Point-of-Interest (POI) recommendation has been a trending research topic as it generates personalized suggestions on facilities for users from large number of candidate venues. Since users' check-in records can be viewed long sequence, methods based recurrent neural networks (RNNs) have recently shown promising applicability this task. However, existing RNN-based either neglect long-term preferences or overlook the geographical relations among visited POIs when modeling short-term...

The state-of-the-art methods in aspect-level sentiment classification have leveraged the graph based models to incorporate syntactic structure of a sentence. While being effective, these ignore corpus level word co-occurrence information, which reflect collocations linguistics like “nothing special”. Moreover, they do not distinguish different types dependency, e.g., nominal subject relation “food-was” is treated equally as an adjectival complement “was-okay” “food was okay”. To tackle above...

Aspect-level sentiment classification aims to determine the polarity of a sentence towards an aspect. Due high cost in annotation, lack aspect-level labeled data becomes major obstacle this area. On other hand, document-level like reviews are easily accessible from online websites. These encode knowledge abundant contexts. In paper, we propose Transfer Capsule Network (TransCap) model for transferring classification. To end, first develop aspect routing approach encapsulate sentence-level...

Aspect-based sentiment analysis (ABSA) involves three subtasks, i.e., aspect term extraction, opinion and aspect-level classification. Most existing studies focused on one of these subtasks only. Several recent researches made successful attempts to solve the complete ABSA problem with a unified framework. However, interactive relations among are still under-exploited. We argue that such encode collaborative signals between different subtasks. For example, when is “delicious”, must be “food”...

The increasing proliferation of location-based social networks brings about a huge volume user check-in data, which facilitates the recommendation points interest (POIs). Time and location are two most important contextual factors in user’s decision-making for choosing POI to visit. In this article, we focus on spatiotemporal context-aware recommendation, considers joint effect time recommendation. Inspired by recent advances knowledge graph embedding, propose translation-based recommender...

Attributed networks are ubiquitous in the real world, such as social networks. Therefore, many researchers take node attributes into consideration network representation learning to improve downstream task performance. In this article, we mainly focus on an untouched "oversmoothing" problem research of attributed learning. Although Laplacian smoothing has been applied by state-of-the-art works learn a more robust representation, these cannot adapt topological characteristics different...

Matrix completion is a classic problem underlying recommender systems. It traditionally tackled with matrix factorization. Recently, deep learning based methods, especially graph neural networks, have made impressive progress on this problem. Despite their effectiveness, existing methods focus modeling the user-item interaction graph. The inherent drawback of such that performance bound to density interactions, which however usually high sparsity. More importantly, for cold start user/item...

Aspect category sentiment analysis (ACSA) is an underexploited subtask in aspect level analysis. It aims to identify the of predefined categories. The main challenge ACSA comes from fact that may not occur sentence most cases. For example, review “ they have delicious sandwiches ” positively talks about food implicit manner. In this article, we propose a novel aware learning (AAL) framework for tasks. Our key idea exploit interaction between and contents under guidance both polarity To end,...

Aspect term extraction (ATE) aims to extract aspect terms from a review sentence that users have expressed opinions on. Existing studies mostly focus on designing neural sequence taggers linguistic features the token level. However, since and context words usually exhibit long-tail distributions, these often converge an inferior state without enough sample exposure. In this paper, we propose tackle problem by correlating with each other through soft prototypes. These prototypes, generated...

Querying cohesive subgraphs on temporal graphs (e.g., social network, finance etc.) with various conditions has attracted intensive research interests recently. In this paper, we study a novel Temporal <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$(k,\mathcal {X})$</tex-math></inline-formula> -Core Query (TXCQ) that extends fundamental notation="LaTeX">$k$</tex-math></inline-formula> (TCQ) proposed in our...

Reasoning is fundamental to human intelligence, and critical for problem-solving, decision-making, thinking. refers drawing new conclusions based on existing knowledge, which can support various applications like clinical diagnosis, basic education, financial analysis. Though a good number of surveys have been proposed reviewing reasoning-related methods, none them has systematically investigated these methods from the viewpoint their dependent knowledge base. Both scenarios bases are...

Graph is ubiquitous in various real-world applications, and many graph processing systems have been developed. Recently, hardware accelerators exploited to speed up systems. However, such hardware-specific are hard migrate across different backends. In this paper, we propose the first tensor-based framework, Tgraph, which can be smoothly deployed run on any powerful (uniformly called XPU) that support Tensor Computation Runtimes (TCRs). TCRs, deep learning frameworks along with their...

This paper studies the problem of identifying users who use multiple userids to post in social media. Since may belong same author, it is hard directly apply supervised learning solve problem. proposes a new method, which still uses but does not require training documents from involved userids. Instead, other for classifier building. The can be applied possible because we transform document space similarity and performed this space. Our evaluation done online review domain. experimental...

Signed directed networks with positive or negative links convey rich information such as like dislike, trust distrust. Existing work of sign prediction mainly focuses on triangles (triadic nodes) motivated by balance theory to predict and links. However, real-world signed can contain a good number "bridge'' edges which, definition, are not included in any triangles. Such ignored previous work, but may play an important role network analysis.%Such serve fundamental building blocks analysis.

Recommender systems have played a vital role in online platforms due to the ability of incorporating users' personal tastes. Beyond accuracy, diversity has been recognized as key factor recommendation broaden user's horizons well promote enterprises' sales. However, trading-off between accuracy and remains be big challenge, data user biases not explored yet. In this paper, we develop an adaptive learning framework for accurate diversified recommendation. We generalize recent proposed...

End-to-end aspect based sentiment analysis (E2E-ABSA) aims to jointly extract terms and predict aspect-level for opinion reviews. Though supervised methods show effectiveness E2E-ABSA tasks, the annotation cost is extremely high due necessity of fine-grained labels. Recent attempts alleviate this problem using domain adaptation technique transfer word-level common knowledge across domains. However, biggest issue in adaptation, i.e., how domain-specific words like <italic...

Querying cohesive subgraphs on temporal graphs with various time constraints has attracted intensive research interests recently. In this paper, we study a novel Temporal k -Core Query (TCQ) problem: given interval, find all distinct -cores that exist within any subintervals from graph, which generalizes the previous historical -core query. This problem is challenging because number of increases quadratically to span interval. For that, propose Core Decomposition (TCD) algorithm...

Despite the recent success achieved by several two-stage prototypical networks in few-shot named entity recognition (NER) task, over-detected false spans at span detection stage and inaccurate unstable prototypes type classification remain to be challenging problems. In this paper, we propose a novel Type-Aware Decomposed framework, namely TadNER, solve these We first present type-aware filtering strategy filter out removing those semantically far away from names. then contrastive learning...