Sarcasm detection is a challenge in review based e-commerce product recommendation systems. Failure to detect sarcasm causes assignment of directionally opposite ratings to reviews and faults the recommendation systems. This work proposes a cross domain hybrid feature fusion based sarcastic opinion recognition system. Hybrid deep feature integrating content and context is formed and feature vector volume is expanded using cross domain adversarial learning. The hybrid features are used for sarcasm detection in two combinations. First combination is hybrid features with bi-directional long short term memory (LSTM)and second is hybrid features with traditional machine learning classifiers. Hybrid feature with bi-directional LSTM is able to capture temporal relation between sentences of the review for better identification of sarcasm. Traditional machine learning classifiers trained with these hybrid features are also able to provide higher accuracy of sarcastic detection in reviews. The effectiveness of the proposed cross domain hybrid feature fusion features is tested against various experimental setups with Bi-directional LSTM and traditional machine learning classifiers. The proposed solution is able to achieve peak accuracy of 95% in classifying e-commerce reviews.