Most widely used approaches use topic modelling i.e. LDA or NMF for finding keywords from a large corpus of documents. But, to get more meaningful, descriptive, precise and fine-grained keywords corresponding to a product description or review document, a little more specific approach is needed.

Dataset Used: https://www.kaggle.com/ak47bluestack/amazonphonedataset

ProductDescription2Keywords

Extracting Search Engine Appropriate Keywords and Key Selling Points from a Product's description in E-Commerce Websites
Problem Statement: Extract from mobile phone description dataset

1. Search Engine Appropriate Keywords

2. Key Selling Points

1. Extracting Search Engine Appropriate Keywords:

Proposed Approach:

 Search engine appropriate keywords can be:

  a. Specific : eg. Samsung Galaxy, Flip cover, Redmi note 3, etc. i.e. proper nouns

  b. General :Features such as: Smartphone, earphone, tangle free earphone (nouns other than proper)

Solution: POS tagging + TFIDF + ngram

 For each document :

   Extract Noun/ Proper nouns from preprocessed (stopwords+ lemmatization) text.

   Join the extracted words to make a new document (pos_filtered_doc)

 On list of pos_filtered_docs:

    Apply ngram TFIDFVectorizer and for each document extract words with top-n TFIDF value.

2. Extracting Key Selling Points:

Proposed Approach:

Key selling points can be treated as general, but little more descriptive, quantified features eg. 1.3ghz processor, bezel less screen, 24 hr battery backup, octa core processor

Solution:

Extract the noun phrases from the original document, corresponding to the keywords found in the previous step (i.e keyword extraction phase) .