https://github.com/legel/words2map

online natural language processing with word vectors
https://github.com/legel/words2map

Last synced: about 1 month ago
JSON representation

online natural language processing with word vectors

• Host: GitHub
• URL: https://github.com/legel/words2map
• Owner: legel
• License: mit
• Created: 2016-06-13T17:40:13.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2024-06-25T15:41:57.000Z (6 months ago)
• Last Synced: 2024-08-02T12:47:55.408Z (4 months ago)
• Language: Python
• Homepage: http://web.archive.org/web/20160806040004if_/http://blog.yhat.com/posts/words2map.html
• Size: 147 MB
• Stars: 310
• Watchers: 21
• Forks: 37
• Open Issues: 4
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.md

Awesome Lists containing this project

• awesome-starred - legel/words2map - online natural language processing with word vectors (others)

README

        


 

> *How words2map derives out-of-vocabulary ([OOV](https://medium.com/@shabeelkandi/handling-out-of-vocabulary-words-in-natural-language-processing-based-on-context-4bbba16214d5)) vectors by searching online:*  

> *(1) Connect NLP vector database with a web search engine API like Google / Bing*  

> *(2) Do a web search on unknown words (just like a human would)*  

> *(3) Parse N-grams (e.g. N = 5) for all text from top M websites (e.g. M = 50)*  

> *(4) Filter known N-grams from pre-trained corpus (e.g. word2vec, with 3 million N-grams)*  

> *(5) Rank N-grams: inverse global frequency x local frequency on M websites (i.e. [TF-IDF](https://en.wikipedia.org/wiki/Tf%E2%80%93idf))*  

> *(6) Derive a new vector: sum vectors for top O known N-grams (e.g. O = 25), i.e.*   




> *(7) Visualize by reducing dimensions to 2D/3D (e.g [t-SNE](https://lvdmaaten.github.io/tsne/) works, but [UMAP](https://github.com/lmcinnes/umap) recommended)*  

> *(8) Finally, show clusters with [HDBSCAN](https://github.com/scikit-learn-contrib/hdbscan), color-coded in a perceptually uniform space*  

> *These OOV vectors were derived in about 15 seconds as explained above:*  





> *See this [archived blog post](http://web.archive.org/web/20160806040004if_/http://blog.yhat.com/posts/words2map.html) for more details on the words2map algorithm.*  

### Derive new vectors for words by searching online

```python

from words2map import *

model = load_model()

words = load_words("passions.csv")

vectors = [derive_vector(word, model) for word in words]

save_derived_vectors(words, vectors, "passions.txt")

```

### Analyze derived word vectors

```python

from words2map import *

from pprint import pprint

model = load_derived_vectors("passions.txt")

pprint(k_nearest_neighbors(model=model, k=10, word="Data_Scientists"))

```

### Visualize clusters of vectors

```python

from words2map import *

model = load_derived_vectors("passions.txt")

words = [word for word in model.vocab]

vectors = [model[word] for word in words]

vectors_in_2D = reduce_dimensionality(vectors)

generate_clusters(words, vectors_in_2D)

```

### Install 

```shell

# known broken dependencies: automatic conda installation, python 2 -> 3, gensim

# feel free to debug and make a pull request if desired

git clone https://github.com/overlap-ai/words2map.git

cd words2map

./install.sh

```