There are number of techniques for opinion mining. But, mining the most frequent features from online reviews is the biggest task. If we use a single domain then it gives poor results. Using two or more domains is very difficult for feature mining process. In this paper, we proposed a novel technique for selecting and identifying most frequent features by using two similar or different domains (Main Domain and Sub Domain) which is based on two statistics. We first extract the features from Main domain by using some rules. For each feature (aspect) we then estimate its Domain Relevance Score which is called as Intrinsic Domain Relevance (IDR). Features those are having the less score (IDR score less than threshold) will be pruned and only those features get extracted which is having High IDR score. In the final step we use our proposed EDR method. The features extracted from main domain having high IDR will be search out in sub domain. At the time of searching it will remove those features which are not available in sub domain. Hence, DR will be compute only for available features. Which known as EDR. Finally, get the most relevant features from sub domain or both the domains. Experimental results based on two different Worldreview domains show the improved EDR approach.