update model card

README.md

@@ -10,16 +10,16 @@ licenses:
 multilinguality:
 - monolingual
 size_categories:
-- 10K<n<100K
+- 1K<n<10K
 source_datasets:
-- extended|other-reuters-corpus
+- 
 task_categories:
 - token-classification
 task_ids:
 - named-entity-recognition
-- part-of-speech-tagging
-paperswithcode_id: conll-2003
-pretty_name: CoNLL-2003
+- named-entity-linking
+paperswithcode_id: ipm-nel
+pretty_name: IPM NEL (Derczynski)
 ---
 
 # Dataset Card for "conll2003"
@@ -50,29 +50,33 @@ pretty_name: CoNLL-2003
 
 ## Dataset Description
 
-- **Homepage:** [https://www.aclweb.org/anthology/W03-0419/](https://www.aclweb.org/anthology/W03-0419/)
+- **Homepage:** []()
 - **Repository:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
-- **Paper:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
-- **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
-- **Size of downloaded dataset files:** 4.63 MB
-- **Size of the generated dataset:** 9.78 MB
-- **Total amount of disk used:** 14.41 MB
+- **Paper:** [http://www.derczynski.com/papers/ner_single.pdf](http://www.derczynski.com/papers/ner_single.pdf)
+- **Point of Contact:** [Leon Derczynski](https://github.com/leondz)
+- **Size of downloaded dataset files:** 120 KB
+- **Size of the generated dataset:** 
+- **Total amount of disk used:** 
 
 ### Dataset Summary
 
-The shared task of CoNLL-2003 concerns language-independent named entity recognition. We will concentrate on
-four types of named entities: persons, locations, organizations and names of miscellaneous entities that do
-not belong to the previous three groups.
+This data is for the task of named entity recognition and linking/disambiguation over tweets. It comprises
+the addition of an entity URI layer on top of an NER-annotated tweet dataset. The task is to detect entities
+and then provide a correct link to them in DBpedia, thus disambiguating otherwise ambiguous entity surface
+forms; for example, this means linking "Paris" to the correct instance of a city named that (e.g. Paris, 
+France vs. Paris, Texas).
 
-The CoNLL-2003 shared task data files contain four columns separated by a single space. Each word has been put on
-a separate line and there is an empty line after each sentence. The first item on each line is a word, the second
-a part-of-speech (POS) tag, the third a syntactic chunk tag and the fourth the named entity tag. The chunk tags
-and the named entity tags have the format I-TYPE which means that the word is inside a phrase of type TYPE. Only
-if two phrases of the same type immediately follow each other, the first word of the second phrase will have tag
-B-TYPE to show that it starts a new phrase. A word with tag O is not part of a phrase. Note the dataset uses IOB2
-tagging scheme, whereas the original dataset uses IOB1.
+The data concentrates on ten types of named entities: company, facility, geographic location, movie, musical
+artist, person, product, sports team, TV show, and other.
+
+The file is tab separated, in CoNLL format, with line breaks between tweets.
+
+* Data preserves the tokenisation used in the Ritter datasets.
+* PoS labels are not present for all tweets, but where they could be found in the Ritter data, they're given. 
+* In cases where a URI could not be agreed, or was not present in
+
+DBpedia, there is a NIL. See the paper for a full description of the methodology.
 
-For more details see https://www.clips.uantwerpen.be/conll2003/ner/ and https://www.aclweb.org/anthology/W03-0419
 
 ### Supported Tasks and Leaderboards
 
@@ -86,148 +90,119 @@ For more details see https://www.clips.uantwerpen.be/conll2003/ner/ and https://
 
 ### Data Instances
 
-#### conll2003
+#### ipm_nel
 
-- **Size of downloaded dataset files:** 4.63 MB
-- **Size of the generated dataset:** 9.78 MB
-- **Total amount of disk used:** 14.41 MB
+- **Size of downloaded dataset files:** 120 KB
+- **Size of the generated dataset:** 
+- **Total amount of disk used:**
 
