In recent years, deep neural networks have yielded immense success on speech recognition, computer vision and natural language processing. However, the exploration of deep neural networks on recommender systems has received relatively less scrutiny. In this work, we strive to develop techniques based on neural networks to tackle the key problem in recommendation -- collaborative filtering -- on the basis of implicit feedback. Although some recent work has employed deep learning for recommendation, they primarily used it to model auxiliary information, such as textual descriptions of items and acoustic features of musics. When it comes to model the key factor in collaborative filtering -- the interaction between user and item features, they still resorted to matrix factorization and applied an inner product on the latent features of users and items. By replacing the inner product with a neural architecture that can learn an arbitrary function from data, we present a general framework named NCF, short for Neural network-based Collaborative Filtering. NCF is generic and can express and generalize matrix factorization under its framework. To supercharge NCF modelling with non-linearities, we propose to leverage a multi-layer perceptron to learn the user-item interaction function. Extensive experiments on two real-world datasets show significant improvements of our proposed NCF framework over the state-of-the-art methods. Empirical evidence shows that using deeper layers of neural networks offers better recommendation performance.
Collaborative filtering recommendation system based on user similarity has been wildly studied because of its broad application. In reality, users keep partial similarity with larger possibility. Computing the whole similarity between users without considering item category is inaccurate when predicting rating for a special category of items by using collaborative filtering recommendation system. Aiming at this problem, a new similarity measurement was given. Based on the new similarity measurement, a new collaborative filtering algorithm named UICF was presented for recommendation. When predicting rating for the special item, UICF chooses the users as nearest neighbors which have the similar rating feature for the items with the same type of the special item, instead of for all the items. Experimental results show the higher quality of the algorithm. Keywords: Recommendation system, collaborative filtering, item classification.Recommendation system has become an important research field of e-commerce IT technology. At present, many kinds of recommendation algorithm have been given by researchers such as Bayesian network、cluster algorithm, association rules, horting based on graph-theoretic, collaborative filtering recommendation algorithm and so on. Bayesian network creates decision tree models relates to recommendation by training set[1]. In the decision tree model, user information is represented by nodes and edges. Cluster algorithm gathers the users with similar taste into one cluster[2,3] and then the rating of target user for special items is predicted according to the rating of users in the same cluster. Online cluster algorithm can generate recommendation with a high speed by completing clustering process offline. Recommendation system based on association rules recommends items to the target user in line with its current behavior and association rules model[4]. Association rules model can be generated offline, so recommendation system based on association rules can ensure real-time requirements. Horting based on graph-theoretic[5] is a recommendation method based on graphic with its nodes representing users and its edges representing similarity between users. And it gives recommendation for target user by searching its neighbor nodes and then integrating the ratings of neighbor nodes.
Among these recommendation algorithms, collaborative filtering, hereinafter referred to as CF, is the most successful one[6]. The basic idea with traditional collaborative filtering algorithms is that rating of items which are not rated by user is predicted based on rating data of the userrsquo;s nearest neighbors and then the item with the highest predicted rating is recommended to the user. In order to find out nearest neighbors for the target users, a method of measuring the similarity between users is necessary. But with the expansion of e-commerce systems and the sharp increase of users and items, the rating data is becoming extremely sparse, which reduces the accuracy of nearest neighbors computed for the object user and then makes collaborative filtering algorithm generate bad recommendation. To address this issue, a number of improved methods have been given by researchers, such as itembased collaborative filtering algorithm and its improvement[7-9], Collaborative filtering based on Matrix dimensionality reduction[10-11] and Collaborative Filtering Based on cloud model[12-13]. All these Methods mentioned above can resolve data sparsity and improve the accuracy of user similarity to a certain extent, but they all ignore a important issue that is users keep partial similarity with larger possibility. Computing the whole similarity between users is inaccurate when predicting rating for special type of items by using collaborative filtering recommendation system. Such as, user A and user B maybe keep the same taste for the S type items , maybe user A and user B are all very like the S type items or vice versa. But for T type items, there are large differences between their tastes, maybe user A likes the T types items very much and it is just Opposite to user B. To address this issue,this paper presents a new Collaborative filtering recommendation algorithm based partial user similarity considering the item types.
