In recent times, Graph Representation Learning (GRL) has become popular due to the advancements in Deep Learning (DL) and Machine Learning (ML) algorithms. Most real-world networks are dynamic in nature, thus temporal information is intrinsic part of the network information. There are very few GRL schemes that take temporal aspect into consideration. In this work, we propose a GRL method, namely, tNode2Vec, in which the popular Node2Vec algorithm is extended by defining an adaptive temporal random walk that extracts the node neighborhood samples using degree-based and temporal distribution. We consider two downstream prediction tasks namely, link prediction and recommendation, to evaluate the quality of the proposed model. Extensive experimentation is performed on two real-world bibliographic datasets, Hep-Ph and DBLP V11. DBLP V11 is a large well-known dataset with a collection of 41, 07, 340 research papers and conference proceedings from computer science and related domains. We have implemented all the baseline methods to solve the link prediction and recommendation problems. The proposed tNode2Vec method achieved better results compared to all the baseline methods for the link prediction task as well as for the recommendation task. We have included the BERT model in the proposed tNode2vec model as well as baseline methods to extract textual node attributes and combine them with node representation to enhance the link prediction performance.