Heterogeneous graphs with different types of nodes and edges are ubiquitous have immense value in many applications. Existing works on modeling heterogeneous usually follow the idea splitting a graph into multiple homogeneous subgraphs. This is ineffective exploiting hidden rich semantic associations between for large-scale multi-relational graphs. In this paper, we propose Relation Structure-Aware Graph Neural Network (RSHN), unified model that integrates its coarsened line to embed both...

User cold-start recommendation is a long-standing challenge for recommender systems due to the fact that only few interactions of users can be exploited. Recent studies seek address this from perspective meta learning, and most them follow manner parameter initialization, where model parameters learned by steps gradient updates. While these gradient-based meta-learning models achieve promising performances some extent, fundamental problem how adapt global knowledge previous tasks...

Applying existing methods to emotional support conversation—which provides valuable assistance people who are in need—has two major limitations: (a) they generally employ a conversation-level emotion label, which is too coarse-grained capture user's instant mental state; (b) most of them focus on expressing empathy the response(s) rather than gradually reducing distress. To address problems, we propose novel model \textbf{MISC}, firstly infers fine-grained status, and then responds...

Relation extraction studies the issue of predicting semantic relations between pairs entities in sentences. Attention mechanisms are often used this task to alleviate inner-sentence noise by performing soft selections words independently. Based on observation that information pertinent is usually contained within segments (continuous a sentence), it possible make use phenomenon for better extraction. In paper, we aim incorporate such segment into neural relation extractor. Our approach views...

Session-based recommendation (SBR) aims to predict a user's next clicked item based on an anonymous yet short interaction sequence. Previous SBR models, which rely only the limited short-term transition information without utilizing extra valuable knowledge, have suffered lot from problem of data sparsity. This paper proposes novel mirror graph enhanced neural model for session-based (MGS), exploit attribute over embeddings more accurate preference estimation.

Existing emotion-aware conversational models usually focus on controlling the response contents to align with a specific emotion class, whereas empathy is ability understand and concern feelings experience of others. Hence, it critical learn causes that evoke users' for empathetic responding, a.k.a. causes. To gather in online environments, we leverage counseling strategies develop an chatbot utilize causal information. On real-world dataset, verify effectiveness proposed approach by...

The East Asian monsoon (EAM) and the Australian (AUM) are two subsystems of Asian-Australian system. EAM AUM can be linked dynamically through cross-equatorial outflow, in addition to their own distinct responses external forcing. Despite previous studies on relationships for different timescales, relationship at orbital timescale has remained poorly explored. In a set simulations, we demonstrate that Summer Monsoon (AUSM) precipitation varies out-of-phase (EASM) precession due local...

Cross-lingual Entity Alignment (CEA) aims at identifying entities with their counterparts in different language knowledge graphs. Knowledge embedding alignment plays an important role CEA due to its advantages of easy implementation and run-time robustness. However, existing methods haven't considered the problem distribution which refers spatial shapes spaces. To this end, we present a new Adversarial Embedding framework (AKE for short) that jointly learns representation, mapping...

The development of AI in mental health is a growing field with potential global impact. Machine agents need to perceive users' states and respond empathically. Since are often latent implicit, building such chatbots requires both knowledge learning utilization. Our work contributes this by developing chatbot that aims recognize empathetically states. We introduce Conditional Variational Autoencoders (CVAE)-based model utilizes relation-aware commonsense generate responses. This model, while...

Graph Neural Networks (GNNs) have achieved state-of-the-art performance in many graph data analysis tasks. However, they still suffer from two limitations for representation learning. First, exploit non-smoothing node features which may result suboptimal embedding and degenerated classification. Second, only neighbor information but ignore global topological knowledge. Aiming to overcome these simultaneously, this paper, we propose a novel, flexible, end-to-end framework, Smoothing Splines...

Code generation aims to automatically generate a piece of code given an input natural language utterance. Currently, among dominant models, it is treated as sequence-to-tree task, where decoder outputs sequence actions corresponding the pre-order traversal Abstract Syntax Tree. However, such only exploits based preceding actions, which are insufficient ensure correct action predictions. In this paper, we first throughly analyze context modeling difference between neural models with different...

Graph classification plays an important role in a wide range of applications from biological prediction to social analysis. Traditional graph models built on kernels are hampered by the challenge poor generalization as they heavily dependent dedicated design handcrafted features. Recently, neural networks (GNNs) become new class tools for analyzing data and have achieved promising performance. However, it is necessary collect large number labeled training accurate GNN, which often...

