In aspect level sentiment, a word is closer to an aspect word, it will have stronger affection to this aspect than words that are far from it. How to use this distance to compute the relevance between words and aspecct terms in a sentence? Paper Discovering Protagonist of Sentiment with Aspect Reconstructed Capsule Network used a method.
This method is below:
Here \(n_i\) = 75, α = 3 β = 10 γ = 1 in this paper.
Problem in this location proximity strategy
There also two main problems in this method, they are:
- If an aspect term contains some words, such as compute battery, how to determine the k?
- If there are some aspect terms in a sentence, how to compute this location weight \(l_t\)? For example, if a sentence contains price and quality, how to compute?