基于大数据的高效并行推荐系统

协同过滤算法广泛应用于电子商务网站的推荐系统中,它基于对大量用户历史行为数据的分析,挖掘用户的兴趣并向用户推荐合适的产品.为了能够处理大数据文件,使用Apache Mahout、MapReduce并行框架和和协同过滤进行有效并行查询处理.实验结果表明,该系统能够在大数据集上实现高效率和高可靠性.系统不仅限于电影推荐,还...     

基于随机森林修正的加权二部图推荐算法

对于基于二部图网络结构的算法忽视了兴趣偏好的影响,只考虑用户与项目之间的关系,结合随机森林分类模型和二部图网络结构,提出了一种基于随机森林修正的加权二部图推荐算法。在二部图网络结构的基础上,利用评分计算边权,充分考虑项目的度和用户共同评分项目的影响改进相似度公式。同时用随机森林算法对用户在项目特征的偏好构建分类模型,根据其对初步得出的推荐列表进行评分修正。对比在Movie Lens数据集上的实验结果,证明该方法比其他算法能够提高推荐的准确性和推荐精度。     

一种基于用户交易行为的隐语义模型推荐算法

通过分析目前推荐技术在电子商务系统中的应用优势,并针对当前产品交易系统的无评分、产品量大和难以分类等现状与问题,设计了一种基于用户交易行为的隐语义模型推荐算法。该算法从用户的隐式交易行为出发,采用隐语义模型推荐算法,构建用户-产品兴趣模型,并加入K均值算法划分隐式特征聚类。实验验证表明,该算法在满足用户的个性化需求的同时,可提高电子商务系统的产品推荐效率。     

一种基于耦合对象相似度的项目推荐算法

推荐系统根据用户的偏好为用户推荐个性化的信息、产品和服务等,能够帮助用户有效解决信息过载问题。基于内容的协同过滤算法缺少合适的度量指标用来计算项目之间的相似度。提出一种基于耦合对象相似度的项目推荐算法,即通过耦合对象相似度捕获项目特征频率分布相似性和特征依赖聚合相似度。首先从项目文本中抽取项目的关键特征,然后利用耦合对象相似度构建项目相似度模型,最后使用协同过滤的方法为活动用户推荐用户可能感兴趣的项目。在真实数据集上的实验结果表明,基于耦合对象相似度的推荐算法可以有效解决基于内容推荐系统的项目相似度度量问题,在缺失大量项目特征数据的情况下改进传统基于内容推荐系统的推荐质量。     

融合用户兴趣表征与注意力机制的推荐算法

为利用用户行为挖掘用户的兴趣,提出一种融合用户兴趣表征与注意力机制的推荐算法.利用CVR算法将传统的用户-项目表征转换为用户-兴趣表征;构建一种应用于用户兴趣预测的深度森林模型,引入兴趣簇重要性作为特征选择权重,融合时间注意力机制进行兴趣预测,将用户-兴趣模型结合基于用户的协同过滤算法预测推荐结果.两个数据集上的实验结果表明,该算法能够提高用户兴趣预测准确率,提升推荐效果.     

基于改进RFM模型的客户筛选及协同过滤推荐算法

大数据背景下，电商平台迅速崛起，如何准确识别高价值客户，并向消费者推荐合适的产品，是平台运营商关注的重点，而精准挖掘用户行为规律是提高推荐系统准确率的关键。本文在RFM模型的基础上，针对电商平台退货率、客单价、客单件和用户活跃时长4个核心特征，提出RFM-RUVA模型，筛选高价值客户，并结合客户实例分析，验证RFM-RUVA模型的有效性。此外，将商品的协同过滤算法和基于用户的协同过滤算法相结合，提出基于商品&用户的协同过滤组合算法（Item&User＿CF），实验结果表明Item&User＿CF相较于基于商品、用户的协同过滤算法准确率分别提高了8.94%、2.63%，可以更准确地为高价值客户推荐商品。     