 An example of 'train' looks as follows.
 
 ```
 {
-    "chunk_tags": [11, 12, 12, 21, 13, 11, 11, 21, 13, 11, 12, 13, 11, 21, 22, 11, 12, 17, 11, 21, 17, 11, 12, 12, 21, 22, 22, 13, 11, 0],
     "id": "0",
     "ner_tags": [0, 3, 4, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
-    "pos_tags": [12, 22, 22, 38, 15, 22, 28, 38, 15, 16, 21, 35, 24, 35, 37, 16, 21, 15, 24, 41, 15, 16, 21, 21, 20, 37, 40, 35, 21, 7],
+    "uris": [12, 22, 22, 38, 15, 22, 28, 38, 15, 16, 21, 35, 24, 35, 37, 16, 21, 15, 24, 41, 15, 16, 21, 21, 20, 37, 40, 35, 21, 7],
     "tokens": ["The", "European", "Commission", "said", "on", "Thursday", "it", "disagreed", "with", "German", "advice", "to", "consumers", "to", "shun", "British", "lamb", "until", "scientists", "determine", "whether", "mad", "cow", "disease", "can", "be", "transmitted", "to", "sheep", "."]
 }
 ```
 
-The original data files have `-DOCSTART-` lines used to separate documents, but these lines are removed here.
-Indeed `-DOCSTART-` is a special line that acts as a boundary between two different documents, and it is filtered out in this implementation.
 
 ### Data Fields
 
-The data fields are the same among all splits.
-
-#### conll2003
 - `id`: a `string` feature.
 - `tokens`: a `list` of `string` features.
-- `pos_tags`: a `list` of classification labels (`int`). Full tagset with indices:
-
-```python
-{'"': 0, "''": 1, '#': 2, '$': 3, '(': 4, ')': 5, ',': 6, '.': 7, ':': 8, '``': 9, 'CC': 10, 'CD': 11, 'DT': 12,
- 'EX': 13, 'FW': 14, 'IN': 15, 'JJ': 16, 'JJR': 17, 'JJS': 18, 'LS': 19, 'MD': 20, 'NN': 21, 'NNP': 22, 'NNPS': 23,
- 'NNS': 24, 'NN|SYM': 25, 'PDT': 26, 'POS': 27, 'PRP': 28, 'PRP$': 29, 'RB': 30, 'RBR': 31, 'RBS': 32, 'RP': 33,
- 'SYM': 34, 'TO': 35, 'UH': 36, 'VB': 37, 'VBD': 38, 'VBG': 39, 'VBN': 40, 'VBP': 41, 'VBZ': 42, 'WDT': 43,
- 'WP': 44, 'WP$': 45, 'WRB': 46}
-```
-
-- `chunk_tags`: a `list` of classification labels (`int`). Full tagset with indices:
-
-```python
-{'O': 0, 'B-ADJP': 1, 'I-ADJP': 2, 'B-ADVP': 3, 'I-ADVP': 4, 'B-CONJP': 5, 'I-CONJP': 6, 'B-INTJ': 7, 'I-INTJ': 8,
- 'B-LST': 9, 'I-LST': 10, 'B-NP': 11, 'I-NP': 12, 'B-PP': 13, 'I-PP': 14, 'B-PRT': 15, 'I-PRT': 16, 'B-SBAR': 17,
- 'I-SBAR': 18, 'B-UCP': 19, 'I-UCP': 20, 'B-VP': 21, 'I-VP': 22}
-```
-
 - `ner_tags`: a `list` of classification labels (`int`). Full tagset with indices:
-
-```python
-{'O': 0, 'B-PER': 1, 'I-PER': 2, 'B-ORG': 3, 'I-ORG': 4, 'B-LOC': 5, 'I-LOC': 6, 'B-MISC': 7, 'I-MISC': 8}
-```
+- `uris`: a `list` of URIs (`string`) that disambiguate entities. Set to `NIL` when an entity has no DBpedia entry, or blank for outside-of-entity tokens.
 
 ### Data Splits
 
-|  name   |train|validation|test|
-|---------|----:|---------:|---:|
-|conll2003|14041|      3250|3453|
+|  name   |train|
+|---------|----:|
+|ipm_nel||
 
 ## Dataset Creation
 
 ### Curation Rationale
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+To gather a social media benchmark for named entity linking that is sufficiently different from newswire data.
 
 ### Source Data
 
 #### Initial Data Collection and Normalization
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The data is partly harvested from that distributed by [Ritter / Named Entity Recognition in Tweets: An Experimental Study](https://aclanthology.org/D11-1141/),
+and partly taken from Twitter by the authors.
 
