随着会话推荐的广泛应用,如何充分利用语义信息、建模用户跨会话兴趣以及抑制数据噪声成为提升推荐性能的关键。为此提出一种新颖的会话推荐增强框架LGSBR,通过整合大语言模型(large language model,LLM)的语义理解能力与图神经网络(gra...随着会话推荐的广泛应用,如何充分利用语义信息、建模用户跨会话兴趣以及抑制数据噪声成为提升推荐性能的关键。为此提出一种新颖的会话推荐增强框架LGSBR,通过整合大语言模型(large language model,LLM)的语义理解能力与图神经网络(graph neural network,GNN)的结构建模能力,实现语义增强与个性化推荐。具体而言,利用大语言模型及微调的语言模型生成项目补充文本嵌入和用户跨会话兴趣嵌入,通过软注意力机制融合文本与ID嵌入,生成语义丰富的项目表示;引入用户兴趣嵌入,结合对齐损失实现个性化推荐;最后通过两阶段权重学习过滤噪声项目,优化会话表示。实验结果表明,在Beauty数据集上,LGSBR的P@20达到21.38%,MRR@20达到6.76%,分别较SR-GNN基线提升23.3%和50.56%;在MovieLen-1M数据集上,P@20为25.86%,MRR@20为7.58%,分别提升12.63%和10.98%;研究验证了LGSBR在多种GNN模型上的通用性和有效性。展开更多
This paper aims at analyzing the security issues that lie in the application layer (AL) protocols when users connect to the Internet via a wireless local area network (WLAN) through an access point. When adversaries l...This paper aims at analyzing the security issues that lie in the application layer (AL) protocols when users connect to the Internet via a wireless local area network (WLAN) through an access point. When adversaries launch deauthentication flood attacks cutting users' connection, the connection managers will automatically research the last access point's extended service set identifier (ESSID) and then re-establish connection. However, such re-connection can lead the users to a fake access point with the same ESSID set by attackers. As the attackers hide behind users' access points, they can pass AL's authentication and security schemes, e.g. secure socket layer (SSL). We have proved that they can even spy on users' account details, passwords, data and privacy.展开更多
The number of Internet users has increased very rapidly due to the scalability of the network. The users demand for higher bandwidth and better throughput in the case of on demand video, or video conference or any rea...The number of Internet users has increased very rapidly due to the scalability of the network. The users demand for higher bandwidth and better throughput in the case of on demand video, or video conference or any real time distributed network system. Performance is a great issue in any distributed networks. In this paper we have shown the performance of the multicast groups or clusters in the distributed network system. In this paper we have shown the performance of different users or receivers belongs to the multicast group or cluster in the distributed network, transfer data from the source node with multirate multicast or unirate multicast by considering packet level forwarding procedure in different sessions. In this work we have shown that how the throughput was effected when the number of receiver increase. In this work we have considered the different types of queue such as RED, Fair queue at the junction node for maintaining the end to end packet transmission. In this work we have used various congestion control protocol at the sender nodes. This paper we have shown the performance of the distributed cluster network by multirate multicast.展开更多
A novel framework for remote service discovery and access of IP cameras with Network address Translation (NAT) traversal is presented in this paper. The proposed protocol, termed STDP (Service Trader Discovery Protoco...A novel framework for remote service discovery and access of IP cameras with Network address Translation (NAT) traversal is presented in this paper. The proposed protocol, termed STDP (Service Trader Discovery Protocol), is a hybrid combination of Zeroconf and SIP (Session Initial Protocol). The Zeroconf is adopted for the discovery and/or publication of local services;whereas, the SIP is used for the delivery of local services to the remote nodes. In addition, both the SIP-ALG (Application Layer Gateway) and UPnP (Universal Plug and Play)-IGD (Internet Gateway Device) protocols are used for NAT traversal. The proposed framework is well-suited for high mobility applications where the fast deployment and low administration efforts of IP cameras are desired.展开更多
