Data plays a big role in operation and organization of our society. However, in the last decade we are facing an issue of excessive amount of data. One way to solve this problem are recommender systems. Recommender systems are systems that try to recommend the most interesting and relevant data to an individual from a mass of data. It is an approach commonly used on the web. The aim of the diploma thesis is to describe the field of recommender system and, using the case of Twitch platform, to determine whether it is possible to recommend content to an individual based on the content consumed by users similar to the individual. Twitch is a live streaming web platform. Due to the enormous amount of video content on Twitch, an individual is not able to choose the most suitable content for himself. We created four models for recommendation using the nearest neighbours method. Two of the models are based on finding the nearest neighbour users according to the Jaccard similarity and the cosine similarity and another two based on finding the nearest channel with the same two similarities. The most accurate model was recommendation based on nearest neighbour users with cosine similarity. We compared our model with recommendations by chance and popularity. The final finding was that the recommendation by comparing the users and channel was more accurate than the recommendations based on chance and popularity. We confirmed the hypothesis that it is possible to recommend content to a user based on the choice of content of similar users.
|
