基于超图卷积网络和目标多意图感知的会话推荐算法

当前先进的会话推荐算法主要通过图神经网络从全局和目标会话中挖掘项目的成对转换关系,并将目标会话压缩成固定的向量表示,忽略了项目间复杂的高阶信息和目标项目对用户偏好多样性的影响。为此提出了基于超图卷积网络和目标多意图感知的会话推荐算法HCN-TMP。通过学习会话表示来表达用户偏好,首先依据目标会话构建会话图,依据全局会话构建超图,通过意图解纠缠技术将原有反映用户耦合意图的项目嵌入表示转换为项目多因素嵌入表示,再经图注意力网络和超图卷积网络分别学习目标会话节点的会话级和全局级项目表示,并使用距离相关性损失函数增强多因素嵌入块间的独立性;然后嵌入目标会话中节点位置信息,加权每个节点的注意力权重,得到全局级和会话级会话表示;利用对比学习最大化两者互信息,经目标多意图感知,针对不同的目标项目自适应地学习目标会话中多意图的用户偏好,得到目标感知级会话表示,最后线性融合三个级别的会话表示得到最终的会话表示。在Tmall和Nowplaying两个公开数据集上进行大量实验,实验结果验证了HCN-TMP算法的有效性。     

基于多模态知识图的用户微行为会话推荐

基于会话推荐的目标是根据给定的会话预测下一个交互项.针对现有基于会话推荐大多只关注某个用户交互项的连续项来捕获相关信息;以及利用知识图谱提供辅助信息的过程中忽略了多模态知识图中各种数据类型的问题.本文提出多模态知识图的用户微行为的会话模型,将用户微行为和多模态知识纳入基于会话推荐的多任务学习中,用户微行为对项目的建模在微观行为级别上,使用一系列项目操作对来充分捕获会话中的用户意图.提出多模态知识图注意力网络,通过使用多模态图注意机制进行信息传播,将得到的聚集嵌入表示进行推荐.考虑候选项目信息来平衡局部和全局表示的门控模块,门控模块利用多层感知器层对候选项目、最近交互项目和每个用户的全局偏好之间的关系进行建模来计算权重.在两个基准数据集的实验结果表明,本文所提模型优于最新基于会话的推荐,有利于提升推荐的准确率.     

低频查询的用户行为分析和类别研究

低频查询是用户提交查询频次非常低的查询.它们占了搜索引擎独立查询的很大比例且对用户体验影响巨大,但由于数据稀疏性,现有的搜索引擎用户行为分析及相关研究中对低频查询涉及很少.结合前人的相关工作,使用商业搜索引擎的大规模用户日志,在会话级别上进行低频查询的用户行为分析以及类别研究.基于目标查询行为、后续相关查询行为、整体会话行为3个方面的12个特征进行了低频查询的用户行为特征分析,首次提出了低频查询类别分析框架,并进一步使用改进的AdaBoost算法对低频查询会话进行分类.实验对2000个随机的低频查询会话样例进行分类,AUC值达到了83%以上.低频查询的用户行为分析和类别研究,将为搜索引擎用户行为分析等网络检索研究提供重要基础.     

基于用户级别的进程调度策略的研究与实现

阐述了Linux 2.6内核的O（1）进程调度策略,分析了它的不足,提出了基于用户级别的进程调度策略,根据需要定义系统内用户级别的个数,通过给不同用户指派不同的级别,区别对待不同级别用户创建的进程,首先满足高级别用户的需求,使高级别用户创建的进程获得更低的静态优先级static_prio和更长的运行时间片timeslice,各级别的用户得到与其级别相适应的服务,更加体现了进程调度的公平性原则.实验结果表明,在原有资源的基础上,基于用户级别的进程调度策略有效地改善了高级别用户获得的服务.     

一种会话理解模型的问题生成方法

会话问题生成(Conversational Question Generation,CQG)不同于根据段落和答案生成单轮问题的问题生成任务,CQG额外考虑由历史问答对构成的会话信息,生成的问题承接会话历史内容,保持较高的一致性.针对这一特性,文中提出了字级别和句级别注意力机制模块来增强对会话历史信息的提取能力,确保当前...     

面向会话推荐的注意力图神经网络

面向会话的推荐方式起源于无法获得用户历史数据的应用场景,它是通过匿名会话来预测用户的行为.现有面向会话的推荐方法,虽然可以准确获得项目嵌入和考虑项目的复杂转换,但不能从多维度提取会话序列中隐藏的用户的长期兴趣和短期偏好,造成推荐性能低.该文引入注意力机制,提出一种多头注意力机制和软注意力机制有机结合的新机制,并据此提出面向会话推荐的注意力图神经网络.该注意力机制通过给不同的输入数据赋予不同权重,实现对当前推荐任务更为关键的信息的聚焦,以此从不同角度提取用户的兴趣和偏好.该模型在电商数据集上进行实验,与已有的基准模型相比,该文所提模型在各项评论指标上均有显著提升.在Dgeca数据集上,P@20可达61.77%,充分表明了所提方法的有效性.     

