First up, long time Kodi fan (Xbox days) and moderators, I am unsure where to post to this so feel free to move it to the appropriate section.
Hi all, I am a physicist learning several topics in data science in my spare time. Recently, I have become interested in the different types of recommendation algorithms and after coming across a paper by Choi et.al., I decided to implement their algorithm which is based on genre correlations. A basic search tells me that their have been previous discussions (1, 2, 3 4) and several attempts (smart(ish) widgets, extended info, various skin helper scripts). So here goes another attempt.
The basic idea is to take your Kodi collection and use the genre information from each movie to compute how one genre is correlated to another. For example, one can argue that there is a strong correlation between Action and Sci-Fi genres, i.e. if a movie has Sci-Fi in it's genre set, it is likely that Action would also be present. Once this correlation matrix has been determined, the user then inputs a set of preferred genres and the algorithm outputs the recommendations. The advantage of this algorithm is that it uses information that is highly accurate since genres are usually set by the movie studios and available right from the start, you don't have to start rating some movies before seeing recommendations, although the algorithm does rely an average movie rating.
The implementation can be found in my repository below. Since I am not a programming guru, I have implemented a basic Python program to test the algorithm. It takes as input Kodi's MyVideos#.db database and generates recommendations based on one's preferred genres which the user inputs. Therefore, this is just a prototype toy model to demonstrate the algorithm. Being a scientist, I am unable to gauge the quality of the recommendations so I ask for your help (my opinion alone is not sufficient). If you are one of the advanced users that gets this to work, play around and please answer the poll. Have fun
.
Lastly, if anyone has any suggestions on how to generate a fake library of movies so I can provide some example datasets in the repo, please let me know.
Repository: https://github.com/NahsiN/MovieReco
Hi all, I am a physicist learning several topics in data science in my spare time. Recently, I have become interested in the different types of recommendation algorithms and after coming across a paper by Choi et.al., I decided to implement their algorithm which is based on genre correlations. A basic search tells me that their have been previous discussions (1, 2, 3 4) and several attempts (smart(ish) widgets, extended info, various skin helper scripts). So here goes another attempt.
The basic idea is to take your Kodi collection and use the genre information from each movie to compute how one genre is correlated to another. For example, one can argue that there is a strong correlation between Action and Sci-Fi genres, i.e. if a movie has Sci-Fi in it's genre set, it is likely that Action would also be present. Once this correlation matrix has been determined, the user then inputs a set of preferred genres and the algorithm outputs the recommendations. The advantage of this algorithm is that it uses information that is highly accurate since genres are usually set by the movie studios and available right from the start, you don't have to start rating some movies before seeing recommendations, although the algorithm does rely an average movie rating.
The implementation can be found in my repository below. Since I am not a programming guru, I have implemented a basic Python program to test the algorithm. It takes as input Kodi's MyVideos#.db database and generates recommendations based on one's preferred genres which the user inputs. Therefore, this is just a prototype toy model to demonstrate the algorithm. Being a scientist, I am unable to gauge the quality of the recommendations so I ask for your help (my opinion alone is not sufficient). If you are one of the advanced users that gets this to work, play around and please answer the poll. Have fun
.Lastly, if anyone has any suggestions on how to generate a fake library of movies so I can provide some example datasets in the repo, please let me know.
Repository: https://github.com/NahsiN/MovieReco
