Suppose we are doing sentiment analysis for a restaurant. Customers can rate the restaurant by #1: how expensive the restaurant is, #2:how good is the food and #3: how likely they will come again. The ratings are dependent,i.e. the more expensive the restaurant is (higher #1), the less likely they will come back (lower #3), but whey will if the food is good (higher #2).

My questions are: is there a good RNN structure(review as input, #1-#3 as output) that can capture and model the dependency among #1 - #3?

  • $\begingroup$ "(review as input, #1-#3 as output) " you want one input and 3 different outputs, are you referring to multi-task learning? And what do you mean exactly with "dependency among #1 - #3"? Can't you just perform classic statistic like correlation between variables and stuff? $\endgroup$ – Edoardo Guerriero Mar 14 at 14:47

Inspired by (1) machine translation where source phrases are first encoded to a feature vector and decoded by another RNN and (2) automatic captioning where a picture is also encoded to a feature vector before it's decoded by the RNN followed. A proper structure for the problem above may be the inverse of these, where the review is encoded by RNN to a vector, and a DNN is followed to decode the vector into #1 - #3.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.