why add the convolution layer?

[YHTtTtao]

LSTM can already capture the position information，why you add a position-aware convolution layer？

[lixin4ever]

Here, the position information refers to the position relevance between the current hidden representation and the given aspect/target representation. LSTM can indeed capture the target-specific position information in an implicit way (i.e., closer (to the target) is more important). Our design is to explicitly incorporate such information into the convolutional feature extractor.