synopsis. It is difficult to determine the genre of a novel due to the lack of genre information on the cover. By these issues, a new classification method was developed by utilizing novel synopsis data and a multi-class Support Vector Machine (SVM) algorithm with a One-Against-All strategy. The TF-IDF and Chi-square approaches are also used for term weighting and features selection. To achieve the highest classification accuracy, this work implements two SVM kernels: the linear kernel and the gaussian kernel. 240 summary texts were used as training and testing dataset, grouped into four different genre categories: horror, romance, science fiction, and history. During tests, the kernel type, Chi-square threshold value, and sequential training parameters were changed to achieve the best classification accuracy result. Based on the test results, the highest classification accuracy value of 94.58% is achieved at the Chi-square threshold of 80%, SVM with a linear kernel, sequential training parameter with lambda (λ) = 0,5, gamma (γ) = 0,05, complexity (C) = 1, epsilon (ε) = 0,0001, and the maximum number of iterations is 100.]]>
Copyrights © 2023
