In task-oriented dialogue systems, Dialogue State Tracking (DST) aims to extract users' intentions from the history. Currently, most existing approaches suffer error propagation and are unable dynamically select relevant information when utilizing previous states. Moreover, relations between updates of different slots provide vital clues for DST. However, rely only on predefined graphs indirectly capture relations. this paper, we propose a Distillation Network (DSDN) utilize states migrate...

Recently, the graph neural network (GNN) has shown great power in matrix completion by formulating a rating as bipartite and then predicting link between corresponding user item nodes. The majority of GNN-based methods are based on Graph Autoencoder (GAE), which considers one-hot index input, maps (or item) to learnable embedding, applies GNN learn node-specific representations these embeddings finally aggregates target users its nodes predict missing links. However, without node content...

The cross-lingual sentiment analysis (CLSA) aims to leverage label-rich resources in the source language improve models of a resource-scarce domain target language, where monolingual approaches based on machine learning usually suffer from unavailability knowledge. Recently, transfer paradigm that can knowledge resource-rich languages, for example, English, resource-poor Chinese, has gained particular interest. Along this line, article, we propose semisupervised with SCL and space...

In the task of cross-language sentiment classification, monolingual machine learning based approaches suffer from shortage available resources in target language. order to reduce cost labeling documents a new language, many proposed transfer knowledge resource-rich languages (e.g. English) resource-poor Chinese). Although labeled data are only source utilization information language is often disregarded. this paper, we propose semi-supervise approach with space tackle above task. The main...

Abstract With the increasingly extensive applications of network, security internal network enterprises is facing more and threats from outside world, which implies importance to master risk assessment skills. To improve accuracy an importent issue. In big data era, there are various protection techniques different types group data. Meanwhile, Online Social Networks (OSNs) Internet Things (SIoT) becoming popular patterns meeting people keeping in touch with friends (Jiang et al. ACM Comput...

In mobile crowdsourcing (MCS), the platform selects participants to complete location-aware tasks from recruiters aiming achieve multiple goals (e.g., profit maximization, energy efficiency, and fairness). However, different MCS systems have there are possibly conflicting even in one system. Therefore, it is crucial design a participant selection algorithm that applies goals. To deal with this issue, we formulate problem as reinforcement learning propose solve novel method, which call...

Abstract In video streaming, bandwidth constraints significantly affect client-side quality. Addressing this, deep neural networks offer a promising avenue for implementing super-resolution (VSR) at the user end, leveraging advancements in modern hardware, including mobile devices. The principal challenge VSR is computational intensity involved processing temporal/spatial data. Conventional methods, uniformly entire scenes, often result inefficient resource allocation. This evident...

We consider the problem of imitation learning from suboptimal demonstrations that aims to learn a better policy than demonstrators. Previous methods usually reward function encode underlying intention demonstrators and use standard reinforcement based on this function. Such can fail control distribution shift between learned since may not generalize well out-of-distribution samples mislead agent highly uncertain states, resulting in degenerated performance. To address limitation, we propose...

Recently, spoken dialogue systems have been widely deployed in a variety of applications, serving huge number end-users. A common issue is that the errors resulting from noisy utterances, semantic misunderstandings, or lack knowledge make it hard for real system to respond properly, possibly leading an unsatisfactory user experience. To avoid such case, we consider proactive interaction mechanism where predicts satisfaction with candidate response before giving user. If not likely be...

Recently, the graph neural network (GNN) has shown great power in matrix completion by formulating a rating as bipartite and then predicting link between corresponding user item nodes. The majority of GNN-based methods are based on Graph Autoencoder (GAE), which considers one-hot index input, maps (or item) to learnable embedding, applies GNN learn node-specific representations these embeddings finally aggregates target users its nodes predict missing links. However, without node content...