大数据驱动的投诉预测模型

随着电信行业市场竞争的不断加剧,用户对服务质量要求逐步提高,导致用户投诉率不断攀升。在此情况下,通过准确预测用户投诉行为来降低用户投诉率成为运营商关注的重点。目前传统的投诉预测模型仅从分类算法和人工调研特征来讨论,而没有充分利用运营商的大数据。因此,提出了在Hadoop/Spark大数据平台上使用并行随机森林来构建用户预测投诉模型,它不仅用到了业务支持系统数据,而且还用到了运营支持系统数据和客服工单数据,并在此基础上进一步增加了反映用户相互关系的图特征和二阶特征。基于上海市某运营商数据的实验结果表明,利用多来源、高维度的特征来训练用户投诉预测模型的精度会明显高于传统方法,在此基础上有针对性地对目标用户采取安抚措施,可以降低用户投诉率,获得较高的商业价值。     

深度随机森林在离网预测中的应用

在电信运营商领域,离网预测模型是企业决策者用来发现潜在离网用户(即停用运营商服务)的主要手段。目前离网预测模型都是基于逻辑回归、决策树、神经网络及随机森林等浅层机器学习算法,但是在大数据的背景下,这些浅层算法在预测问题上很难取得更高的精度。因此,提出了一种新型的深层结构模型——深度随机森林,通过将传统浅层随机森林堆积成深层结构模型,获得更高的预测精度。在运营商真实数据上进行了大量实验,结果证明深层随机森林模型比传统浅层机器学习算法在离网预测问题上可以得到更好的效果。同时,增大训练数据量可以进一步提升深层随机森林的预测能力,从而证明了在大数据环境下深层模型的潜力。     

基于聚类和随机森林的协同过滤推荐算法

针对基于邻近关系的协同过滤算法在线推荐效率低的问题,提出了一种可离线训练评分预测模型的算法。通过聚类算法降低用户-项目评分矩阵中用户向量和项目向量的维数,并对数据进行转换使其适用于监督模型;利用转换后的数据离线训练随机森林模型,在线推荐时只需根据随机森林模型的规则进行评分预测,无需查找最邻近用户或项目。实验结果表明,该算法在不降低评分预测精度的情况下,在线推荐效率远高于基于邻近关系的协同过滤算法。     

一种基于用户播放行为序列的个性化视频推荐策略

本文针对在线视频服务网站的个性化推荐问题,提出了一种基于用户播放行为序列的个性化推荐策略.该策略通过深度神经网络词向量模型分析用户播放视频行为数据,将视频映射成等维度的特征向量,提取视频的语义特征.聚类用户播放历史视频的特征向量,建模用户兴趣分布矩阵.结合用户兴趣偏好和用户观看历史序列生成推荐列表.在大规模的视频服务系统中进行了离线实验,相比随机算法、基于物品的协同过滤和基于用户的协同过滤传统推荐策略,本方法在用户观看视频的Top-N推荐精确率方面平均分别获得22.3%、30.7%和934%的相对提升,在召回率指标上分别获得52.8%、41%和1065%的相对提升.进一步地与矩阵分解算法SVD++、基于双向LSTM模型和注意力机制的Bi-LSTM+Attention算法和基于用户行为序列的深度兴趣网络DIN比较,Top-N推荐精确率和召回率也得到了明显提升.该推荐策略不仅获得了较高的精确率和召回率,还尝试解决传统推荐面临大规模工业数据集时的数据要求严苛、数据稀疏和数据噪声等问题.     

