Cosine similarity compares two units of text to get the semantic relation between them. This comparison is based on the numerical value (features) represented by semantic vectors. Orthogonality between the feature vectors makes them inefficient for semantic comparisons. Modifying the metrics to handle orthogonality perform better taking the advantage of representations. This article, proposed modified cosine similarity metrics for comparing sentences based on multi-feature embedding vectors. Our approach relies on the assumption that linguistic units may have multiple aspects of semantics which should be considered while calculating the similarity between the two units.