1
$\begingroup$

I am working on e-commerce, how to rank smartphones in the same category. I want to calculate a weighted average of sentiment score from reviews posted by buyers.

Weights will be based on how many days before review was posted, as people might give more importance to recent reviews than old reviews. Now to find weight for previous day review, i want to understand impact of previous reviews on current review. i.e. impact of t-1, t-2, t-3 on t.

I want to know for how many days should i aggregate review sentiment?

$\endgroup$
1
$\begingroup$

Welcome to DataScience.SE. Let's see if I've interpreted your question correctly:

i want to understand impact of previous reviews on current review

Keep track of which reviews a reviewer has read (randomly withhold some if you have to), and use this information together the rating of the new review to create a regressor to estimate the score of a new review given the age and score of selected previous reviews.

You can do similar regressions with the topic and sentiment, if you use embeddings.

$\endgroup$
2
  • $\begingroup$ My objective is to model impact of online rating and review sentiment on price of smartphone. I have multiple smartphone with almost similar configuration. My client is smartphone seller. He wants to know how much discount i should give on my product if my online review or rating is better or worse than competing products. Also, please note that a buyer will not read thousands of reviews. He will skim through a)few recent reviews and will look at b)average rating. $\endgroup$ Jul 12 '16 at 7:47
  • $\begingroup$ My challenge is mostly for creating a feature to model for a), as each review will have sentiment score. I want to give higher weightage to review sentiment from first page and significantly low weightage to review sentiment from page 2 on wards. My problem is to quantify the weights that should be given to reviews in page1 and reviews from page 2 onwards. Can you suggest any idea on how to proceed? $\endgroup$ Jul 12 '16 at 7:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.