Gensim-LSI-Word-Similarities

Two simple little functions to create word-word similarities from Gensim's latent semantic indexing in Python. Both functions produce an inverted cosine similarity score (0 = low, 1 = high) between two words in a Gensim-generated LSA/LSI space across the total number of dimensions specified in the creation of the model (i.e., num_topics from gensim.models.LsiModel).

Both require Gensim, Pandas, and SciPy.

Includes four functions:

• wordsim: Create cosine-derived similarity score (from 0-1) between individual words. Input:
• word1 (string or string variable)
• word2 (string or string variable)
• target_dictionary (Gensim-created LSI dictionary)
• target_lsi_model (Gensim-created LSI model)


• wordvectsim: Same as wordvect but created to calculate similarity scores (from 0-1) for word pairs in a 2-dimensional word vector (e.g., using numpy.apply_along_axis). Input:
• word_vector2d (2D string vector or 2D string vector variable)
• target_dictionary (Gensim-created LSI dictionary)
• target_lsi_model (Gensim-created LSI model)


• Two additional functions/series of functions added (detailed documentation available in each function and will be added here soon):
• word2vec_vect_sim_fun: similarity score function for gensim's word2vec
• word_pair_similarity_matrix: word-word similarity matrix function for gensim's LSI (LSA) model