Emotional classification and visualization of movies based on their IMDb reviews
| DOI | https://doi.org/10.1108/IDD-05-2017-0045 |
| Pages | 149-158 |
| Date | 21 August 2017 |
| Published date | 21 August 2017 |
| Author | Kamil Topal,Gultekin Ozsoyoglu |
| Subject Matter | Library & information science,Library & information services,Lending,Document delivery,Collection building & management,Stock revision,Consortia |
Emotional classification and visualization of
movies based on their IMDb reviews
Kamil Topal
Center for Proteomic and Bioinformatics, Department of Nutrition, Department of Electrical Engineering and Computer Science,
Case Western Reserve University, Cleveland, OH, USA, and
Gultekin Ozsoyoglu
Department of Electrical Engineering and Computer Science, Case Western Reserve University, Cleveland, Ohio, USA
Abstract
Purpose – The purpose of this study is to detect these reviews’ complex emotions, visualize and analyze them. Movie reviewers’ moviescores and
reviews can be analyzed with respect to their emotion content, aggregated and projected onto a movie, resulting in an emotion map for a movie.
It is then possible for a moviegoer to choose a movie, not only on the basis of movie scores and reviews, but also on the basis of aggregated
emotional outcome of a movie as reflected by its emotion map displaying certain emotion map patterns desirable for the moviegoer.
Design/methodology/approach – The authors use the hourglass of emotion model to find the emotional scores of words of a review, then they use
singular value decomposition to reduce the data dimension into singular scores. Once, they have the emotional scores of reviews, the authors cluster them
by using k-means algorithm to find similar emotional levels of movies. Finally, the authors use heat maps to visualize four dimensions in a figure.
Findings – The authors are able to find the emotional levels of movie reviews, represent them in single scores and visualize them. The authors look
the similarities and dissimilarities of movies based on their genre, ranking and emotional statuses. They also find the closest emotion levels of movies
to a given movie.
Originality/value – The authors detect complex emotions from the text and simply visualize them.
Keywords Information retrieval, Data visualization, Emotion analysis, Emotion clustering, Emotion detection, Recommended systems
Paper type Research paper
1. Introduction
It is common knowledge that, usually, moviegoers, called users
or reviewers in the rest of this paper, utilize movie ratings and
reviews in selecting their next movie to see/watch. This is indeed
the case for the authors of this paper. And, unfortunately,
sometimes movie reviews and ratings do not help users make the
right choices, as evidenced by their emotional feelings after
watching the movie. This is perhaps because users desire a
certain emotional state after watching a movie, which does not
match the emotions evoked by the selected movie. And, this is
indeed what happened to the second author of this paper, after
going over IMDb (2016a) ratings and (some) reviews, and
picking, rather quickly and unfortunately, to watch the extremely
highly rated and indeed excellent Oscar-winning movie The
Revenant – simply because the emotions evoked by the movie on
him did not match his desired emotional state [as characterized
by Cambria et al.’s (2012) four emotional dimensions:
pleasantness, attention, sensitivity and aptitude].
Clearly, users’ reviews and ratings for a movie are strongly tied
to their emotions evoked by the movie. This paper argues that:
●it may be useful for users in their decision-making process
to choose the next movie to watch if a movie also comes
with an (expected) “emotion signature” or an “emotion
map”;
●toward goal 1, we can build automated software tools that
analyze movie reviews and ratings and provide an
emotional signature for a movie as evidenced from the
reviews and ratings of that movie; and
●clearly, once emotion maps for all movies are at hand, if a
user perhaps submits his/her desired emotion state and,
possibly, the desired genre of the movie, it is easy to build
a “personalized movie recommender system” for each
user.
This paper is a first step for goals 1 and 2 (but not for goal 3,
due to space limitations) and makes an attempt to analyze the
relationships between:
●users’ ratings for a movie and their emotions evoked by
watching the movie as evidenced in their movie reviews
and ratings; and
●movies’ genres and users’ emotional responses from their
movie reviews.
For our experimental evaluation, we used movie reviews from
IMDb, the world’s most popular content source for movie,
TV and celebrity content (IMDb, 2016a) with reviews for
more than 3.5 million movies. IMDb members provide
reviews and usefulness scores for other reviews. However, the
overall rating of a movie is calculated by IMDb’s own rating
The current issue and full text archive of this journal is available on
Emerald Insight at: www.emeraldinsight.com/2398-6247.htm
Information Discovery and Delivery
45/3 (2017) 149–158
© Emerald Publishing Limited [ISSN 2398-6247]
[DOI 10.1108/IDD-05-2017-0045]
Received 1 May 2017
Revised 5 August 2017
Accepted 8 August 2017
149
To continue reading
Request your trial