Difference usage of document level, sentence level and aspect level in sentiment analysis

Question

I'm studying about Sentiment analysis. What is the purpose of using Document Level, Sentence Level and Aspect Level?

I read this definition in Web Data Mining book by Bing liu.

In this book it was stated that,

Sentence Level sentiment analysis is to classify a sentence to negative, positive, neutral class.

Document level sentiment analysis to classify a document.

Aspect Level sentiment analysis is to classify each aspect of entity mentioned in a review.

Thanks.

Answer 1

I am building on Toros91's answer. He has said:

Use case of all these 3 different levels are:

Sentence Level: Used to classify a sentences for instance, whether a tweet is positive or negative or neutral.
Document Level: Used to classify the whole document level to say whether that document sentiment is positive or negative or neutral.
Aspect Level: Used to select an aspect and give the outcome of that respective Aspect, for instance let us consider Iphone X as an

aspect and mine tweets(this is an example you can use documents and so on) and carry out Sentiment Analysis.

Now, to answer your question further about how it can be beneficial, just take an example of some review. I am just forming my own review on something as an example, I do not know the specifics, so please do not quote me here. This has some business logic implied as well.

I feel the latest laptop from Mac is really good overall. It has amazing resolution. The computer is really very sleek and can slide into bags easily. However, I feel the weight is a let down. The price is a bit expensive given the configurations. I expect an SSD storage. However, the processor seems really good.

At a document level, I can say the entire paragraph looks more positive than negative. So, the review, if has to be labeled as one thing, is positive.

At a sentence level, I can break the document into 6 sentences and classify each of them as positive, negative or neutral. I get a bit more specific here.

Finally, at an aspect level, I get information about price, storage, processor and so on.

Given N customer reviews, you can know how well different aspects of a particular product are received. Say, 10 customers feel the price is fine, but 4 feel it is expensive. 13 of your 14 customers really love the resolution while 1 feel it is okayish. A lot of products in day to day life have multiple things to be considered. And this is where fine grained labeling help analysts or scientists.