Web日志挖掘中的数据预处理技术

数据预处理是Web日志挖掘中的重要步骤,一般分为数据清理、用户识别、会话识别和路径补充。为消除代理服务器、防火墙和本地缓存对Web日志带来的影响,采用基于引用的分析方法完成用户会话识别和路径补充。实验结果表明,在Web访问日志中的记录引用信息较完整的情况下,该方法可以高效地获得用户的访问路径。     

Web日志挖掘中的会话识别方法

为更好地实现会话识别,从而为后续模式发现提供准确的挖掘数据,在分析现有常用的会话识别方法后,提出一种基于待挖掘站点首页的用户会话识别方法.该方法根据用户浏览站点的习惯,以站点首页作为用户新会话开始标识,并增强了用户会话的定义.最后利用数据库编程实现该方法,将识别出的会话存储在数据库中,便于后续数据挖掘使用.实验结果表明,该方法能识别出更多的用户会话,且识别会话的准确率也更高.     

认证群密钥协商协议编译器的分析与改进

认证群密钥协商协议能为群用户产生一个共享的会话密钥,使群用户在公共数据网络中进行安全通信。该文分析证明Abdalla等人的密钥协商协议不能抵抗假冒攻击：某用户的左右邻居在获得与该用户运行群密钥交换协议副本后,可以假冒该用户与其他群成员运行该协议,产生一个新的共享会话密钥,且不会被群中其他成员发现。并针对该协议编译器的缺点进行改进。     

一种基于SIP的升空平台通信系统的优化切换

在升空平台通信系统中,当移动用户终端移动到另一个平台时,它的IP地址会发生改变。当用户终端改变接入网络时,能够自动获得一个新的IP地址。用户终端被重新分配一个新的地址之后,要向归属域进行重新IMS注册,而且向通信对端节点发送重新会话邀请请求RE_INVITE,这样会话才能继续。这样的做法会给正在进行的会话带来很长的中断和延时。通过引入一种新的通过共享用户注册信息和会话状态信息的解决方法来降低切换时延。在重新注册和重新建立会话请求时,并不需要所有的REGISTER和INVITE流,因为IMS服务器已经通过服务器之间的信令传输获得用户状态信息,这样能减少切换时延,在切换时信令包的数量和重建通信连接的延时方面作了理论分析,并在OPNET仿真环境下进行了检验。     

基于VRML的产品虚拟模型协同操作方法研究

针对基于Web的虚拟设计中用户和企业设计人员交流的需要,提出了采用VRML对产品的虚拟模型进行同步浏览和协同设计的新方法,并设计了相应的控制方式.同时结合矢量化的电子图板、文字及语音通信对产品的虚拟模型进行协同操作和讨论,实现了协同操作的系统原型.     

保护信道和排队相结合的信道分配策略

提出了一种适用于语音、数据呼叫的蜂窝移动通信系统的信道分配策略。该策略为数据呼叫提供保护信道,降低数据呼叫的阻塞率。同时,采取语音呼叫排队策略抑制数据保护信道引起的语音呼叫阻塞率的恶化。为了进一步提高系统的性能,在策略中引入了不耐烦顾客,并建立了带有不耐烦顾客的排队模型。仿真结果表明该策略能够有效地降低语音呼叫和数据呼叫阻塞率,改善系统性能。     

Discovering patterns of online purchasing behaviour and a new‐product‐launch strategy

Spurred by rapid development of computers and Internet technology, online shopping is gradually overtaking in‐store shopping, because of advantages such as convenience, more choice of products or services etc. Online stores must devote a great deal of time and resources to locating and attracting new customers. Growing a customer base requires first understanding customers and then providing the products or services they need, thus encouraging customers to purchase more. This paper develops a system to analyse customers’ purchasing behaviour and track shifts in their interests. Customers’ purchasing behaviour is measured using proposed standard product loyalty status and standard brand loyalty status. Using these metrics, together with the preference map established for each customer, a marketing specialist can easily locate potential customers to target when a company launches a new product. The new‐product‐launch strategy proposed in this paper can be used to create a list of potential customers for a product being launched under a variety of conditions. A prototype system has been built to test the feasibility of the proposed new‐product‐launch strategy. The result shows almost 40% of potential customers respond to the recommendation positively.     

Pricing and inventory strategies under quick response with strategic and myopic consumers

We divide consumers in the selling period into two types according to their purchasing behavior: strategic customers and myopic customers. We address the optimal inventory and pricing decision problem of a retailer considering strategic and myopic consumers with and without a quick response. The results indicate that the retailer should establish a higher price to sell only to myopic customers if there is a sufficient presence of myopic customers in the market, and the retailer should set a lower price to sell to myopic and strategic customers if the number of myopic customers in the market is relatively low. A quick response can decrease the initial ordering quantity and increase the retailer's profit when selling only to myopic customers or selling to both myopic and strategic customers. Moreover, a quick response is beneficial for the retailer to improve the product's retail price if the retailer wishes the two types of customers to purchase the product during the selling period. We also find that the pricing strategy considering strategic and myopic consumers under a quick response will increase profits more than the pricing strategy only considering myopic customers. Finally, numerical experiments are conducted to illustrate and validate the proposed models and provide managerial insights.     

