The Anatomy of a Restaurant Review

Evaluating Supervised Classifiers for Fine-grained Sentiment Analysis

This data source has 8000 reviews in total.

Review #639

I am very grateful to the King’s meal (the first one).It’s a little excited.First of all, the exterior building is very characteristic.When I enter the building, I see the phone booth and the mail box.I feel that girls will like to take photos here, and the cough will return to the topic.The Kellett Chimney Circle is very easy to find next to the elevator on the 2nd floor.The owner of the store will ask you if you like sweet and salty, and then recommend the taste you like.The tea is also recommended by the boss for the mint flower tea.First, the boss recommended me that I chose the cheese chimney circle.I would like to dry it when I eat the chimney ring.But with the feeling of flower tea, there is still no boss.The boss is very good, and I will talk about it.I will not feel when I go alone.Bored, when I was eating, I didn’t say hello when I saw the boss.I’m here to say thank you for your loved ones.I can go and try it out.The shops in the collection are full of features.

Click or select sentences to see sentiment predictions for the clicked/selected sentence(s).

Colors highlight the sentiment polarity of key phrases ( Negative or Positive ) as predicted by TextBlob and SnowNLP. These sentiment predictions are not used in the classification models that power the visualization on the right. They are added only for referrence purposes.

Base Classifiers are fitted for each fine-grained category one by one.

Sentiment
Not mentioned
Negative
Neutral
Positive

predict: the predicted probablity (for LDA and Logistic Regression) or final labels of the sentiments.
P: the predicted sentiment labels.
A: the actual sentiment labels.
global: the global distribution of the predicted and actual labels in the selected dataset.

F1: F1 score of predictions on the selected dataset (highlighted are scores more than one standard deviation below the average).