SAMEntics : Tools for paraphrase detection

on YouTube ↗31m

" summary

SAMEntics claims unsupervised paraphrase detection and generation that combines augmented structural metrics (Jaro, Jaro‑Winkler, Dice, cosine, Levenshtein applied at the word level) with semantic models: Word2Vec++ (Word2Vec + LSA + LDA), WordNet similarity scored via BLEU, and a recursive autoencoder to collapse sentences into phrasal vectors. Hierarchical centroid linkage clustering yields 74.8% accuracy and 82.6 F1 on MSRP+SemEval style corpora. Paraphrase generation uses logical forms turned into disjunctive forms and a bi‑directional bigram realizer with OpenCCG to produce grammatical variants for data augmentation.

AI-written summary · about

" chapters (5)

1. 0:00Introduction and background
2. 2:02Paraphrase detection pipeline
3. 16:44Demo and IBM Watson use case
4. 24:23Paraphrase generation approach
5. 29:14Q&A and closing

" related talks

• 36mSAMENVATTR, or, How to Make a Text Summarization Tool for your Language of ChoiceRobert Rodger · PyData Amsterdam 2018
• 58mAutomatic text summarizationMasa Nekic · NDC Oslo 2019
• 01h 28mDocument/Sentence Similarity SolutionAde Idowu · PyData London 2022
• 01h 31mNatural Language Understanding: Foundations and State-of-the-ArtPercy Liang · Foundations of Machine Learning Boot Camp 2017
• 01h 00mHuman and machine inductive biases for compositional linguistic generalizationNajoung Kim · Emerging Generalization Settings 2024
• 39mBeyond word2vec: GloVe, fastText, StarSpaceKonstantinos Perifanos · PyData London 2018
• 41mSemantic Multimodal Multilingual Similarity EngineVladimir Orshulevich · PyData Yerevan 2022
• 14mSami stopwords - How far have we gotten and why does it matter?Espen Klem · NDC Oslo 2022