Hey Kingsley! You got a lovely name :) The dataset we are using is the actual Amazon Alexa Product Reviews & exhibits the true distribution of Review sentiments there. Therefore we went ahead with the same data and got a good accuracy score of 91%. However, definitely you should try out data balancing. Do share the results here. :)
@@skillcate Thank you for your response and the compliment. In my own case, I am built a model following your simple and easy to understand steps, The model was based on the review of Google play store app of Piggyvest. It is a Nigerian based fin tech company. I got an accuracy of 92% when I used the rating as my labelled column. I said if a rating is greater than 2 then label it 1 else label it 0. However, when I used a new column remained review which I created from the sentiment polarity. This time if the polarity is zero or negative label it 0 else label it 1, this time I got accuracy of 63% with TP TN being 0. I would have loved to share the link, but it is my In-Course Assessment for my Intelligence Decision Support System module in Master course. Please let me get your thoughts.
Thank you, I just stumbled on your channel yesterday. Your tutorials are truly genius.
Glad you like them!
Good job.
Thank you for your tutorials.
I was wondering, the labelled classes are imbalanced, you didn't consider resampling it.
Why is that please?
Hey Kingsley! You got a lovely name :)
The dataset we are using is the actual Amazon Alexa Product Reviews & exhibits the true distribution of Review sentiments there. Therefore we went ahead with the same data and got a good accuracy score of 91%.
However, definitely you should try out data balancing. Do share the results here. :)
@@skillcate Thank you for your response and the compliment.
In my own case, I am built a model following your simple and easy to understand steps,
The model was based on the review of Google play store app of Piggyvest.
It is a Nigerian based fin tech company. I got an accuracy of 92% when I used the rating as my labelled column.
I said if a rating is greater than 2 then label it 1 else label it 0.
However, when I used a new column remained review which I created from the sentiment polarity.
This time if the polarity is zero or negative label it 0 else label it 1, this time I got accuracy of 63% with TP TN being 0.
I would have loved to share the link, but it is my In-Course Assessment for my Intelligence Decision Support System module in Master course.
Please let me get your thoughts.