Search code examples
pythonnlpspacysimilarityspacy-3

Getting similar words no longer working in spacy

I have a Google Colab notebook from a while ago which uses spacy 2.2.4 and successfully gets the most similar words for a list of words:

import spacy
import spacy.cli
spacy.cli.download("en_core_web_lg")
import en_core_web_lg
nlp = en_core_web_lg.load()
import numpy as np
import pandas as pd

print(spacy.__version__)

all_search_terms = ["technology", "internet", "smartphone"]

# define a function to get the x most similar words to a word
def most_similar(word, topn=2):
    print(word)
    word = nlp.vocab[str(word)]
    print(word.prob)
    queries = [
        w for w in word.vocab 
        if w.is_lower == word.is_lower and w.prob >= -15 and np.count_nonzero(w.vector)
    ]

    by_similarity = sorted(queries, key=lambda w: word.similarity(w), reverse=True)
    return [(w.lower_,w.similarity(word)) for w in by_similarity[:topn+1] if w.lower_ != word.lower_]


# create function to receive a list of words and return the 
# top 2 similar words for each word in the list

def get_similar_words(list_of_words):
    
    all_similar_words = []
    
    for word in list_of_words:
        spacy_word = nlp.vocab[str(word)]
        if spacy_word.has_vector:
        
            # find similar words to the word, and store them in a dataframe along with their scores
            similar_words = pd.DataFrame(most_similar(word, topn=2), columns=["word", "similarity_score"])

            # save the list of similar words
            similar_words_list = list(similar_words["word"])

            # append the list of similar words to the list to be returned
            all_similar_words.append(similar_words_list)
        
    # flatten the list of lists to one list
    all_similar_words = [item for sublist in all_similar_words for item in sublist]
    
    # remove duplicates from the list
    all_similar_words = list(dict.fromkeys(all_similar_words))
    
    # sort list in alphabetical order
    all_similar_words.sort()

    return all_similar_words


# run the function on the search terms entered by the user
new_search_terms = get_similar_words(all_search_terms)
new_search_terms

The output is:

technology
-10.063644409179688
internet
-8.897857666015625
smartphone
-12.11159896850586
['handset', 'online', 'smartphones', 'technological', 'technologies', 'web']

THE PROBLEM: I've just tried running the same code in a different environment within RStudio (i.e. NOT using Google Colab) where the version of spacy is 3.0.6 and the list of similar words (new_search_terms) is empty. I've also noticed that the word probabilities are all the same (-20).

The output with spacy 3.0.6:

technology
-20.0
internet
-20.0
smartphone
-20.0
[]

What do I need to do differently in this new version of spacy to get the same output as before?

Solution

  • The token probabilities are not loaded by default in v3 and so you have to do some stuff to load them.

    import spacy
from spacy.lookups import load_lookups
nlp = spacy.load("en_core_web_sm")
lookups = load_lookups("en", ["lexeme_prob"])
nlp.vocab.lookups.add_table("lexeme_prob", lookups.get_table("lexeme_prob"))

    After this your code should work, though I am not sure why you are using .prob here.