Step 1 – Relevant Opinionated Words Extraction: All the review comments posted by various buyers on the website are fed, one by one, into the process of relevant opinionated word extraction.
Not all the words written in the comment are useful from the perspective of judging positivity, negativity, or objectivity towards the seller. So, in a review comment, the words that express positivity, negativity, or objectivity regarding the buying experience, product quality, genuineness, packaging, or delivery are considered as the relevant opinionated words.
Read more on
Step 2 – Sentiment Classification: The list of opinionated words for each user comment is then fed into the process of sentiment classification. Each opinionated word is examined for possessing positive, neutral, or negative sentiment. The positive sentiment word gets a score in the range (0,1), negative word gets a negative score, and the neutral word gets zero.
Step 3 – Trust Score Evaluation: The positivity in review comments is measured to produce the Seller Trust profile. The profile would contain the computed seller trust using two metrics; (1) the positive review percentage and, (2) the out-of-5 rating.
The positive review percentage is the percentage of reviews having the out-of-5 ratings 4 or above. The out-of-5 rating is the positive review percentage normalized to 5.
The neutral review percentage is the percentage of reviews having the out-of-5 ratings equal to 3.
The negative review percentage is the percentage of reviews having the out-of-5 ratings 2 or below. The out-of-5 rating is the negative review percentage normalized to 5.
The final output is the Seller Trust profile, which is a combination of positive review percentage as well as rating out-of-5. The Seller Trust Score is then fed into yet another algorithm that processes thus calculated Seller Trust Score, with overall on-Page Relevancy and Recency to calculate Weightage of a Review Rating that would eventually contribute towards Listing Quality Rank.
Author – Rizwan Zaffar (Originally posted in Ecommerce Outset)