Few-Shot Entity Extraction

Utility scripts to extract entities from sentences in few-shot dataset using tagme, wikidata, kgsearch APIs.

Requirements

• python3
• tagme==0.1.3
• Tagme's api key: to use tagme api, you need to register an account here, and overwrite the token as TAGME_TOKEN in config.py.
• Google KG's api key: to use google KG-search api, you need to register an accound here, and overwrite tokens as KG_KEY in config.py.
• VPN is required if you are inside the GFW to access Google KG api. If you use proxies, overwrite the PROXIES in config.py, in the format: {'https': 'x.x.x.x:xxxx'}

Fewshot dataset

• Data source: https://github.com/zxlzr/FewShotNLP/tree/master/data

• Command: python tag_fewshot.py

• Output: original file named A.train(dev/test) -> A.train(dev/test).json

• Data format (after processing):

  • sentence: sentence.
  • class: type of sentence.
  • entities: list of entities' information
    • start_pos: begin position of entity in sentence.
    • end_pos: end position of entity in sentence.
    • score: probability of entity derived from tagme.
    • title: title of entity.
    • tagme_id: entity id according to tagme.
    • wiki_id: entity id according to wiki data.
    • kg_id: entity id according to google KG.

• Sample data:

[{"sentence": "it advertised not stock so i unable get it", "class": "-1", "entities": [{"start_pos": 3, "end_pos": 13, "score": 0.1549598127603531, "title": "Advertising", "tagme_id": 2861, "wiki_id": "Q37038", "kg_id": "kg:/m/019rl6"}, {"start_pos": 18, "end_pos": 23, "score": 0.09428920596837997, "title": "Stock and flow", "tagme_id": 888140, "wiki_id": "Q2714114", "kg_id": "kg:/m/01116q1q"}, {"start_pos": 36, "end_pos": 42, "score": 0.08518906682729721, "title": "Get It On (T. Rex song)", "tagme_id": 5590319, "wiki_id": "", "kg_id": "kg:/m/01972m"}]}]

Fewrel dataset

• Data source: https://github.com/thunlp/FewRel/tree/master/data

• Command: python tag_fewrel.py

• Output: train(val).json -> train(val)_entity.json

• Data format (after processing):

  • $relation_type$: name of each relation type, e.g., 'P931'.
    • tokens: words / segments split from sentences.
    • h: first item in relation.
    • t: second item in relation.
    • entities: list of entities' information.
      • index_begin: index of starting word in sentence.
      • index_end: index of ending word in sentence.
      • start_pos: begin position of entity in sentence.
      • end_pos: end position of entity in sentence.
      • score: probability of entity derived from tagme.
      • title: title of entity.
      • tagme_id: entity id according to tagme.
      • wiki_id: entity id according to wiki data.
      • kg_id: entity id according to google KG.

• Sample data:

{"P931": [{"tokens": ["Merpati", "flight", "106", "departed", "Jakarta", "(", "CGK", ")", "on", "a", "domestic", "flight", "to", "Tanjung", "Pandan", "(", "TJQ", ")", "."], "h": ["tjq", "Q1331049", [[16]]], "t": ["tanjung pandan", "Q3056359", [[13, 14]]], "entities": [{"start_pos": 0, "end_pos": 7, "score": 0.2455686628818512, "title": "Merpati Nusantara Airlines", "tagme_id": 1289337, "wiki_id": "Q619118", "kg_id": "", "index_begin": 0, "index_end": 1}, {"start_pos": 8, "end_pos": 14, "score": 0.16067400574684143, "title": "Flight International", "tagme_id": 3080620, "wiki_id": "Q1428742", "kg_id": "", "index_begin": 1, "index_end": 2}, {"start_pos": 19, "end_pos": 27, "score": 0.00844799354672432, "title": "The Departed", "tagme_id": 1349086, "wiki_id": "Q172975", "kg_id": "", "index_begin": 3, "index_end": 4}, {"start_pos": 28, "end_pos": 35, "score": 0.5443639755249023, "title": "Jakarta", "tagme_id": 16275, "wiki_id": "Q3630", "kg_id": "", "index_begin": 4, "index_end": 5}, {"start_pos": 38, "end_pos": 41, "score": 0.35298147797584534, "title": "Soekarno\u2013Hatta International Airport", "tagme_id": 1203751, "wiki_id": "Q749497", "kg_id": "", "index_begin": 6, "index_end": 7}, {"start_pos": 49, "end_pos": 64, "score": 0.3290479779243469, "title": "1982 Garuda Fokker F28 crash", "tagme_id": 15540624, "wiki_id": "Q4580452", "kg_id": "", "index_begin": 10, "index_end": 12}, {"start_pos": 68, "end_pos": 82, "score": 0.5, "title": "Tanjung Pandan", "tagme_id": 20621756, "wiki_id": "Q3056359", "kg_id": "", "index_begin": 13, "index_end": 15}]}]}

Note

• Max workers: specified in config.py. Using larger numbers would improve efficiency, but not all the time.
• Termination: The program terminates when all jobs are done. You may stop processing by typing ctrl+c whenever you want, but please make sure all workers exited and the program output Saved data., which means the extracted entities has been written into output files. You may restart and the program would process from where you stopped last time.