Initial upload: ERKG deduplication results (single table).

README.md

@@ -0,0 +1,27 @@
+---
+pretty_name: UkrainIA canonicalization
+task_categories:
+  - Open Knowledge Graph Canonicalization
+tags:
+  - entity-resolution
+  - knowledge-graphs
+  - deduplication
+license: mit
+---
+
+# Entity Duplicate Candidates
+
+A dataset of candidate duplicate entity pairs with features and labels, sorted by `original`.
+
+## Columns
+- **original** *(string)*: Source entity surface form.
+- **candidate** *(string)*: Candidate entity surface form.
+- **levenshtein** *(float/int)*: Levenshtein distance (or normalized score if applicable).
+- **normalized** *(string/bool)*: Normalized surface form or flag (as in your source).
+- **hashtag** *(string/bool)*: Hashtag feature (as in your source).
+- **cossim_score** *(float)*: Cosine similarity between sentence embeddings.
+- **len_shortest_path** *(int)*: Length of shortest path between entities in the KG (if any).
+- **nb_shortest_path** *(int)*: Number of shortest paths.
+- **is_duplicate** *(bool)*: Ground-truth label (True if the pair is a duplicate).
+
+```