服务号码捆绑特征在离网预测系统中的应用

用户流失问题是电信运营商面临的亟待解决的问题,针对不同的场景,业界研究开发了多个用户离网预测系统。服务号码捆绑指用户在使用运营商服务期间,与银行、电商、便利店等第三方服务提供商通过绑定手机号产生联系。通过研究发现用户在服务存续期间普遍会绑定多种第三方服务提供商,这些商家会不定时给用户推送短信,当用户即将流失时,多数用户会逐渐取消这类服务的绑定。因此,服务号码捆绑特征对于离网用户的甄别起到了重要的作用。采用随机森林算法构建离网预测模型,利用逻辑回归算法对服务号码捆绑特征进行降维,并加入模型,进行离网用户分析,从而辅助决策者制订相应的客户维挽策略,降低客户离网率。实验结果表明,服务号码软捆绑特征能够提高系统的分析预测能力。     

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.     

基于客户喜好的双向个性化推荐算法

为了解决现有推荐算法仅考虑同类产品间单向推荐所缺乏的灵活性,提升产品的销量及用户的购物体验,提出一种基于客户喜好的双向个性化推荐算法,不仅可以为客户精准推荐产品,还可以为商家推荐潜在客户.首先,基于产品购买网络中客户及其邻居的购买信息,扩展客户购买信息;其次设计客户产品喜好权重计算办法,分析客户的购买喜好,并在客户喜好的指导下为客户提供个性化的产品推荐;最后,基于商家提供的样本客户,挖掘与样本客户相似的客户构成社区,为商家提供潜在客户推荐以及精准客户维护.在真实数据集上的实验验证了算法的有效性.该算法从客户和商家两个维度出发实现了产品与客户的双向推荐,为个性化推荐领域的研究提供有益的帮助.     

基于RFM模型的随机森林算法对民航客户的流失分析

近几年,随着航空市场的快速发展,对于航空公司而言,如何在增加市场占有率的同时,对客户的流失进行有效的控制也刻不容缓.基于随机森林算法,根据航空客户数据,建立流失预测模型,对客户是否已流失进行预测研究,将传统的RFM客户价值模型进行改进,结合随机森林算法对客户流失进行预测.实验结果表明,基于RFM模型的随机森林算法构建的...     

Ontology-based personalized couple clustering for heterogeneous product recommendation in mobile marketing

Plethora of cellular phones has been increasingly driving the spread of e-commerce mechanisms running on mobile devices. For instance, mobile marketing fulfills the wireless delivery (to the devices of mobile users) of the recommended product information and even one-to-one recommendations. One-to-one recommendation not only reduces the time that customers have to expend to for attaining appropriate products, but also is a method to engender customer values and develop the long-term customer relationships. This paper presents a one-to-one recommendation mechanism that iteratively takes as inputs the audio customer messages (together with product information) and produces personalized product analogy structures (that subsequently drive the generation of personalized heterogeneous product recommendations) based on the coupled clustering algorithm. The personalized product analogy structures also evolve as the messages (of the correspondent customer) grow. We have implemented the mechanism with J2EE Web Service that has produced fairly promising evaluation results.     

The recommender system for virtual items in MMORPGs based on a novel collaborative filtering approach

The recommendation system for virtual items in massive multiplayer online role-playing games (MMORPGs) has aroused the interest of researchers. Of the many approaches to construct a recommender system, collaborative filtering (CF) has been the most successful one. However, the traditional CFs just lure customers into the purchasing action and overlook customers’ satisfaction, moreover, these techniques always suffer from low accuracy under cold-start conditions. Therefore, a novel collaborative filtering (NCF) method is proposed to identify like-minded customers according to the preference similarity coefficient (PSC), which implies correlation between the similarity of customers’ characteristics and the similarity of customers’ satisfaction level for the product. Furthermore, the analytic hierarchy process (AHP) is used to determine the relative importance of each characteristic of the customer and the improved ant colony optimisation (IACO) is adopted to generate the expression of the PSC. The IACO creates solutions using the Markov random walk model, which can accelerate the convergence of algorithm and prevent prematurity. For a target customer whose neighbours can be found, the NCF can predict his satisfaction level towards the suggested products and recommend the acceptable ones. Under cold-start conditions, the NCF will generate the recommendation list by excluding items that other customers prefer.     

