For the upcoming Berlin MIR Meetup, we are happy to announce the following talk.
Martin Herzog (Audio Communication Group, TU Berlin):
A Knowledge-Based Recommender System for Music Branding
In this talk, we present the development and initial evaluation of a knowledge-based music branding recommender system. Its requirements significantly differ from traditional music recommenders: In our case, the perceived semantic expression of music titles is of main interest since it has to meet marketing intentions, whereas consumers’ personal preferences or emotional responses are of rather minor importance. In order to address the ‘semantic gap’ between audio signal analysis and complex brand identities to be communicated by music to heterogeneous target groups, our system combines machine learning of music branding expert knowledge with audio signal analysis toolboxes’ output and population-representative ground truth data gathered by multinational online listening experiments. A combination of random forest regression and traditional regression models is able to predict perceived brand image dimensions with an accuracy of up to 80%, which appears sound, given that our recommender does not rely on any form of social or expert tagging when new music titles are entered into the pool. By using adaptive sub-models for different target groups, we further approach the challenge of a socially heterogeneous understanding of music. A first operational scenario will be a commercial software tool that automatically generates playlists for the point of sale on basis of any given brand identity and target group profile.
Martin Herzog is a research associate at the Audio Communication Group, TU Berlin, and is part of the project ABC_DJ (Artist-to-Business-to-Business-to-Consumer Audio Branding System). His PhD research focusses on predicting musical meaning from high-level music features.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement №[masked]