In this work, we propose a context-based bilingual word embedding framework that leverages the information of large amount of parallel sentence pairs which share the same semantic meaning. Such information is abundantly available but has not been fully utilized in previous work of context-based bilingual word embedding models, which only exploit local contextual information through a short window sequence at the word level. To incorporate such information, we define a sentence similarity matching objective which is enforced as a constraint into the original bilingual word embedding objective. They are jointly optimized to better learn the bilingual word embedding. Experimental results show that the proposed model is superior to previous methods on machine translation quality.