I am trying to calculate the semantic similarity by inputting the word list and output a word, which is the most word similarity in the list. E.g.
input:
list = ['portugal', 'spain', 'belgium', 'country', 'netherlands', 'italy']
output:
['country']
Follow these steps:
1. Use one of the neural embedding methods (e.g. SkipGram, CBOW, BERT, etc.)
2. Find the embedding vectors of your input list.
3. Map Step 2 values to 2-dimensional vectors (i.e use T-SNE for reduction).
4. Get the Cosine similarity of Step 3.
5. Print the highest value from Step 4.
6. Print the corresponding values associated with Step 5.
Javad PourMostafa Roshan Sharami thank you for your effort, can you explain more?
E.g. Use the BERT to find the embedding vector of each input words. It maps them into n-dimensional vectors. Then reduce n-dim to 2-dim utilizing T-SNE or any other dimensionality reduction algorithms. Next plot all 2-dim vectors on the coordinate system and compute their cosine similarities. It shows the similarities between vectors representing your words.
https://stackoverflow.com/questions/62546815/how-to-use-word-embeddings-i-e-word2vec-glove-or-bert-to-calculate-the-most
Hi Alhassan,
From the example you give, it sounds like rather than lexical similarity, what you really want is to find the hypernym (the broader category) of all other elements within the list. If this is the case, it might be worth taking a look at a SemEval shared task on hypernym discovery Conference Paper SemEval-2018 Task 9: Hypernym Discovery
https://towardsdatascience.com/from-pre-trained-word-embeddings-to-pre-trained-language-models-focus-on-bert-343815627598
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python