Danlp

:sparkles: Features
- Add BERT HateSpeech model (detection of offensive language & classification of hate speech)
- Add ELECTRA Offensive model (detection of offensive language)

:sparkles: Features
- Add DaNED dataset
- Add DaWikiNED dataset
- Add XLM-R NED model

:warning: Upgrades
- upgraded package versions for dependencies:
- gensim == 3.8.3
- flair==0.9

:sparkles: Features
- Add DaUnimorph dataset
- Add AngryTweets dataset
- Add BERT Offensive model

:warning: Upgrades
- default python version for development is now 3.8
- upgraded package versions for dependencies:
- torch==1.7.1
- flair==0.8
- transformers==4.3.3
- allennlp==2.5.0

:sparkles: Features
- Add DKHate dataset for hate speech detection
- Add a feature for finding `domains` with DanNet
- Add a model for coreference resolution

:bug: Bugfixes
- 108 Fix download of datasets/models to a custom directory
- Fix warning unclosed file in Dacoref
- Fix UserWarning "Maximum length exceeded" with BERT

:sparkles: Features
- Add DaCoref dataset for coreference resolutions
- Add the DanNet wordnet to make it easy to find e.g. synonyms
- Add a pre-trained Bert model to use as embeddings, mask word predictions and next sentence predictions
- Update BERT ner predict methods to provide the tags combined

:bug: Bugfixes
- 80 Fix bug with missing argument in loading bert model
- Update requirements to fix mistakes in Flair


v.0.0.10
:sparkles: New features
- Add spaCy sentiment model with the tags "positive, neutral and negative" trained using hard distil of the "Tone BERT" model
- Add spaCy function for noun phrase chunking using the dependency model in spacy
- Add predict proba method in BERT models for tone and emotions

:mag: Minor updates
- Made it posible to choose any static embeddings form DaNLP with the Flair embeddings
- Allowed for longer sequences than 512 in BERT sentiment models, and add warning for doing so

:hotel: Storage
- Updatet to permanent URL for storage of models and dataset to http://danlp-downloads.alexandra.dk

v.0.0.9
:hotel: changes in storage
- The hosting storage is removed from Amazon to local

:sparkles: Features
- Add a first version of BERT Tone model which inlc a model for polarity (positive, neutral, negative) and a model for analytic (objective, subjective)
- Add a small dataset annotated from Twitter with sentiment, and intergation to download tweets based on ID from Twitter
- Add a small dataset from Europarl annotated with sentiment