# ConceptNet Commonsense relation prediction tasks derived from ConceptNet 5.7 (English subset), encoded as ILP problems. Each task targets one ConceptNet relation type; the background knowledge provides the other nine relations restricted to the same entity set, enabling reasoning across relation types. Entities are filtered to those appearing in at least 3 triples; each relation is capped at 2000 positive examples. Negatives are sampled from the cross-product of entities appearing in background knowledge. **Tasks:** 10 (`r001_at_location` through `r010_used_for`) | Task | Relation | Description | |------|----------|-------------| | `r001_at_location` | `AtLocation` | Typical location where a concept can be found | | `r002_capable_of` | `CapableOf` | Actions a concept is capable of performing | | `r003_causes` | `Causes` | Events or states that a concept causes | | `r004_has_a` | `HasA` | Things a concept possesses or contains | | `r005_has_prerequisite` | `HasPrerequisite` | Prerequisites required before an action | | `r006_has_property` | `HasProperty` | Properties or attributes of a concept | | `r007_is_a` | `IsA` | Hypernym or category membership | | `r008_part_of` | `PartOf` | Meronymy: concept is part of a larger entity | | `r009_receives_action` | `ReceivesAction` | Actions that can be done to a concept | | `r010_used_for` | `UsedFor` | Purpose or function of a concept | **Head relation:** `f(Concept1, Concept2)` — the target ConceptNet relation holds between `Concept1` and `Concept2`. **Background relations:** For each task, the other nine ConceptNet relations serve as background. Each is a binary relation over concepts: | Relation | Arity | Description | |----------|-------|-------------| | `at_location(C1, C2)` | 2 | `C1` is typically found at location `C2` | | `capable_of(C1, C2)` | 2 | `C1` is capable of `C2` | | `causes(C1, C2)` | 2 | `C1` causes `C2` | | `has_a(C1, C2)` | 2 | `C1` has `C2` | | `has_prerequisite(C1, C2)` | 2 | `C1` has prerequisite `C2` | | `has_property(C1, C2)` | 2 | `C1` has property `C2` | | `is_a(C1, C2)` | 2 | `C1` is a kind of `C2` | | `part_of(C1, C2)` | 2 | `C1` is part of `C2` | | `receives_action(C1, C2)` | 2 | `C1` receives action `C2` | | `used_for(C1, C2)` | 2 | `C1` is used for `C2` | (The relation being predicted is excluded from the background for that task.) ## Task statistics | task_name | num_bk_facts | num_bk_relations | num_pos_examples | num_neg_examples | |---|---:|---:|---:|---:| | `r001_at_location` | 16094 | 9 | 1997 | 5991 | | `r002_capable_of` | 16094 | 9 | 2000 | 6000 | | `r003_causes` | 16094 | 9 | 2000 | 6000 | | `r004_has_a` | 16841 | 9 | 1253 | 3759 | | `r005_has_prerequisite` | 16094 | 9 | 2000 | 6000 | | `r006_has_property` | 16094 | 9 | 2000 | 6000 | | `r007_is_a` | 16094 | 9 | 1932 | 5796 | | `r008_part_of` | 16094 | 9 | 1910 | 5730 | | `r009_receives_action` | 17253 | 9 | 841 | 2523 | | `r010_used_for` | 16094 | 9 | 1991 | 5973 | **Source:** Speer, R., Chin, J., & Havasi, C. (2017). ConceptNet 5.5: An open multilingual graph of general knowledge. *Proceedings of the AAAI Conference on Artificial Intelligence*, 31(1). Data retrieved from the ConceptNet S3 release: https://s3.amazonaws.com/conceptnet/downloads/2019/edges/conceptnet-assertions-5.7.0.csv.gz