Geographical characteristics derived from the historical check-in data have been reported effective in improving location recommendation accuracy. However, previous studies mainly exploit geographical a user's perspective, via modeling distribution of each individual check-ins. In this paper, we are interested exploiting by neighborhood location. The is modeled at two levels: instance-level defined few nearest neighbors location, and region-level for region where exists. We propose novel...

Trajectory similarity computation is fundamental functionality with many applications such as animal migration pattern studies and vehicle trajectory mining to identify popular routes similar drivers. While a continuous curve in some spatial domain, e.g., 2D Euclidean space, trajectories are often represented by point sequences. Existing approaches that compute based on matching suffer from the problem they treat two different sequences differently even when represent same trajectory. This...

Knowledge graph (KG) plays an increasingly important role in recommender systems. Recently, neural networks (GNNs) based model has gradually become the theme of knowledge-aware recommendation (KGR). However, there is a natural deficiency for GNN-based KGR models, that is, sparse supervised signal problem, which may make their actual performance drop to some extent. Inspired by recent success contrastive learning mining signals from data itself, this paper, we focus on exploring KG-aware and...

Multivariate Time Series (MTS) forecasting plays a vital role in wide range of applications. Recently, Spatial-Temporal Graph Neural Networks (STGNNs) have become increasingly popular MTS methods due to their state-of-the-art performance. However, recent works are becoming more sophisticated with limited performance improvements. This phenomenon motivates us explore the critical factors and design model that is as powerful STGNNs, but concise efficient. In this paper, we identify...

This paper addresses the problem of temporal sentence grounding (TSG), which aims to identify boundary a specific segment from an untrimmed video by query. Previous works either compare pre-defined candidate segments with query and select best one ranking, or directly regress timestamps target segment. In this paper, we propose novel localization framework that scores all pairs start end indices within simultaneously biaffine mechanism. particular, present Context-aware Biaffine Localizing...

Humans can only interact with part of the surrounding environment due to biological restrictions. Therefore, we learn reason spatial relationships across a series observations piece together environment. Inspired by such behavior and fact that machines also have computational constraints, propose COnditional COordinate GAN (COCO-GAN) which generator generates images parts based on their coordinates as condition. On other hand, discriminator learns justify realism multiple assembled patches...

Session-based recommendation (SBR) is a challenging task, which aims at recommending items based on anonymous behavior sequences. Almost all the existing solutions for SBR model user preference only current session without exploiting other sessions, may contain both relevant and irrelevant item-transitions to session. This paper proposes novel approach, called Global Context Enhanced Graph Neural Networks (GCE-GNN) exploit item transitions over sessions in more subtle manner better inferring...

Bundle recommendation aims to recommend the user a bundle of items as whole. Previous models capture user’s preferences on both and association items. Nevertheless, they usually neglect diversity intents adopting fail disentangle in representations. In real scenario recommendation, intent may be naturally distributed different bundles that (Global view). And contain multiple (Local Each view has its advantages for disentangling: 1) global view, more are involved present each intent, which...

Incorporating Knowledge Graphs (KG) into recommeder system as side information has attracted considerable attention. Recently, the technical trend of Knowledge-aware Recommendation (KGR) is to develop end-to-end models based on graph neural networks (GNNs). However, extremely sparse user-item interactions significantly degrade performance GNN-based models, from following aspects: 1) interaction, itself, means inadequate supervision signals and limits supervised models; 2) combination (CF...

Aspect-based sentiment analysis (ABSA) is a fine-grained task that aims to align aspects and corresponding sentiments for aspect-specific polarity inference. It challenging because sentence may contain multiple or complicated (e.g., conditional, coordinating, adversative) relations. Recently, exploiting dependency syntax information with graph neural networks has been the most popular trend. Despite its success, methods heavily rely on tree pose challenges in accurately modeling alignment of...

