This system aims to, given a short-text input, adapt and select Portuguese expressions (e.g., proverbs and movie titles) in order to approximate them to the input, enhancing relatedness, originality and, possibly, funniness.
Files needed to put in folder 'models_db':
- To put in folder 'bert_pretrained_models', download BERT MultiLingual Cased 'multi_cased_L-12_H-768_A-12'
- To put in folder 'we_models', downloadable here
Under 'teco_config', the configuration file is 'config.properties', where besides the paths to the models, there are also options for running TECo:
- Adaptation methods and their order, split by commas: VecDiff, Analogy, Subs
- Amount of expressions to be selected from the corpus, for adaptation
- Final Selection Method: TF-IDF, BERT
- Interval between tweets, in seconds. Useful if trying to run the twitter-bot.
- Run 'bert_server_run.py' (only if BERT is required)
- Run 'teco_main.py'
- TECo is described in two research papers:
TECo: Exploring Word Embeddings for Text Adaptation to a given Context, included in the proceedings of the 11th International Conference on Computational Creativity, which can be cited as follows:
@inproceedings{mendes_goncalooliveira:iccc2020b,
author = {Rui Mendes and Hugo {Gon{\c c}alo Oliveira}},
booktitle = {Proceedings of the 11th International Conference on Computational Creativity, September 7-11, 2020, Coimbra},
pages = {185--188},
publisher = {ACC},
series = {ICCC 2020},
title = {TECo: Exploring Word Embeddings for Text Adaptation to a given Context},
year = {2020}}
Amplifying the Range of News Stories with Creativity: Methods and their Evaluation, in Portuguese, included in the proceedings of the 13th International Conference on Natural Language Generation, which can be cited as follows:
@inproceedings{mendes-goncalo-oliveira-2020-amplifying,
title = "Amplifying the Range of News Stories with Creativity: Methods and their Evaluation, in {P}ortuguese",
author = "Mendes, Rui and Gon{\c{c}}alo Oliveira, Hugo",
booktitle = "Proceedings of the 13th International Conference on Natural Language Generation",
month = dec,
year = "2020",
address = "Dublin, Ireland",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.inlg-1.32",
pages = "252--262",
}