Evaluation of Reviews Appropriateness for Outbound Marketing
Description
Project aimed at building a service to rank consumer reviews of a product in order of their relevance for a customer.
Studied about the current reviews ranking algorithm used by amazon and proposed a method to predict helpfulness of reviews by considering only the structural features of review text.
Filter out inappropriate and non-English reviews. Used readability formulas to get readability score of the reviews.
Assigned a content score to each review. This score depends upon how much does a review talks about the product (this is also used to filter out reviews which does not belong to this product).
Used content score and readability score to rank the reviews and extract top n (given by the user) reviews.
Created python package using nltk to generate ranked list of reviews. Used AWS’s Dynamodb and S3 to get and store data.
Created Java Service which first fetches reviews of a product and then uses the python package to rank the reviews. Used Amazon Jython to integrate Java service with python package.