基于自适应遗传神经网络的银行客户分类研究

银行产品的营销行为都是针对广大客户的。若能提前分辨出哪些是优质客户,再为其定制合理的营销策略,那银行就能获得更大的竞争力。文中将遗传算法与BP神经网络结合用于对银行客户分类进而预测客户是否会购买银行产品。该方法有效地克服了BP神经网络容易陷入局部极小值和收敛速度慢的问题,并且针对其中遗传算法的计算时间和精度问题提出了一种新的自适应遗传算法。实验结果表明,基于这种自适应的遗传神经网络的方法用更短的计算时间达到了更高的预测精度,可以准确地为银行客户分类。     

Risk assessment of co-creating value with customers: A rough group analytic network process approach

Co-creating value with customers is becoming an important competition strategy for companies. It provides a feasible way to meet customers’ personalized requirements. However, the strategy needs effective management to obtain benefits because it may involve many potential risks. Neglecting these risks may lead to extra-budgetary costs, wrong decision of product investment or loss of customers. To the best of our knowledge, little attention has been paid so far to the risk assessment of co-creating value with customers.This study mainly focuses on assessing the risks of co-creating value with customers under uncertainty. First, the scattered literature is combed systemically to determine the risk factors of co-creating with customers. Then, a novel approach called the rough group analytic network process is proposed to assess these risks quantitatively. The proposed approach can intelligently handle decision maker's subjectivity and vagueness, and the interdependences/feedbacks among risk factors. Finally, an industrial case study is presented to illustrate the application of the proposed approach, and the proposed approach is compared with other existing methods to demonstrate its the advantages.     

User security enhancement: A worldwide security project

The Society for Worldwide Interbank Financial Telecommunication (S.W.I.F.T.) provides financial processing and communication services of the highest quality and integrity. Its network and value-added services enable customers to reduce costs, raise productivity and control risks. Security is one of the key benefits that S.W.I.F.T. offers to customers, and the User Security Enhancement (USE) programme is a major part of the company's security strategy.     

改进k-means 算法在电信CRM 客户分类中的应用

电信市场的竞争在于客户的竞争,为了在激烈的竞争中保持优势,必须将客户进行细分,针对不同的客户,研究出相应的营销策略．K-means算法能对大型数据集进行高效分类,但对“噪声”敏感,聚类结果不准确,本文对该算法进行改进,使其能够实现更加准确和全面的客户细分．     

Expectation-based coopetition approach to service experience design

Service experience design is a key issue in the service industry. Satisfactory service experiences provide customers with good memories and increase customer satisfaction, thus increasing customer loyalty and service provider profitability. However, providing customers with quality service experiences requires considering numerous factors, and thus is a complex and difficult issue both in academia and in real service contexts. Customer satisfaction results from the difference between customer perceptions and customer expectations. Hence, customer expectation management is important in delivering quality service experiences. Additionally, a competitive relationship exists among different service providers as well as between service providers and customers. The service experience delivery process can be regarded as a value co-creation process for service providers and customers who can implement either a competitive or cooperative strategy based on their goals and needs. This study accordingly presents an expectation based coopetition approach by using a real exhibition data of AutoTronics 2009 for simulations. The evaluation results show that this expectation-based coopetition approach can help service providers design and deliver quality service experiences and co-create value with customers, yielding a high performance ecosystem.     

Feature-based recommendations for one-to-one marketing

Most recommendation systems face challenges from products that change with time, such as popular or seasonal products, since traditional market basket analysis or collaborative filtering analysis are unable to recommend new products to customers due to the fact that the products are not yet purchased by customers. Although the recommendation systems can find customer groups that have similar interests as target customers, brand new products often lack ratings and comments. Similarly, products that are less often purchased, such as furniture and home appliances, have fewer records of ratings; therefore, the chances of being recommended are often lower. This research attempts to analyze customers' purchasing behaviors based on product features from transaction records and product feature databases. Customers' preferences toward particular features of products are analyzed and then rules of customer interest profiles are thus drawn in order to recommend customers products that have potential attraction with customers. The advantage of this research is its ability of recommending to customers brand new products or rarely purchased products as long as they fit customer interest profiles; a deduction which traditional market basket analysis and collaborative filtering methods are unable to do. This research uses a two-stage clustering technique to find customers that have similar interests as target customers and recommend products to fit customers' potential requirements. Customers' interest profiles can explain recommendation results and the interests on particular features of products can be referenced for product development, while a one-to-one marketing strategy can improve profitability for companies.