Maybe you should provide the data to analyze the feedback. You can use some of the expert system (Bayes) or neural network (Backpropagation) algorithms to train the previous data. The positive or negative response depends on the accuracy of the data training.

In feedback, user should give some rating like 1,2.... upto 5 or 10. Based on that rating value one can decide the sentiment of feedback. If rating value is lower then sentiment is negative else positive. So , rating maybe the act as keyword here.. Several other solutions are also available. Just bing liu paper 

One way to implement is, use dictionary. Make a list of words and categorize based on their negativity and positivity. For example, 1 is lease negative and 5 is highly negative. Mcategorize all words in dictionary and give them mentioned weightage. Based on number of string or word use entered in site ,we can take average of weightage and see if users sentiment falls under positive or negative area. 

you should learn your machine some knowledge. probably using Naive Bayes is so much powerfull and usefull to gain maximum result of your sentiment analysis

https://jurnal.uns.ac.id/index.php?journal=itsmart&page=article&op=view&path%5B%5D=630

Thanks for sharing..