Environmental sustainability issues and the costs of new power generation and transmission have increased the interest in evolving current power grid to new technologies. The Smart Grid is a promising technology, sinc...Environmental sustainability issues and the costs of new power generation and transmission have increased the interest in evolving current power grid to new technologies. The Smart Grid is a promising technology, since it allows a distributed computing approach with potentials for self-diagnosing/-healing, reliable multi-user communication and fast hard real-time control. However, the missing standardization associated with heterogeneity of legacy systems and wide-area service demands, makes very challenging to adopt Smart Grid in a cost-effective way. By considering this, we propose the Session-Oriented Communication System (SOCSys) to overcome the above issues by enhancing Smart Grid with truly reliable and robust capabilities over heterogeneous environments. SOCSys achieves this goal by orchestrating session-control with innovative network-centric facilities operating over a wireless mesh Information Network compliant with IEEE 802.11e/s standard. The simulation results show that SOCSys improved network performance in terms of bandwidth utilization and minimization of delay, while consuming low network resources. Graphical analyses showed that SOCSys supported multimedia sessions with excellent quality, where it outperforms the experiments with regular settings.展开更多
目的现有的基于图神经网络(GNN)的推荐方法忽略了会话中有价值用户在项目上的时间驻留信息,无法解决用户无意识点击带来的影响,同时忽略图神经网络中隐藏因素的表达能力,针对以上问题,提出一种融合时间驻留信息的图神经网络会话推荐模型...目的现有的基于图神经网络(GNN)的推荐方法忽略了会话中有价值用户在项目上的时间驻留信息,无法解决用户无意识点击带来的影响,同时忽略图神经网络中隐藏因素的表达能力,针对以上问题,提出一种融合时间驻留信息的图神经网络会话推荐模型(Graph Neural Network Session-based Recommendation Based on Fusion of Time Resident Information,TRGNN)。方法首先,对用户在各个项目上的驻留时间信息进行处理,通过时间图神经网络得到时间特征;其次,应用多头注意力机制增强因素的表达能力更好地提取项目特征,TRGNN将时间特征与项目特征进行融合得到最终特征,通过注意力网络得到全局上下文和局部上下文;最后,通过预测层得到最终推荐结果。结果在Diginetica和Yoochoose两个真实数据集上进行对比实验,实验结果表明:相较于最优基线模型,本模型在Mrr@20评价指标下分别提升了1.57%和3.30%,在Recall@20指标下分别提升了1.10%和0.66%。结论本模型实现了更好的推荐效果,能更好地挖掘隐藏信息,充分应用时间特征和项目隐藏特征来提高推荐准确率,降低用户误触对推荐准确率的影响。展开更多
文摘随着会话推荐的广泛应用,如何充分利用语义信息、建模用户跨会话兴趣以及抑制数据噪声成为提升推荐性能的关键。为此提出一种新颖的会话推荐增强框架LGSBR,通过整合大语言模型(large language model,LLM)的语义理解能力与图神经网络(graph neural network,GNN)的结构建模能力,实现语义增强与个性化推荐。具体而言,利用大语言模型及微调的语言模型生成项目补充文本嵌入和用户跨会话兴趣嵌入,通过软注意力机制融合文本与ID嵌入,生成语义丰富的项目表示;引入用户兴趣嵌入,结合对齐损失实现个性化推荐;最后通过两阶段权重学习过滤噪声项目,优化会话表示。实验结果表明,在Beauty数据集上,LGSBR的P@20达到21.38%,MRR@20达到6.76%,分别较SR-GNN基线提升23.3%和50.56%;在MovieLen-1M数据集上,P@20为25.86%,MRR@20为7.58%,分别提升12.63%和10.98%;研究验证了LGSBR在多种GNN模型上的通用性和有效性。
基金the National Science Council (No. NSC-99-2219-E-033-001)the Foundation of the Chung Yuan Christian University (1004) (No. CYCU-EECS.9801)
文摘This paper aims at analyzing the security issues that lie in the application layer (AL) protocols when users connect to the Internet via a wireless local area network (WLAN) through an access point. When adversaries launch deauthentication flood attacks cutting users' connection, the connection managers will automatically research the last access point's extended service set identifier (ESSID) and then re-establish connection. However, such re-connection can lead the users to a fake access point with the same ESSID set by attackers. As the attackers hide behind users' access points, they can pass AL's authentication and security schemes, e.g. secure socket layer (SSL). We have proved that they can even spy on users' account details, passwords, data and privacy.
