Comparison between cbow, skip gram and skip-gram with subword information
0 likes105 views
The document compares CBOW, Skip-gram, and Skip-gram with subword information regarding their training data, model settings, and performance on tasks such as related concepts and similarity. It highlights the limitations of plain word embeddings in capturing fine-grained semantic similarities and their dependency on training data quality. Visual comparisons and metrics such as precision, recall, and F-score are also discussed.
![Comparison Between CBOW, Skip-
Gram and Skip-Gram with Subword
Information [ See article ]
By: Kavita Ganesan](https://image.slidesharecdn.com/comparisonbetweencbowskip-gramandskip-gramwithsubwordinformation-200424150406/85/Comparison-between-cbow-skip-gram-and-skip-gram-with-subword-information-1-320.jpg)
Comparison between cbow, skip gram and skip-gram with subword information
- 1. Comparison Between CBOW, Skip- Gram and Skip-Gram with Subword Information [ See article ] By: Kavita Ganesan
- 2. CBOW This is a visual comparison sum visual thisis acomparison visual thisis acomparison visual visual visual target word By: Kavita Ganesan target word SkipGram
- 3. This is a visual comparison sum visual th ishi target word SkipGram + Subword (SkipGramSI) this<t s>
- 4. Training Data & Model Settings
- 6. Differences in Training time With Gensim
- 7. Task 1: Most related concepts
- 12. Task 2: Similarity between words
- 14. Task 3: Similarity between phrases
- 15. Dataset
- 16. Precision, Recall & F-Score Cosine threshold = 0.6
- 17. Raw Predictions
- 18. Final Thoughts Plain word embeddings: ● Does not capture fine-grained semantic similarity ● Words are not directly replaceable with words closest in vector space ● Quality of embeddings is only as good as the training data