Alan Höng

Hello,
I am a young, driven and open-minded computer scientist with a passion for machine learning and pattern recognition. Mathematical models, I find highly interesting as well as implementing and evaluating them.

This site holds my personal and academic projects. Feel free to contact me. I am always open to discuss collaborations or future oportunities.

This work tries to approach music recommendation in a content based way. Most today`s recommender systems use user data on such as listening history and likes to recommend new music to users. Many of todays music platforms such as Soundcloud, Mixcloud or Youtube grow rapidly and contain many new music pieces of lesser known artists. As traditional recommender systems require a great amount of user data to predict good recommendations, these approaches might perform poorly on these newer platforms, as lesser known tracks won't be recommended at all or only appear to very few users. This work presents a recommender system which requires very few user data to make recommendations. By analysing audio data and clustering the analysed data using a neural network, recommendations for each user are generated. At the end we evaluate the achieved results using offline evaluation techniques.

A prototype trying to change the way how to choose and experience musical events. Often information about events are scattered across the web, hidden on social media websites or only available from a selected site which is used regionally. This makes it hard for the individual to find events suited to his personal tase. Through scraping the web and accumulating and untangling all necessary information like music, genres, lineups, dates, times and locations an interactive and medial rich envionment is created. Users can listen to the artists profile without leaving the site. An automatic playlist is genreated and presented to the user according to his preferences.