제품 개선 및 신제품 개발을 위한 시장분석 방법 : 머신러닝 방법론을 중심으로

Online reviews provide valuable insights for product improvement and new product development as they contain both strengths and weaknesses of the product, as well as information about competing products. However, most methods proposed in previous research have shown low accuracy or limitations in scalability. Additionally, there is limited research on the use of machine learning algorithms with excellent predictive performance for product performance prediction through online review analysis, and there is insufficient empirical research applying explainable machine learning techniques in the context of online review analysis. Therefore, this study proposes a framework that extracts a product attribute system from online review text reflecting product attributes based on natural language processing technology and predicts and explains performance through machine learning models. To validate the framework, 'running shoes' products and online reviews were collected from Coupang and the product's attributes were measured using Word2Vec, and the attribute hierarchy was classified through hierarchical clustering. Using review attributes and main keywords, product attributes and sentiment scores were predicted at the sentence level using LSTM, and scores for each top attribute were calculated for each product. The relationship between product attribute scores and product performance (i.e., overall satisfaction [rating], popularity [number of online reviews]) was learned as a predictive model, and the importance of product attributes and the relationship between product attributes and product performance were analyzed at the entire product group level. This study provides theoretical and practical implications by proposing a framework that can obtain implications for product improvement and new product development by analyzing online reviews with natural language processing technology and machine learning algorithms.