- Some words in paper are cut because of typesetting, and these words are represented as prehalf-backhalf.
- These words are taken into consideration when pre-preprocessing???
- If not, this will be a problem when look up in Embedding table.
- prediction of relations and corefs' entities aren't in the prediction of entities. But ground truth is different: each entity of relation and coref is definately in the ground truth entity list.
- annotation guideline
Greedy
:Spans can be embedded in longer spans, only if the shorter span is involved in a relation.- If generic phrases are not involved in a relation, do not tag them
- ??? involved in a relation ??? in pdf1 Figure 1 sentence #3 approach-Generic
- This can be done by post-delete operation.
- ??? some terms are difficult to recognize. ???
- "Multi-...KGC" mentioned that the dataset is annotated by experts.
- Expression following the phrase "such as" has three components or more.
- all of them are involved in the relation of conjunction
- Entity Literature
- Relation Evaluate-for in pdf1 Figure 3 sentence #6
- Relation Proposed-by, Synonym-of
- BRAT: a annotation tool. paper
- pandoc: a
tex to word
tool. download- usage:
pandoc -s paper.tex -o paper.docx
Entity | Relation |
---|---|
1. Task | 1. Used-for |
2. Method | 2. Feature-of |
3. Evaluation Metric | 3. Hyponym-of |
4. Material | 4. Part-of |
5. Other Scientific Terms | 5. Compare |
6. Generic | 6. Conjunction |
7. Literature | 7. Evaluate-for |
8. Proposed-by | |
9. Synonym-of |