Citation based approaches have seen good progress for recommending research papers using citations in the paper. Citation proximity analysis which uses the in-text citation proximity to find relatedness between two research papers is better than co-citation analysis and bibliographic analysis. However, one common problem which exists in each approach is that paper should be well cited. If documents are not cited properly or not cited at all, then using these approaches will not be helpful. To overcome the problem, this paper discusses the approach Virtual Citation Proximity (VCP) which uses Siamese Neural Network along with the notion of citation proximity analysis and content-based filtering. To train this model, the actual distance between the two citations in a document is used as ground truth, this distance is the word count between the two citations. VCP is trained on Wikipedia articles for which the actual word count is available which is used to calculate the similarity between the documents. This can be used to calculate relatedness between two documents in a way they would have been cited in the proximity even if the documents are uncited. This approach has shown a great improvement in predicting proximity with basic neural networks over the approach which uses the Average Citation Proximity index value as the ground truth. This can be improved by using a complex neural network and proper hyper tuning of parameters.