Knowledge Center

This is a page to explain the key concepts in the package and how aspects and opinions are extracted.

---

Comments and Sentences

A review dataset is mostly like this:

We can see the level of granularity:

Review data → comment by a single customer → sentences in a comment

Here we assume that each customer only comments once. “Comment by a single customer” is an important level when we do sentiment analysis. Every customer has an overall attitude towards the product. Maybe they feel happy about one aspect but not satisfied with another but they still love to use the product, then it’s a positive feedback (but the sentiment score will be lower than that of the customer who has no complaints). Some customers writes a lot and some customers only leave one sentence. If we simply calculate sentiment score on each sentence regardless of who made the sentence, the sentiment from the customer who writes a lot will be over-weighted and the sentiment from the other type will be under weighted.

“Sentences in a comment” is important when we do analysis on aspects. We will talk about aspects later. But imagine we would like to know a specific aspect of a product. Usually the customers will mention an aspect for only one time in their reviews. We just break down the comments to sentences to make it easier.