Unrestricted accessibility to the Web has brought increases in quantity and quality of movies that are available to the audiences all over the world. This is why movie-related information services, such as movie retrieval and movie recommender systems, have developed. Meanwhile, the WWW also has captured and accumulated abundant Word-of-Mouth about movies such as user reviews, user comments, and ‘tweets’. This paper aims to explore the potential of the Word-of-Mouth, as cultural metadata, to be used as movie features that represent the various characteristics of movies.
In order to identify the association between cultural metadata and movies, we use a periphrastic way of examining correlations between the similarity of cultural metadata and that of movie characteristics. To do this, we collected movie metadata - cultural, cinematic, editorial, rating, and IMDb recommendations. Then, we selected 96 movies as a probe set considering their representativeness and influence in the movie industry. For each probe movie, lists of similar movies were made on the basis of every metadata type. We then evaluated which metadata type represented which movie characteristics the best.
User reviews are indeed cultural metadata in that they provide colorful qualitative discourses on movies, and are generated culturally and spontaneously by the general people. It is found that user reviews represent the various characteristics, namely cinematic, affective, preference-related and people-related, of movies well. A rating predictor using user reviews also showed a remarkable performance. In conclusion, user reviews has the potential to be used as movie features for various purposes.
The present work have performed an in-depth study to characterize user reviews as cultural metadata. Moreover, F-Emotions that is the original data of affective annotations was built, which can contribute not only to the field of computer science but also to the field of social science such as communica...