Small Bug Fixes

• Sentence separation was reverted, but now fixed. Test to ensure.

Version 0.10.0 More flexible changes for custom transformer models

• Refactoring to make it easier for adding new transformer models.
• Bug fix: fixed issue with incorrect sentences.
• Fixed issue with spacing from sentences.
• Improved Doc strings.
• Cleaned up code with new proper naming and appropriate python packages.

Version 0.9.0: SBert Implementation

In this version, a few minor fixes are added, along with the SBert implementation.

Paper: Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
Project: https://github.com/UKPLab/sentence-transformers

All features are available for SBERT that are available to the existing summarizer.

Bug Fix

0.8.1

Bump to transformers 4

Transformers 4 Support

This release has Transformers 4 support.

KMeans Elbow support.

This release contains two different features:

1. Elbow support. This includes returning inertias plus finding optimal cluster amount.
2. Optional return sentences as a list.

Spacy 3.0 Support

This release supports 3.0 for the sentence handler.

Update on setup

0.6.1

Added Hidden Layer Processing with correct version

Multiple layer processing. Added new Transformers.

In this release, there are two new arguments for summarizer processing. One is the hidden parameter which now accepts an integer or a list. If supplied a list, it will combine the layers for the aggregation. Also, another parameter called hidden_concat allows the user to concat hidden layers before clustering.

Also a couple of new options for TransformerSummarizer.

2 new embedding reduction options.

In this release, two new reduce options are included: concat_last_4 and reduce_last_4. The concat_last_4 concats the last 4 BERT hidden layers and takes the mean for aggregation. The reduce last 4 option sums the last four layers and takes the mean across tokens.