Cross-modal hashing has been widely used in multimedia retrieval tasks due to its fast speed and low storage cost. Recently, many deep unsupervised cross-modal methods have proposed deal the unlabeled datasets. These usually construct an instance similarity matrix by fusing image text modality-specific matrices as guiding information train networks. However, most of them directly use cosine similarities between bag-of-words (BoW) vectors datapoints define matrix, which fails mine semantic...

Aspect Sentiment Triplet Extraction (ASTE) has become an emerging task in sentiment analysis research, aiming to extract triplets of the aspect term, its corresponding opinion and associated polarity from a given sentence. Recently, many neural networks based models with different tagging schemes have been proposed, but almost all them their limitations: heavily relying on 1) prior assumption that each word is only single role (e.g., or etc. ) 2) word-level interactions treating...

Multivariate time series forecasting (MTSF) is crucial for decision-making to precisely forecast the future values/trends, based on complex relationships identified from historical observations of multiple sequences. Recently, Spatial-Temporal Graph Neural Networks (STGNNs) have gradually become theme MTSF model as their powerful capability in mining spatial-temporal dependencies, but almost them heavily rely assumption data integrity. In reality, due factors such collector failures and...

Most of the unsupervised hashing methods usually map images into semantic similarity-preserving hash codes by constructing local similarity structure as guiding information, i.e., treating each point similar to its k nearest neighbours. However, for an image, some neighbours may be dissimilar it, they are noisy datapoints which will damage retrieval performance. Thus, tackle this problem, in paper, we propose a novel deep method, called MLS3RDUH, can reduce further enhance Specifically,...

Due to their high retrieval efficiency and low storage cost, cross-modal hashing methods have attracted considerable attention. Generally, compared with shallow methods, deep can achieve a more satisfactory performance by integrating feature learning hash codes optimizing into same framework. However, most existing either cannot learn unified code for the two correlated data-points of different modalities in database instance or guide feedback function procedure, enhance accuracy. To address...

Representing and rendering dynamic scenes has been an important but challenging task. Especially, to accurately model complex motions, high efficiency is usually hard guarantee. To achieve real-time scene while also enjoying training storage efficiency, we propose 4D Gaussian Splatting (4D-GS) as a holistic representation for rather than applying 3D-GS each individual frame. In 4D-GS, novel explicit containing both 3D Gaussians neural voxels proposed. A decomposed voxel encoding algorithm...

Graph ranking plays an important role in many applications, such as page on web graphs and entity social networks. In besides graph structure, rich information nodes edges explicit or implicit human supervision are often available. contrast, conventional algorithms (e.g., PageRank HITS) compute scores by only resorting to structure information. A natural question arises here, that is, how effectively efficiently leverage all the more accurately calculate than algorithms, assuming is also...

The growth of the Web in recent years has resulted development various online platforms that provide healthcare information services. These contain an enormous amount information, which could be beneficial for a large number people. However, navigating through such knowledgebases to answer specific queries consumers is challenging task. A majority might non-factoid nature, and hence, traditional keyword-based retrieval models do not work well cases. Furthermore, many scenarios, it desirable...

Minhao Cheng, Wei Wei, Cho-Jui Hsieh. Proceedings of the 2019 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019.

Answering questions in many real-world applications often requires complex and precise information excerpted from texts spanned across a long document. However, currently no such annotated dataset is publicly available, which hinders the development of neural question-answering (QA) systems. To this end, we present MASH-QA, Multiple Answer Spans Healthcare Question consumer health domain, where answers may need to be multiple, non-consecutive parts text We also propose MultiCo, architecture...

Incorporating Knowledge Graphs into Recommendation has attracted growing attention in industry, due to the great potential of KG providing abundant supplementary information and interpretability for underlying models. However, simply integrating recommendation usually brings negative feedback ignorance following two factors: i) users' multiple intents, which involve diverse nodes KG. For example, e-commerce scenarios, users may exhibit preferences specific styles, brands, or colors. ii)...