Feature Word Extraction Method for Book Difficulty

With the rapid development of the E-Commerce market in recent years, product selection has become increasingly difficult. Therefore, various studies have been conducted to assist in the selection, but few have focused on the level of difficulty. However, if the level of difficulty related to content is not considered when selecting books, it is impossible to provide products that meet user needs. Therefore, this study used product reviews to identify the characteristic words for the difficulty level. As an extraction method, a distributed representation of words was obtained based on Word2Vec, and clustering was performed using DBSCAN with two-dimensional compression by t-SNE to provide stable clustering while considering the meanings of words. In addition, the experimental results show that it is possible to extract not only feature words directly related to the difficulty level but also words that indirectly affect the evaluation of the difficulty level.