The need for a visual question answering framework in the era of Web 3.0 is realized by using staged addition of auxiliary knowledge in the perspective of two distinct datasets namely the dataset of documents and the images which is enriched through informative term extraction and further using structural topic modelling, Wikidata, NELL and Linked Open Data (LOD) Cloud, knowledge stores and tag generation to classify the dataset obtained informative terms. Subsequently, the image dataset terms are subjected to extraction of categories and labels to contribute to the informative terms. The random forest classifier at either ends of the two datasets helps in controlled classification of the datasets independently and assures computational inexpensiveness because of the lightweighted nature of the machine learning classifiers. The RNN to classify the tags helps in automatizing the exponentially generated tags and semantic network formulated helps in organizing the auxiliary knowledge. Semantic relevance computation using Jaccard Similarity, Morisitas Overlap Index, NPMI and SOC-PMI facilitates in semantic reasoning. The proposed VQARS model's achieves an overall precision of 94.81%; recall of 95.75%; average accuracy of 95.28%; F-measure of 95.28%; and the lowest FDR of 0.06. In comparison to the other baseline models, this model provides the best-in-class outcomes.