This extended RISM ground truth has been prepared in Spring 2016 using the then available RISM Linked Open Data. The ground truth and its preparation are described in detail in:
Jelmer van Nuss, Geert Jan Giezeman, Frans Wiering (2017). Melody retrieval and composer attribution using sequence alignment on RISM incipits. TENOR Conference, A Coruña, Spain, 25-26 May 2017. link to full paper.This publication also employs the ground truth for a comparative evaluation of 27 different melodic similarity methods, all of them based on sequence alignment. This evaluation was also presented at the Music Encoding Conference in Tours (May 2017), see abstract and slides.
The original round truth was ranked by similarity to the query. Items below a certain threshold were considered not to be relevant. For the new items a much simpler evaluation method was used, resulting in a binary score: relevant or not relevant. There is thus no ranking amongst the new items. For each of the 11 queries, we list old and new items separately, but ranking is not preserved for the old items. Therefore, for each query, a link is provided to the ranked version of the 2005 ground truth.
The ground truth is also available as a CSV file. Values are separated by colons; the meaning of the three columns is as follows:
| 000.109.446-1.1.1 | |
| 000.111.706-1.1.1 | |
| 230.005.489-1.1.1 | |
| 240.001.397-1.13.1 | |
| 270.000.749-1.19.1 | |
| 400.187.395-1.1.1 | |
| 600.053.699-1.1.2 | |
| 600.066.687-1.1.2 | |
| 700.006.825-1.1.1 | |
| 700.007.852-1.1.1 | |
| 705.001.214-1.1.1 | |