머신러닝 기법을 활용한 국내 영화의 흥행 요인에 관한 연구

Abstract

국문 요지

국내 영화시장은 세계에서 5번째로 큰 거대 시장이며, 영화 시장 규모는 지속적인 성장 추세를 보였다. 최근에는 OTT와 IPTV 등 미디어 시장의 성장으로 수익 채널이 다양화되었다. 또 영화 산업은 일반적으로 하이 리스크 하이 리턴이라는 특성을 가지고 있어 흥행 여부에 따라 실적의 격차가 매우 커질 수밖에 없다. 따라서 영화 흥행 예측에 관한 연구는 중요성은 매우 크고 필수적이다. 최근의 영화 산업의 트렌드는 수직적 통합을 통해 일부 대형작품들이 시장을 독점하고 있다. 본 연구에서는 영화 산업의 다양성 재고와 정확한 흥행 예측 연구를 위해 가치사슬에 따라 영화 흥행 요인을 분류한 뒤 제작단계에서 새로운 예측 변수를 제시하고 흥행 예측 모델을 제시한다. 특히, LSA 잠재의미분석과 TF-IDF 개념을 활용해 시나리오 텍스트 데이터를 분석하여 기존 연구에서는 잘 다루지 않았던 제작단계에서의 새롭고 다양한 변수를 추가하였다. 이 후 영화 산업의 가치사슬 단계별로 영화 흥행 예측 변수들을 나누고 순서형 로지스틱 회귀분석을 통해 변수의 통계적 유의성을 검증하고 머신러닝 모델들을 통해 비교한다.

이를 통해 시나리오를 활용한 영화 예측 연구와 영화 제작 과정에서의 예측 및 투자 결정을 지원하는데 기여할 것으로 기대된다.

keyword : 기계 학습(Machine Learning), 순서형 로지스틱 회귀분석(Ordinal Logistic Regression), LSA(Latent Sentiment Analysis), TF-IDF, 영화 흥행 예측(Predicting box office Performace)|Abstract

The Determinants of Box Office Performance in Korea with Machine Learning Technique

Korea film market is the fifth largest market in the world and the scale of the industry has steadily grown up. Recently, revenue streams have diversified due to the growth of the media market such as OTT and IPTV. In addition, as the characteristics of film industry are generally high-risk and high-return, a box office record of the movies has a major impact on difference in performance. Therefore, research on predicting box office success of movies is important and essential. Recent trends in film market are that a few of large films dominate the market through vertical integration.

This paper categorizes the determinants of box office performance according to the value chain and proposes new prediction variable in production section and forecasting model for box office success in order to reconsider the diversity of film industry and improve the quality of research on forecasting box office success.

In particular, after scenario text data were analysed by using latent sentiment analysis and the concepts of TF-IDF, new and diverse variables in the production phase that were not well addressed in previous studies were added. Then, the predictors of box office success are divided by value chain stage of the movie industry and are compared through machines learning models after validating the statistical significance of the variables through ordinal logistic regression.

This is expected to contribute to supporting research on film prediction using scenarios and decision making for investment in the process of film production.

keyword : Machine Learning, Ordinal Logistic Regression, LSA(Latent Sentiment Analysis), TF-IDF, Predicting box office Performace; Abstract

The Determinants of Box Office Performance in Korea with Machine Learning Technique

Korea film market is the fifth largest market in the world and the scale of the industry has steadily grown up. Recently, revenue streams have diversified due to the growth of the media market such as OTT and IPTV. In addition, as the characteristics of film industry are generally high-risk and high-return, a box office record of the movies has a major impact on difference in performance. Therefore, research on predicting box office success of movies is important and essential. Recent trends in film market are that a few of large films dominate the market through vertical integration.

This paper categorizes the determinants of box office performance according to the value chain and proposes new prediction variable in production section and forecasting model for box office success in order to reconsider the diversity of film industry and improve the quality of research on forecasting box office success.

In particular, after scenario text data were analysed by using latent sentiment analysis and the concepts of TF-IDF, new and diverse variables in the production phase that were not well addressed in previous studies were added. Then, the predictors of box office success are divided by value chain stage of the movie industry and are compared through machines learning models after validating the statistical significance of the variables through ordinal logistic regression.

This is expected to contribute to supporting research on film prediction using scenarios and decision making for investment in the process of film production.

keyword : Machine Learning, Ordinal Logistic Regression, LSA(Latent Sentiment Analysis), TF-IDF, Predicting box office Performace