Add missing transitive links and adapt to highlight multi-source use-…

…case (#9)

* Add missing links and make adaptations

* Remove cluster id

* Add cluster files for binary

* Update README

* Add new fold data and adapt loading from release

* Adapt tests!

* Update README

* Test downloading

* Adapt changelog

CHANGELOG.md

@@ -7,9 +7,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.1.0] - 2024-03-13
+
+### Fixed
+
+- Remove intra-dataset links from ent_links
+
+### Added
+
+- intra_ent_links to ERDataset
+- cluster files
+- multi_source_cluster file
+
 ## [1.0.2] - 2023-03-24
 
-### Changed 
+### Changed
 
 - Loosen dependency restrictions to allow every Python >=3.7.1
 
@@ -24,6 +36,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 Make packagable and enable loading of datasets
 
+[1.1.0]: https://github.com/ScaDS/MovieGraphBenchmark/releases/tag/v1.0.2
 [1.0.2]: https://github.com/ScaDS/MovieGraphBenchmark/releases/tag/v1.0.2
 [1.0.1]: https://github.com/ScaDS/MovieGraphBenchmark/releases/tag/v1.0.1
 [1.0.0]: https://github.com/ScaDS/MovieGraphBenchmark/releases/tag/v1.0.0

README.md

@@ -1,3 +1,10 @@
+# !! Update 2024-02-24 (fixed in 1.1.0) !!
+We found that `ent_links` in some cases contained intra-dataset links, which is not immediately noticable by the user.
+Another round of clerical review was performed, transitive links, which were previously missed are added and the `ent_links` files now only contain entity links _between_ the datasets. The `721_5fold` directories have been adapted accordingly.
+The intra-dataset links are now in `{dataset_name}_intra_ent_links` for each of the three datasets.
+What might also not be immediately obvious is that this dataset can be used as multi-source entity resolution task.
+We therefore provide a `multi_source_cluster` file with each line consisting of a cluster id and comma-seperated cluster members of the three datasets, which can also include multiple entries for a single dataset.
+
 # Dataset License
 Due to licensing we are not allowed to distribute the IMDB datasets (more info on their license can be found [here](https://help.imdb.com/article/imdb/general-information/can-i-use-imdb-data-in-my-software/G5JTRESSHJBBHTGX?pf_rd_m=A2FGELUUNOQJNL&pf_rd_p=3aefe545-f8d3-4562-976a-e5eb47d1bb18&pf_rd_r=2TNAA9FRS3TJWM3AEQ2X&pf_rd_s=center-1&pf_rd_t=60601&pf_rd_i=interfaces&ref_=fea_mn_lk1#))
 What we can do is let you build the IMDB side of the entity resolution datasets yourself. Please be aware, that the mentioned license applies to the IMDB data you produce.
@@ -36,12 +43,27 @@ print(ds.rel_triples_2)
 print(ds.ent_links)
 for fold in ds.folds:
     print(fold)
+
+# the intra-dataset links are stored as tuple of dataframes
+print(ds.intra_ent_links[0])
+print(ds.intra_ent_links[1])
 ```
 
+Alternatively this dataset (among others) is also available in [`sylloge`](https://github.com/dobraczka/sylloge).
+
 # Dataset structure
 There are 3 entity resolution tasks in this repository: imdb-tmdb, imdb-tvdb, tmdb-tvdb, all contained in the `data` folder. 
-The data structure follows the structure used in [OpenEA](https://github.com/nju-websoft/OpenEA).
-Each folder contains the information of the knowledge graphs (`attr_triples_*`,`rel_triples_*`) and the gold standard of entity links (`ent_links`). The triples are labeled with `1` and `2` where e.g. for imdb-tmdb `1` refers to imdb and `2` to tmdb. The folder 721_5fold contains pre-split entity link folds with 70-20-10 ratio for testing, training, validation.
+The data structure mainly follows the structure used in [OpenEA](https://github.com/nju-websoft/OpenEA).
+Each folder contains the information of the knowledge graphs (`attr_triples_*`,`rel_triples_*`) and the gold standard of entity links between the datasets(`ent_links`). The triples are labeled with `1` and `2` where e.g. for imdb-tmdb `1` refers to imdb and `2` to tmdb. The folder 721_5fold contains pre-split entity link folds with 70-20-10 ratio for testing, training, validation.
+Furthermore, there exists a file for each dataset with intra-dataset links called `*_intra_ent_links`.
+For the binary cases each dataset has a `cluster` file in the respective folder. Each line here is a cluster with comma-seperated members of the cluster. This includes intra- and inter-dataset links.
+For the multi-source setting, you can use the `multi_source_cluster` file in the `data` folder.
+Using [`sylloge`](https://github.com/dobraczka/sylloge) you can also easily load this dataset as a multi-source task:
+
+```
+from sylloge import MovieGraphBenchmark
+ds = MovieGraphBenchmark(graph_pair='multi')
+```
 
 # Citing
 This dataset was first presented in this paper:

data/imdb_intra_ent_links

@@ -0,0 +1 @@
+https://www.scads.de/movieBenchmark/resource/IMDB/nm3361496	https://www.scads.de/movieBenchmark/resource/IMDB/nm3361266