 #### Who are the source language producers?
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+English-speaking Twitter users, between October 2011 and September 2013
 
 ### Annotations
 
 #### Annotation process
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The authors were allocated documents and marked them for named entities (where these were not already present) and then attempted to find
+the best-fitting DBpedia entry for each entity found. Each entity mention was labelled by a random set of three volunteers. 
+The annotation task was mediated using Crowdflower (Biewald, 2012). Our interface design was to show each volunteer the text of the tweet, any URL links contained
+therein, and a set of candidate targets from DBpedia. The volunteers were encouraged to click on the URL links from the
+tweet, to gain addition context and thus ensure that the correct DBpedia URI is chosen by them. Candidate entities were
+shown in random order, using the text from the corresponding DBpedia abstracts (where available) or the actual DBpedia
+URI otherwise. In addition, the options ‘‘none of the above’’, ‘‘not an entity’’ and ‘‘cannot decide’’ were added, to allow the
+volunteers to indicate that this entity mention has no corresponding DBpedia URI (none of the above), the highlighted text
+is not an entity, or that the tweet text (and any links, if available) did not provide sufficient information to reliably disambiguate the entity mention.
 
 #### Who are the annotators?
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The annotators are 10 volunteer NLP researchers, from the authors and the authors' institutions.
 
 ### Personal and Sensitive Information
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The data was public at the time of collection. User names are preserved.
 
 ## Considerations for Using the Data
 
 ### Social Impact of Dataset
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+There's a risk of user-deleted content being in this data. The data has NOT been vetted for any content, so there's a risk of [harmful text](https://arxiv.org/abs/2204.14256) content.
 
 ### Discussion of Biases
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The data is annotated by NLP researchers; we know that this group has high agreement but low recall on English twitter text [C16-1111](https://aclanthology.org/C16-1111/).
 
 ### Other Known Limitations
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The above limitations apply.
 
 ## Additional Information
 
 ### Dataset Curators
 
-[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
+The dataset is curated by the paper's authors.
 
 ### Licensing Information
 
-From the [CoNLL2003 shared task](https://www.clips.uantwerpen.be/conll2003/ner/) page:
+The authors distribute this data under Creative Commons attribution license, CC-BY 4.0. You must
+acknowledge the author if you use this data, but apart from that, you're quite
+free to do most things. See https://creativecommons.org/licenses/by/4.0/legalcode .
 
-> The English data is a collection of news wire articles from the Reuters Corpus. The annotation has been done by people of the University of Antwerp. Because of copyright reasons we only make available the annotations. In order to build the complete data sets you will need access to the Reuters Corpus. It can be obtained for research purposes without any charge from NIST.
-
-The copyrights are defined below, from the [Reuters Corpus page](https://trec.nist.gov/data/reuters/reuters.html):
-
-> The stories in the Reuters Corpus are under the copyright of Reuters Ltd and/or Thomson Reuters, and their use is governed by the following agreements:
->
-> [Organizational agreement](https://trec.nist.gov/data/reuters/org_appl_reuters_v4.html)
->
-> This agreement must be signed by the person responsible for the data at your organization, and sent to NIST.
->
-> [Individual agreement](https://trec.nist.gov/data/reuters/ind_appl_reuters_v4.html)
->
-> This agreement must be signed by all researchers using the Reuters Corpus at your organization, and kept on file at your organization.
 
 ### Citation Information
 
 ```
-@inproceedings{tjong-kim-sang-de-meulder-2003-introduction,
-    title = "Introduction to the {C}o{NLL}-2003 Shared Task: Language-Independent Named Entity Recognition",
-    author = "Tjong Kim Sang, Erik F.  and
-      De Meulder, Fien",
-    booktitle = "Proceedings of the Seventh Conference on Natural Language Learning at {HLT}-{NAACL} 2003",
-    year = "2003",
-    url = "https://www.aclweb.org/anthology/W03-0419",
-    pages = "142--147",
+@article{derczynski2015analysis,
+  title={Analysis of named entity recognition and linking for tweets},
+  author={Derczynski, Leon and Maynard, Diana and Rizzo, Giuseppe and Van Erp, Marieke and Gorrell, Genevieve and Troncy, Rapha{\"e}l and Petrak, Johann and Bontcheva, Kalina},
+  journal={Information Processing \& Management},
+  volume={51},
+  number={2},
+  pages={32--49},
+  year={2015},
+  publisher={Elsevier}
 }
-
 ```
 
 
 ### Contributions
 
-Thanks to [@jplu](https://github.com/jplu), [@vblagoje](https://github.com/vblagoje), [@lhoestq](https://github.com/lhoestq) for adding this dataset.
+Author-added dataset [@leondz](https://github.com/leondz)