Table of Contents
Quick start guide
- 1. Add a new experimental set
- 2. Click into the experiment name
- 3. Add a new questionnaire giving an ID to the participant
- 4. Fill the questionnaire
- 5. Return to the main page
- 6. Click on the analyze link
- 7. Create a model based on the preselected layers
- 8. Give the number of desired Likert steps
- 9. Wait while your data is processed
- 10. View the results
The functionality of this version of AMP (Artist-base Music Profiler) is divided in three modules: Forager, Modeler, and Analyzer. Explained next:
The forager's function is to gather data to be used as input to the system. In order to construct an individual's profile, AMP requires the names of three liked and three disliked artists. To this end, it is provided a questionnaire with auto-complete functionality (powered by the Echo Nest), to avoid spelling errors.
Ideally, you would use it to input data, but batch forms of doing it are under development and will be available very soon.
AMP's output is defined with this module. It is the equivalent to a list of genres to be rated by participants to assess their musical preferences. The main difference is that AMP's modeler offers apart from genres, a large set of target terms. The set of terms can be a preset featuring some of the most popular terms, or it can be constructed manually by choosing them from a list of near eight thousand tags.
While having such variety maximizes flexibility, it may become very soon unpractical if it distracts from the main focus i.e., constructing an output model for the AMP. For this reason the modeler provides visualizations, to aid in the construction of a semantically balanced model. These visualizations offer an introspection on how the terms are related, each term is defined as a vector with a depth of near a quarter million songs.
To aid in the construction of AMP's output model, the system offers two visualizations that inform about the distance and relations between the selected terms. The reliability of a test, can be determined a posteriori with a meta-analysis of subject's answers. AMP's visualizer offers the possibility of an improvement in reliability by offering a unique perspective on the semantic content. For instance:
This similarity matrix indicates whether a combination of terms is suitable to be included in the same model. For instance, if two terms are to closely related (red), it may be advisable to choose only one of them, otherwise, the whole model would be unbalanced. In other words, if two or more terms are semantically to close, they are probably measuring the same thing.
This spring graph, offers another perspective on the same data. It is composed of two visual channels: the diameter of the circles represent their popularity, and the connections between the circles show the strength of the relation. Note that if no line is drawn between circles, it does not imply that a relation is non existent, but that the relation is weak in this particular context.
This module's purpose is to provide the main functionality of transforming liked and disliked artist names into rated terms. AMP takes the list of liked and disliked artists expressed by each individual and use them to retrieve tags from Last.fm, these tags are then filtered and weighted to emulate rates for the semantic model.
Thus, the analyzer is conceived as the space where you can combine individual's responses with semantic models containing terms of relevance for your research.