Conversational recommendation system aims to recommend appropriate items user by directly asking preference on attributes or recommending item list. However, most of existing methods only employ the flat and attribute relationship, ignore hierarchical relationship connected similar which can provide more comprehensive information. And these usually use accepted represent conversational history information sequential transition in historical turns. In this paper, we propose Hierarchical...

In few-shot relational triple extraction (FS-RTE), one seeks to extract triples from plain texts by utilizing only few annotated samples. Recent work first extracts all entities and then classifies their relations. Such an entity-then-relation paradigm ignores the entity discrepancy between To address it, we propose a novel task decomposition strategy, Relation-then-Entity, for FS-RTE. It detects relations occurred in sentence corresponding head/tail of detected instantiate this further...

The lack of sufficient training data in many domains, poses a major challenge to the construction domain-specific machine reading comprehension (MRC) models with satisfying performance. In this paper, we propose novel iterative multi-source mutual knowledge transfer framework for MRC. As an extension conventional one-to-one correspondence, our focuses on many-to-many transfer, which involves synchronous executions multiple many-to-one transfers manner.Specifically, update target-domain MRC...

Open Information Extraction (OpenIE) aims to discover textual facts from a given sentence. In essence, the contained in plain text are unordered. However, popular OpenIE systems usually output sequentially way of predicting next fact conditioned on previous decoded ones, which enforce an unnecessary order and involve error accumulation between autoregressive steps. To break this bottleneck, we propose MacroIE, novel non-autoregressive framework for OpenIE. MacroIE firstly constructs graph...

Recommender systems have been widely adopted in various online personal e-commerce applications for improving user experience. A long-standing challenge recommender is how to provide accurate recommendation users cold-start situations where only a few user-item interactions can be observed. Recently, meta learning methods promising solution, and most of them follow way parameter initialization predictions fast adapted via multiple gradient descent steps. While these meta-learning...

Semantic parsing is a mainstream method of knowledge base question answering task that first generates set logical forms according to and (KB), then selects the most matching one get answers. However, existing selection methods are usually based on word-level matching, which cannot capture structural information or solve long-term dependency problem entities. To this problem, we propose syntax-based graph method, explicitly models both form as graphs, performs at structure-level. The...

In multi-turn dialog, utterances do not always take the full form of sentences (Carbonell 1983), which naturally makes understanding dialog context more difficult. However, it is essential to fully grasp generate a reasonable response. Hence, in this paper, we propose improve response generation performance by examining model's ability answer reading comprehension question, where question focused on omitted information dialog. Enlightened multi-task learning scheme, joint framework that...

Cross-domain slot filling is a challenging task in spoken language understanding due to the differences text genre across domains. In this paper, we attempt solve by exploiting syntactic structures of user utterances, because these are actually accessible and can be shared between utterances from different To end, propose novel Syntactic Structure Encoder (SSE) module incorporate it into detection-prediction framework. SSE introduces graph convolutional network (GCN) learn common multiple...

Starting from an abelian group $G$ and a factorizable ribbon Hopf $G$-bialgebra $H$, we construct TQFT $J_H$ for connected framed cobordisms between surfaces with boundary decorated cohomology classes coefficients in $G$. When restricted to the subcategory of trivial decorations, our functor recovers special case Kerler-Lyubashenko TQFTs, namely those associated algebras. Our result is inspired by work Blanchet-Costantino-Geer-Patureau, who constructed non-semisimple TQFTs admissible using...

Recent deep learning based methods have achieved impressive performance on paraphrase identification (PI), a fundamental NLP task, judging whether two sentences are semantically equivalent or not. However, their success heavily relies massive labeled samples, which time-consuming and expensive to obtain. To alleviate this problem, study explores the effect of word alignment information (WAI), extracted by existing monolingual tools, PI baseline models. Apart from directly encoding WAI into...

Unsupervised cross-domain NER task aims to solve the issues when data in a new domain are fully-unlabeled. It leverages labeled from source predict entities unlabeled target domain. Since training models on large corpus is time-consuming, this paper, we consider an alternative way by introducing syntactic dependency structure. Such information more accessible and can be shared between sentences different domains. We propose novel framework with dependency-aware GNN (DGNN) learn these common...