semantic-counter 0.0.3

A wrapper around CountVectorizer which counts semantically similar tokens

Wrapper around CountVectorizer to count phrases based on semantic similarity

---

SemanticCountVectorizer extends CountVectorizer by counting the frequency of semantically similar tokens. This is useful when searching for frequency of similar but not exact terms.

SemanticCountVectorizer uses sentence_transformers to compute embedding of n-grams and then clusters similar terms using Agglomerative Clustering using cosine similarity as the similarity metric

---

Installation

pip install semantic_counter

Usage

from semantic_counter.semantic_vectorizer import SemanticCountVectorizer 
sentences = ['this is a test sentences','this is another test sentence']
svect = SemanticCountVectorizer(embedding_model_name='all-MiniLM-L6-v2', similarity_threshold=0.7, ngram_range=(3,3))
counts=svect.fit_transform(sentences)

Arguments

• embedding_model_name: Name of the sentence_transformer pretrained model to use. Refer to package website for a list of available models
• similarity_threshold: value between 0-1 which indicates minimum cosine similarity needed to group tokens.
• Arguments of sklearn's CountVectorizer as listed here