Today, developing new software variant to meet new demands of customers by ad-hoc copying of already existing variants of a software system is a frequent phenomenon in the software industry. Typically, maintaining such variants becomes difficult and expensive over the time. To re-engineer such software variants into a software product line (SPL) for systematic reuse, it is important to identify source code elements that implement a specific feature in order to understand product variants code. Information Retrieval(IR) methods have been used widely to support this purpose in a single software. This paper proposes a new approach to improve the performance of IR methods in a collection of similar software variants. Our proposal produces following two improvements. First, increasing the accuracy of IR results by exploiting commonality and variability across software variants. Secondly, increasing the number of retrieved links that are relevant by reducing the abstraction gap between feature and source code levels. We have validated our approach with a set of variants of two different systems. The experimental results showed that the proposed approach outperforms the conventional application of IR as well as the most relevant work on the subject.