Help

1. Introduction

This is the help page for the SoundSuggest application. SoundSuggest is a visual explanation system. Such a system uses visual elements to explain and provide insight into a complex concept. The concept in this case is the Last.fm recommender.

The goal of this explanation system is to provide insight into the process of calculating suggestions by the Last.fm recommender. By providing an approximation of the underlying concept of collaborative filtering-based recommendation, the user can make an estimation of how the system calculated this recommendation.

2. The Last.fm recommender

2.1 What is a recommender system?

A recommender system automatically generates item recommendations for a certain user. A recommender system may use various approaches, i.e., algorithms, to calculate these recommendations based on the user’s preferences and/or behaviour.

2.2 What are neighbours?

The Last.fm recommender is based on an algorithm called collaborative filtering (CF). In this approach the system looks for similar users of the active user, i.e., you, based on similarities between artist libraries. A neighbour is then a user with a high similarity score to the active user.

3. Installation

The extension becomes active on the page http://www.last.fm/home/recs if you are logged in with your Last.fm account. Note that you’ll need to authenticate the application the first time you visit the page.

To remove the selection, simply click on the selected item again, or use the clear selection button in the menu bar.

4. Functionality

The navigation through the graph is quite straightforward: simply hover or select item on the graph. The graph will highlight edges and nodes that occur in the same profile. The relevant profile(s) in the list of neighbours will be highlighted as well.

Similarly, when hovering or selecting a neighbour, the items that are in his/her profile will be highlighted on the graph.

A demo can be found here.

4.2 Settings

The application settings can be adjusted on in the settings menu, which is available under the settings button. Note that you can only change settings once the visualization has finished loading.

There are two types of settings you can change: settings related to the data, and settings related to the visual elements in the visualization. Under the first category, three options are available:

1. Neighbours : the amount of neighbours involved in the visualization. More neighbours will likely produce a better overview of how certain recommendations are related to you.
2. Recommendations : the amount of recommendations you would like to see in the graph.
3. Top artists : the amount of top artists involved in the graph. Note that increasing this number may give better insight in how certain recommendations are related to your artists through your neighbours. If a neighbour has an item that is in your profile, and one that is not, you will see a connection between the items in your profile, and those particular suggestions.

With the second type of settings, you can change the colours related to the available interactions, items, and users in the visualization:

1. Active user : the colour of your items in the graph and user list.
2. Mouseover interaction : the colour of items and users when hovering over them.
3. Click interaction : the colour of items and users when selecting them.

Three colours are available: blue, green and red.

Another option for to alter the visualization layout is to change the edge-bundling strength by changing the tension parameter. A tension of $0$ will result in straight lines between the nodes, a tension of $1$ will cause all the edge to attract each other.

A demo can be found here. The images below shows the effect of the tension parameter on the visualization.

To add an artist recommendation to your library, select one of the artists in the graph. Next click the button View Full Profile and click Add to library. The item is added to your library. The visualization will not reload until you refresh the page.

Undoing this operation is done in exactly the same manner as you would normally remove items from your library. Simply navigate to your library, and remove the relevant artist(s) from it.

5. Known issues

Loading the data may sometimes take quite a while if you selected a high number for one or more data settings. To estimate how long it may take to load all the data, for $N$ neighbours, $R$ recommendations and $T$ top artists, the total time is approximately $2Nt(R + T)$, with $t$ the time for processing one request to the Last.fm API.

If your Last.fm account has barely any or no listening history, you may not have any ‘top artists’ in your profile. If this is the case, the visualization will not load.