Unsupervised Path Representation Learning with Curriculum Negative Sampling

Abstract

Path representations are critical in a variety of transportation applications, such as estimating path ranking in path recommendation systems and estimating path travel time in navigation systems. Existing studies often learn task-specific path representations in a supervised manner, which require a large amount of labeled training data and generalize poorly to other tasks. We propose an unsupervised learning framework Path InfoMax (PIM) to learn generic path representations that work for different downstream tasks. We first propose a curriculum negative sampling method, for each input path, to generate a small amount of negative paths, by following the principles of curriculum learning. Next, PIM employs mutual information maximization to learn path representations from both a global and a local view. In the global view, PIM distinguishes the representations of the input paths from those of the negative paths. In the local view, PIM distinguishes the input path representations from the representations of the nodes that appear only in the negative paths. This enables the learned path representations encode both global and local information at different scales. Extensive experiments on two downstream tasks, ranking score estimation and travel time estimation, using two road network datasets suggest that PIM significantly outperforms other unsupervised methods and is also able to be used as a pre-training method to enhance supervised path representation learning.

Weakly-supervised Temporal Path Representation Learning with Contrastive Curriculum Learning -- Extended Version

“…In addition, their path representations are task specific. Most recently, the unsupervised path representation learning framework PIM [20] learns path representations. However, it does not include temporal information.…”

“…Recently, the most effective approaches for learning representations with or without labeled data have been supervised or unsupervised contrastive learning [20], [24], [29]- [32], which have shown impressive performance in computer vision and graph learning. As a form of metric learning [33], contrastive approaches achieve representations in a discriminating manner through contrasting positive data pairs against negative data pairs.…”

“…This illustrates the potential of path representation learning for improving intelligent transportation applications. Indeed, initial studies of path representation learning [20], [21] already exist.…”

“…Unsupervised approaches do not rely on task specific labels and are thus able to offer generic path representations. Existing unsupervised path representation approaches rely heavily on unsupervised graph representation learning, where the representations of the edges in a path are aggregated into a path representation [20], [24]. Since existing graph representation learning does not consider temporal information, the obtained path representations also lacks the temporal aspect.…”

“…• We report on extensive experiments using three realworld data sets in the settings of three downstream tasks to assess in detail the effectiveness of the proposed framework. A preliminary version [1] has been accepted by the 38th IEEE International Conference on Data Engineering (ICDE 2022). This version offers more detailed technical specifics and additional experimental results.…”