Information diffusion modeling on social media is modeled using directly observable characteristics at one extreme, such as over activities (retweets and mentions) and inbuilt content characteristics (hashtags), or over methodologies that are algorithmically obtained on the other side, such as LDA-based similarity of large buckets of tweets. Initial works have been conducted by researchers that attempt to model information diffusion using similarity of a restricted set of hashtags, where they have manually grouped known hashtags. In the current work, we observe that socially connected users on Twitter tend to use the same words over time, under the influence of their peers. We also observe that, users that tend to use the same words (unigrams) for the same given topics, tend to belong to the same user communities, compared to the users using different words with sufficient LDA similarity. Encouraged by these research trends and observations, we aim to inspect intermediate models and opinion similarities for information diffusion. To this, we propose a simplistic method to identify paraphrases from tweets, with a view to exploring the impact of such paraphrases in information diffusion in the future. The current talk briefs the above, and details our approach for detecting paraphrases and semantic similarities of given pairs of tweets.