Progress
- new plugin set up: GitHub - joplin/plugin-ai-summarisation
- created and added a idea for the summarisation feature: Allow users to summarise particular part of the note by highlighting the text
- implemented single note summarisation with TextRank
- explored more algorithms for extractive summarisation: LexRank, LSU, Luhn, TD-IDF, KL-Sum, BERTSUM
- experimented with Tensorflow.js in a plugin
- used and tested pre-trained qna model for the summarisation feature
- with 1000 characters, it takes 1 min for a model to give an answer => very slow for large notes
Plans
- implementing "summarising multiple notes simultaneously" feature
- learn more about webpack
- making a table of all extractive summarisations and discover their strengths and weaknesses
- TextRank could be enhanced:
- [Interactive Text Ranking with Bayesian Optimisation: A Case Study on Community QA and Summarisation] ([1911.10183] Interactive Text Ranking with Bayesian Optimisation: A Case Study on Community QA and Summarisation) [1911.10183] Interactive Text Ranking with Bayesian Optimisation: A Case Study on Community QA and Summarisation
- [The enhancement of TextRank algorithm by using word2vec and its application on topic extraction] (Redirect Notice
- [Enhancement of TextRank Algorithm using Coreference Resolution] (https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://ijrp.org/filePermission/fileDownlaod/4/be915dd3ab99f5f1ed933e0765edbfe2/2&ved=2ahUKEwi8gsjGwqyGAxVKQkEAHZtUA2UQFnoECCEQAQ&usg=AOvVaw3KVBOsJ9jXwSOHtLZikBln)https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://ijrp.org/filePermission/fileDownlaod/4/be915dd3ab99f5f1ed933e0765edbfe2/2&ved=2ahUKEwi8gsjGwqyGAxVKQkEAHZtUA2UQFnoECCEQAQ&usg=AOvVaw3KVBOsJ9jXwSOHtLZikBln
- Algorithms that need to be written from scratch: Luhn, LSU, KL-Sum and BERTSUM. (For improving TextRank, we might have to write that from scratch too)
- TextRank could be enhanced:
Problems
- I am still having issues with setting up Transformers.js in a plugin