No Need to Look Back: An Efficient and Scalable Approach for Temporal Network Representation Learning