文摘The number of Internet users has increased very rapidly due to the scalability of the network. The users demand for higher bandwidth and better throughput in the case of on demand video, or video conference or any real time distributed network system. Performance is a great issue in any distributed networks. In this paper we have shown the performance of the multicast groups or clusters in the distributed network system. In this paper we have shown the performance of different users or receivers belongs to the multicast group or cluster in the distributed network, transfer data from the source node with multirate multicast or unirate multicast by considering packet level forwarding procedure in different sessions. In this work we have shown that how the throughput was effected when the number of receiver increase. In this work we have considered the different types of queue such as RED, Fair queue at the junction node for maintaining the end to end packet transmission. In this work we have used various congestion control protocol at the sender nodes. This paper we have shown the performance of the distributed cluster network by multirate multicast.
文摘A novel framework for remote service discovery and access of IP cameras with Network address Translation (NAT) traversal is presented in this paper. The proposed protocol, termed STDP (Service Trader Discovery Protocol), is a hybrid combination of Zeroconf and SIP (Session Initial Protocol). The Zeroconf is adopted for the discovery and/or publication of local services;whereas, the SIP is used for the delivery of local services to the remote nodes. In addition, both the SIP-ALG (Application Layer Gateway) and UPnP (Universal Plug and Play)-IGD (Internet Gateway Device) protocols are used for NAT traversal. The proposed framework is well-suited for high mobility applications where the fast deployment and low administration efforts of IP cameras are desired.
文摘Environmental sustainability issues and the costs of new power generation and transmission have increased the interest in evolving current power grid to new technologies. The Smart Grid is a promising technology, since it allows a distributed computing approach with potentials for self-diagnosing/-healing, reliable multi-user communication and fast hard real-time control. However, the missing standardization associated with heterogeneity of legacy systems and wide-area service demands, makes very challenging to adopt Smart Grid in a cost-effective way. By considering this, we propose the Session-Oriented Communication System (SOCSys) to overcome the above issues by enhancing Smart Grid with truly reliable and robust capabilities over heterogeneous environments. SOCSys achieves this goal by orchestrating session-control with innovative network-centric facilities operating over a wireless mesh Information Network compliant with IEEE 802.11e/s standard. The simulation results show that SOCSys improved network performance in terms of bandwidth utilization and minimization of delay, while consuming low network resources. Graphical analyses showed that SOCSys supported multimedia sessions with excellent quality, where it outperforms the experiments with regular settings.
文摘目的现有的基于图神经网络(GNN)的推荐方法忽略了会话中有价值用户在项目上的时间驻留信息,无法解决用户无意识点击带来的影响,同时忽略图神经网络中隐藏因素的表达能力,针对以上问题,提出一种融合时间驻留信息的图神经网络会话推荐模型(Graph Neural Network Session-based Recommendation Based on Fusion of Time Resident Information,TRGNN)。方法首先,对用户在各个项目上的驻留时间信息进行处理,通过时间图神经网络得到时间特征;其次,应用多头注意力机制增强因素的表达能力更好地提取项目特征,TRGNN将时间特征与项目特征进行融合得到最终特征,通过注意力网络得到全局上下文和局部上下文;最后,通过预测层得到最终推荐结果。结果在Diginetica和Yoochoose两个真实数据集上进行对比实验,实验结果表明:相较于最优基线模型,本模型在Mrr@20评价指标下分别提升了1.57%和3.30%,在Recall@20指标下分别提升了1.10%和0.66%。结论本模型实现了更好的推荐效果,能更好地挖掘隐藏信息,充分应用时间特征和项目隐藏特征来提高推荐准确率,降低用户误触对推荐准确率的影响。
