In English

Spelrekommendationssystem med maskininlärning - Kollaborativ filtrering

Robert Felczak ; Rami Jabor
Göteborg : Chalmers tekniska högskola, 2018. 57 s.
[Examensarbete på grundnivå]

In an age of information overflow, the demand for filtering increases every day. This is also the case for videogames, where new games are released everyday. There are many filtering techniques and one of them is Collaborative Filtering which this thesis focuses on. The purpose is to explore the possibilities of collaborative filtering for the company 8 Dudes in A Garage ABs Recommendation System. The tests performed consist of training a amount of recommendation system models with the Turi Graphlab Framework and evaluate how good the recommendations are for each model. The Datasets that the tests are performed on are taken from the company and Amazons Videogame Category. The goal was to also test the optimal model on the company's website with the community users, but it was not able to be done due to the projects timelimit. The results show that the ItemSimilarity model gave the best values in evaluations, even though the results gave low values on average for all models. The report goes through the problems these results indicate and other problems that were not shown in the results. Suggested solutions that the company could implement in future endeavours with their recommendation system are included.

In conclusion the report shows:

1. In what ways the company would benefit from Collaborative Filtering.

2. That with the Framework Graphlab these changes wouldn’t be hard to implement to the company website.

3. That a hybridsystem would be a solution towards an optimal Recommendation System.

Nyckelord: Videogames, Collaborative Filtering, Recommendation Systems, Machine Learning,



Publikationen registrerades 2018-06-28. Den ändrades senast 2018-06-28

CPL ID: 255312

Detta är en tjänst från Chalmers bibliotek