An approach to a Multi-Category Recommendation System using Machine Learning : With the caveat of having limited knowledge in related areas

Abstract: Machine learning is one of many buzz words in todays tech-world. Huge company resources are allocated to the field in order to discover its potential. Everything from cameras to cars tries to use this technology. However, the question is if developers with little experience in the field can use this technology in a useful way? And how would one proceed with that? This thesis tries to answer these questions by having two third year undergraduate students attempt to implement a multi-category movie recommendation system using machine learning. With the important caveat of neither student having any previous knowledge in machine learning, recommendation systems nor the chosen programming language (Python). An extensive background study was performed in order to obtain knowledge in the different areas. Recommendation systems often use either collaborative, or content-based filtering, or a hybrid of the two. Machine learning uses different algorithms, a selection of these where studied together with available frameworks. In order to implement and design a system, a data-set from MovieLens, containing ratings of movies, and the framework SciKit-learn was used. The implementation tried to use genres in order to give movie recommendations. This was done in two systems, one where every user got a genre-weight and the other system used Nearest Neighbor in order to use the collaborative filtering approach. However, due to the limited time the implementation was not implemented for multiple categories, but the results showed that this should be highly applicable using the proposed design. The thesis showed that even two third year undergraduate students with no prior knowledge in the areas could make use of machine learning in an system implementation. The results of the project was presented in two different parts; Firstly, the system implementation result showed that the accuracy metric was not at a satisfactory level. Even though the concept of using genres as a metric for giving recommendations worked, it was seemingly to simple and broad. Secondly, the project result showed that the majority of the time was spent on the preliminary work and the system implementation. Finally the economical cost of the project was presented.