This page contains the data that is needed to reproduce the experiments and examples that have been reported in the following paper:
Joaquín Derrac and Steven Schockaert. Inducing semantic relations from conceptual spaces: a data-driven approach to commonsense reasoning, Artificial Intelligence, vol. 228, pages 66-94, 2015
The data consists of (spatial relations in) conceptual space representations in three domains: place types, movies, and wines. We refer to the aforementioned paper for more details about the data or the relevance of the considered relations.
We have made available the PPMI weighted feature vectors and the MDS representations of the place types (20D, 50D, 100D, 200D. In each case, the MDS representation contains one line for each considered place type, with each line encoding the coordinates of the corresponding point. A mapping is available which specifies the order in which the place types have been considered (i.e. which lines in the MDS files correspond to which place types).
Furthermore, we have made available the directions in the MDS space corresponding to each term which is sufficiently frequent in the original text corpus (20D, 50D, 100D, 200D), the chosen salient properties and corresponding clusters of terms (20D, 50D, 100D, 200D), the directions corresponding to these properties (20D, 50D, 100D, 200D), and the projections of the place type representations on the lines corresponding to these salient directions (20D, 50D, 100D, 200D).
Finally, we have also made available the classes that we have used for the experiments (Foursquare, GeoNames, OpenCYC).
We have made available the PPMI weighted feature vectors and the MDS representations of the movies (20D, 50D, 100D, 200D. In each case, the MDS representation contains one line for each considered movie, with each line encoding the coordinates of the corresponding point. A mapping is available which specifies the order in which the movies have been considered (i.e. which lines in the MDS files correspond to which movies).
Furthermore, we have made available the directions in the MDS space corresponding to each term which is sufficiently frequent in the original text corpus (20D, 50D, 100D, 200D), the chosen salient properties and corresponding clusters of terms (20D, 50D, 100D, 200D), the directions corresponding to these properties (20D, 50D, 100D, 200D), and the projections of the movie representations on the lines corresponding to these salient directions (20D, 50D, 100D, 200D).
Finally, we have also made available the classes that we have used for the experiments (genres, ratings, keywords).
We have made available the PPMI weighted feature vectors and the MDS representations of the wines (20D, 50D, 100D, 200D. In each case, the MDS representation contains one line for each considered wine, with each line encoding the coordinates of the corresponding point. A mapping is available which specifies the order in which the wines have been considered (i.e. which lines in the MDS files correspond to which wines).
Furthermore, we have made available the directions in the MDS space corresponding to each term which is sufficiently frequent in the original text corpus (20D, 50D, 100D, 200D), the chosen salient properties and corresponding clusters of terms (20D, 50D, 100D, 200D), the directions corresponding to these properties (20D, 50D, 100D, 200D), and the projections of the wine representations on the lines corresponding to these salient directions (20D, 50D, 100D, 200D).
Finally, we have also made available the classes that we have used for the experiments.