Does reliable information matter? Towards a trustworthy co-created recommendation model by mining unboxing reviews

Online review forums provide customers with powerful platforms to express opinions and influence business trends, while allowing firms to collaborate and co-create value with customers. However, information overload due to the huge amount of reviews posted daily complicates the efforts of consumers to locate reliable information when making a purchase decision. Therefore, this study develops a trustworthy co-created recommendation model. The proposed model mines unboxing reviews, calculates the trust scores of the reviewers, and then generates the recommended products by combing this information with customer preferences using a multi-criteria decision-making method. An illustrative example of mobile phones demonstrates the recommendation procedure of the proposed model. The proposed model is evaluated via an empirical experiment to examine the satisfaction of study participants by using a seven-point Likert scale. An analysis of the structural equation modelling results indicates that three factors (i.e. confidence in decision quality, enhanced problem-solving ability, and satisfaction with resource expenditure) significantly and positively affect the purchase decision-making process. Moreover, the proposed model outperforms a baseline model in all four factors, ultimately increasing user satisfaction. In addition to its theoretical framework for co-creating value with customers to develop a trustworthy co-created recommendation model, as supported by various theories of trust, the proposed model provides further insights into the role of customer reviews in designing recommendation models, as well as the extent to which such models impact user decisions.     

Rank order-based recommendation approach for multiple featured products

Recommendation methods, which suggest a set of products likely to be of interest to a customer, require a great deal of information about both the user and the products. Recommendation methods take different forms depending on the types of preferences required from the customer. In this paper, we propose a new recommendation method that attempts to suggest products by utilizing simple information, such as ordinal specification weights and specification values, from the customer. These considerations lead to an ordinal weight-based multi-attribute value model. This model is well suited to situations in which there exist insufficient data regarding the demographics and transactional information on the target customers, because it enables us to recommend personalized products with a minimal input of customer preferences. The proposed recommendation method is different from previously reported recommendation methods in that it explicitly takes into account multidimensional features of each product by employing an ordered weight-based multi-attribute value model. To evaluate the proposed method, we conduct comparative experiments with two other methods rooted in distance-based similarity measures.     

推荐系统中的新颖性问题研究

准确性推荐中存在商品类型单一、流行商品多、缺乏新意的问题,因而新颖性推荐得到重视。但已有研究在设计算法时未考虑项的特征,无法针对不同用户帮其区分和挑选具备较高新颖度的项。为提高推荐系统的性能,对基于随机游走的方法进行改进,提出融合新颖性特征的推荐算法。从兴趣扩展和预测角度分析项的特征,给出完善的新颖度定义,并结合用户需求构建新的转移概率,产生个性化的推荐列表,提高了列表内容的新意。实验结果表明,提出的算法较现有算法对准确率影响较小,同时在新颖性指标上有明显提升,并得出通过融合新颖性特征能够在兼顾准确性的情况下有效改善推荐内容的结论。     

多子域随机森林在情境感知推荐中的应用研究

情境感知推荐系统通过增加情境信息来提高推荐精度,在实际应用中得到广泛的应用。然而,传统的情境感知推荐方法存在赋予情境因素相同权重,忽略了用户在不同情境下所偏好项目的不同,以及情境因素在推荐过程中所起的影响作用不同的问题。提出一种基于多子域随机森林算法的情境感知推荐方法。该方法对特征重要性按权值大小进行排序,将权值的取值区域分为多个大小相等的子区域,在这些子区域中随机选择特征,构造特征子空间来改进随机森林算法;通过改进的随机森林算法来分解并降低用户、项目和情境的特征维度;使用协同过滤推荐算法来进行冷链物流配载个性化推荐。对LDOS-CoMoDa和Cycle Share两个数据集进行仿真实验,结果表明该方法相比传统方法平均绝对误差减少近10%,有效地提高了推荐系统的预测精度,为情境感知推荐的应用提供借鉴。