Traditional knowledge graphs (KGs) provide each entity with a unique embedding as a representation, which contains a lot of redundant information. Meanwhile, the space complexities of the KGs are positively related to the number of entities. In this work, we propose a hierarchical representation learning method, namely HRL, which is a parameter-efficient model where the number of model parameters is independent of dataset scales. Specifically, we propose a hierarchical model comprising a Meta Encoder and a Context Encoder to generate the representation of entities and relations. The Meta Encoder captures the common representations shared across entities, while the Context Encoder learns entity-specific representations. We further provide a theoretical analysis of model design by constructing a structural causal model (SCM) when completing a knowledge graph. The SCM outlines the relationships between nodes, where entity embeddings are conditioned on both common and entity-specific representations. Note that our model is designed to reduce model scale while maintaining competitive performance. We evaluate HRL on the knowledge graph completion task using three real-world datasets. The results demonstrate that HRL significantly outperforms existing parameter-efficient baselines, as well as traditional state-of-the-art baselines of similar scale.

Urban fine-grained data map inference, leveraging information from coarse-grained maps, has emerged as a significant area of research due to the growing complexity and data heterogeneity in urban environments.Existing methods have a priori assumption that a coarse-grained data map, one fixed-size granularity, transforms into a fine-grained data map, also one fixed-size granularity.However, in actual scenarios, the collected coarse-grained data maps are often incomplete and have significantly distinct granularities in various urban areas, which results in incomplete heterogeneous data, i.e., multi-granularity data maps in terms of spatial information.Meanwhile, different granularity data maps are needed for various urban downstream tasks, which is a multi-task problem.To that end, this paper proposes a novel framework, a multi-granularity super-resolution data map inference framework (MGSR), designed to harness spatio-temporal information to transform incomplete coarse-grained multi-granularity data maps into fine-grained multigranularity data maps.Specifically, we design a granularity alignment network to align multi-granularity information and address missing data on each granularity data map by leveraging the other granularity data maps with a well-designed self-supervised task.Then, we introduce a feature extraction network to capture spatiotemporal dependencies and extract features.Finally, we devise a recurrent super-resolution network with shared parameters to infer multi-granularity data maps.We conduct extensive experiments on three real-world benchmark datasets and demonstrate that MGSR significantly outperforms the state-of-the-art methods for multigranularity urban data map inference and reduces RMSE and MAE by up to 40.1% and 50.3%, respectively.

Passengers move between urban places for diverse interests and drive the metropolitan regions as the aggregation of urban places to group into network communities. This paper aims to examine the relationship between the spatial patterns (represented by the network communities) of mobility flows and places of interest (POIs). Furtherly, it intends to identify the categories of POIs that play the most significant role in shaping the spatial patterns of mobility flows. To achieve these purposes, we partition the study area into disjoint regions and construct the network with each partitioned region as a node and connection between them as links weighted by the mobility flows. The community detection algorithm is implemented on the network to discover spatial mobility patterns, and the multiclass classification based on the logistic regression method is adopted to classify spatial communities featured by POIs. Taking the taxi systems of Shanghai and Beijing as examples, we detect spatial communities based on the movement strengths among regions. Then we investigate their correlations with POIs. It finds that communities’ modularity correlates linearly with POIs; particularly governments, hotels, and the traffic facilities are of the most significance for generating the mobility patterns. This study can provide valuable insight into understanding the spatial mobility patterns from the perspective of POIs.

Radio maps play a vital role in fingerprint-based indoor positioning systems (IPSs) in terms of the localization accuracy and computational overheads. Most existing studies either directly eliminate redundant APs or adopt unsupervised dimension reduction methods, say principal component analysis (PCA), to obtain a low-dimension representation of fingerprints, which consumes less storage and computational overheads. In this paper, we propose to reduce the dimensions of radio maps based on the Gaussian Process Manifold Kernel Dimension Reduction (GPMKDR) which is a supervised dimension reduction technique in comparison with the well known PCA-based method. Specifically, GPMKDR is employed to find a nonlinear and optimal embedding into the received signal strength (RSS) sample space during the offline phase, such that any RSS sample vector obtained in the online localization phase can be projected onto the optimal subspace with a lower dimension, with the result that the fingerprint-based localization can be efficiently realized based on a low-dimension radio map. Experiments show that the nonlinear GPMKDR-based method significantly improves the localization performance in comparison with the PCA-based method.