Final Activity Report Summary - CONTACT (Consumer-oriented access to media content)
The CONTACT project had the overall goal to train four young researchers from the wide area of human-media interaction towards a research PhD degree. The project was based on the cooperation between researchers from diverse scientific disciplines, ranging from machine learning to human factors.
The individual research topics were selected to address the social and technological problems faced by consumers due to vastly increasing amounts of audio and video content. Technological developments led to dramatically increased storage capacities of multimedia devices used at home, as well as to large increases in the bandwidth of internet connections and broadcast channels. These developments posed great problems for the consumer in terms of selecting, archiving and retrieving content. They also posed a challenge for the consumer electronics' industry in that, so far, no solutions were realised for consumer devices to overcome these problems. Therefore, each of the four subprojects had the following major outcomes.
Fellow 1 (Novello) focussed on the problem of music similarity using songs from a wide range of western popular music. Music similarity was a concept addressed in the fields of musicology as well as music information retrieval. This project made important contributions to the methodology of large-scale experiments on perceived music similarity by optimising the trade-off between a number of different sound examples and experimentation time using balanced incomplete block designs. Within the experimental and selected-song context, statistically significant evidence was found for a hierarchical salience of the control variables used in the stimulus selection on participants' rankings, implying that genre had greater influence than tempo which was more important than timbre.
Fellow 2 (Campanella) investigated how automation and user control could be balanced in a home video editing system. The goal of this research consisted in finding an easy to use, effective and efficient solution for home video editing. To accomplish it, a semi-automatic application for editing videos was developed, following a user-centred approach, and several iterations of design and evaluation were realised. The project also reviewed the literature relevant to home video editing and performed a user-need analysis regarding home video editing.
Fellow 3 (Dunn) concentrated on how personality characteristics could be used to predict music preferences. The goal of this project sought to improve our understanding of the relation between individuals' personality and the music that they liked, which could be used to help improve current music recommender technologies. To improve on previous models relating music preferences and personality, this project incorporated music stimuli to build its own model of music preferences, which was then related to extracted music features. Furthermore, personality was measured in greater detail than previously to produce algorithms that more accurately predicted music preferences. Thus, this project provided a comprehensive and detailed model of music preferences related to personality which made important contributions to an overall understanding of music preferences and its relation to personality, as well as provided accurate algorithms able to predict music preferences based on personality information.
In the project of fellow 4 (Tiemann), it was investigated how recommendation algorithms could be tailored for the music domain and, in particular, for selecting music for personalised music radio channels. The aim of this project was to improve the quality of the created personalised music radio channels and to reduce user effort while using a system providing the aforementioned functionality. Several novel algorithm variants were devised in order to achieve these goals. An extensive user evaluation was carried out to evaluate several aspects that might influence the perceived quality of generated personal music radio channels.
The individual research topics were selected to address the social and technological problems faced by consumers due to vastly increasing amounts of audio and video content. Technological developments led to dramatically increased storage capacities of multimedia devices used at home, as well as to large increases in the bandwidth of internet connections and broadcast channels. These developments posed great problems for the consumer in terms of selecting, archiving and retrieving content. They also posed a challenge for the consumer electronics' industry in that, so far, no solutions were realised for consumer devices to overcome these problems. Therefore, each of the four subprojects had the following major outcomes.
Fellow 1 (Novello) focussed on the problem of music similarity using songs from a wide range of western popular music. Music similarity was a concept addressed in the fields of musicology as well as music information retrieval. This project made important contributions to the methodology of large-scale experiments on perceived music similarity by optimising the trade-off between a number of different sound examples and experimentation time using balanced incomplete block designs. Within the experimental and selected-song context, statistically significant evidence was found for a hierarchical salience of the control variables used in the stimulus selection on participants' rankings, implying that genre had greater influence than tempo which was more important than timbre.
Fellow 2 (Campanella) investigated how automation and user control could be balanced in a home video editing system. The goal of this research consisted in finding an easy to use, effective and efficient solution for home video editing. To accomplish it, a semi-automatic application for editing videos was developed, following a user-centred approach, and several iterations of design and evaluation were realised. The project also reviewed the literature relevant to home video editing and performed a user-need analysis regarding home video editing.
Fellow 3 (Dunn) concentrated on how personality characteristics could be used to predict music preferences. The goal of this project sought to improve our understanding of the relation between individuals' personality and the music that they liked, which could be used to help improve current music recommender technologies. To improve on previous models relating music preferences and personality, this project incorporated music stimuli to build its own model of music preferences, which was then related to extracted music features. Furthermore, personality was measured in greater detail than previously to produce algorithms that more accurately predicted music preferences. Thus, this project provided a comprehensive and detailed model of music preferences related to personality which made important contributions to an overall understanding of music preferences and its relation to personality, as well as provided accurate algorithms able to predict music preferences based on personality information.
In the project of fellow 4 (Tiemann), it was investigated how recommendation algorithms could be tailored for the music domain and, in particular, for selecting music for personalised music radio channels. The aim of this project was to improve the quality of the created personalised music radio channels and to reduce user effort while using a system providing the aforementioned functionality. Several novel algorithm variants were devised in order to achieve these goals. An extensive user evaluation was carried out to evaluate several aspects that might influence the perceived quality of generated personal music radio channels.