Document Similarity

Document similarity measures are basis the several downstream applications in the area of natural language processing (NLP) and information retrieval (IR).

Sentence-level

• ERCNN: Enhanced Recurrent Convolutional Neural Networks for Learning Sentence Similarity

BERT and other Transformer Language Models

• BERT
• GPT
• Generative Pre-Training-2 (GPT-2)
• Universal Language Model Fine-tuning (ULMFiT)
• XLNet

Overcoming BERT's 512 token limit:

• Long-form document classification with BERTr/bert_document_classification)
• BERT-AL: BERT for Arbitrarily Long Document Understanding
• Blockwise Self-Attention for Long Document Understanding
• BP-Transformer: Modelling Long-Range Context via Binary Partitioning. (2019).
• Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context.
• Longformer: The Long-Document Transformer
• Reformer
• Compressive Transformers for Long-Range Sequence Modelling

Siamese Networks

• Siamese Recurrent Architectures for Learning Sentence Similarity
• SMASH-RNN: [Jiang, J. et al. 2019. Semantic Text Matching for Long-Form Documents. The World Wide Web Conference on - WWW ’19 (New York, New York, USA, 2019), 795–806.]
• [Liu, B. et al. 2018. Matching Article Pairs with Graphical Decomposition and Convolutions. (Feb. 2018).]

Text matching

• [Simple and Effective Text Matching with Richer Alignment Features]
• [Enhanced Text Matching Based on Semantic Transformation]