The rest of the paper is structured as follows. Section 2 introduces the main framework of collective filtering recommendation system based on us
剩余内容已隐藏，支付完成后下载完整资料
近年来，深层神经网络在语音识别、计算机视觉和自然语言处理等方面取得了巨大的成功。然而，在推荐系统上对深层神经网络的探索却得到了相对较少的关注。在这项工作中，我们努力开发基于神经网络的技术，在隐式反馈的基础上解决推荐中的关键问题——协作过滤。尽管最近的一些工作采用了深度学习的推荐，他们主要使用它来建模辅助信息，如项目的文字描述和音乐的声学特征。在构建协同过滤的关键因素----用户和项特征之间的相互作用时，他们仍然采用矩阵因数分解，并在用户和项的潜在特征上应用了一个内积。通过用一个可以从数据中学习任意函数的神经结构来代替内积，我们提出了一个名为ncf的通用框架，它是基于神经网络的协作过滤的简称。ncf是通用的，可以在其框架下表达和推广矩阵因式分解。为了对非线性的ncf模型进行超荷，我们提出利用多层感知器来学习用户项交互功能。在两个实际数据集上进行的大量实验表明，我们提出的ncf框架比最先进的方法有了很大的改进。经验表明，使用更深层次的神经网络提供了更好的推荐性能。
基于用户相似性的协同过滤推荐系统由于其广泛的应用而得到了广泛的研究。在现实生活中，用户保持部分相似度的可能性更大。在不考虑项类别的情况下计算用户之间的整体相似性是不准确的，在预测一个特殊类别的项的评级时，使用协同过滤推荐系统。针对这一问题，提出了一种新的相似度测量方法。在新的相似度测量的基础上，提出了一种名为uicf的协同滤波算法作为推荐。在预测特殊项目的评级时，uicf选择用户作为最近的邻居，对具有相同类型特殊项目的项目具有类似的评级特征，而不是对所有的项目。实验结果表明，该算法具有较高的质量。关键字：推荐系统、协同过滤、项目分类，推荐系统已成为电子商务信息技术的一个重要研究领域。目前，研究者们给出了很多类型的推荐算法，如贝叶斯网络、聚类算法、关联规则、基于图论的点阵算法、协同滤波推荐算法等。贝叶斯网络创建决策树模型与推荐相关的培训集[1]。在决策树模型中，用户信息由节点和边缘表示。聚类算法将具有相似趣味的用户聚集到一个聚类[2,3]中，然后根据同一聚类中用户的评分来预测特殊物品的目标用户的评分。在线聚类算法可以通过离线完成聚类过程，以较快的速度生成推荐。
基于关联规则的推荐系统根据目标用户当前的行为和关联规则模型向目标用户推荐项[4]。关联规则模型可以离线生成，因此基于关联规则的推荐系统可以保证实时需求。基于图形理论的点阵法是一种基于图形的推荐方法，其节点代表用户，边缘代表用户之间的相似性。通过搜索目标节点的邻域节点，然后对邻域节点的评级进行集成，为目标用户提供建议。在这些推荐算法中，协同过滤（以下简称cf）是最成功的一种。传统协同过滤算法的基本思想是，根据用户最近邻居的评分数据预测未被用户评分的项的评分，然后向用户推荐预测评分最高的项。为了找出目标用户最近的邻居，需要一种测量用户之间相似性的方法。但随着电子商务系统的扩展和用户、项目的急剧增加，用户的评分数据越来越少，降低了对象用户计算的最近邻居的准确性，使得协同过滤算法产生了不良的推荐率。为了解决这一问题，研究人员提出了一些改进的方法，如基于协同过滤算法及其改进[7-9]、基于矩阵维数约[10-11]的协同过滤和基于云模型[12-13]的协同过滤。这些方法在一定程度上解决了数据稀缺性问题，提高了用户相似性的准确性，但都忽视了用户保持局部相似性的可能性较大这一重要问题。在使用协同过滤推荐系统预测特殊类型项的评分时，计算用户之间的整体相似性是不准确的。例如，用户a和用户b可能对s类项保持相同的口味，用户a和用户b可能都很像s类项，反之亦然。但是对于t型物品，他们的口味有很大的差异，也许用户a非常喜欢t型物品，而它正好与用户b相反来解决这个问题，针对项目类型，提出了一种基于部分用户相似性的协同过滤推荐算法。
这篇论文的其余部分的结构如下。第二节介绍了基于用户相似度的集体过滤推荐系统的主要框架，分析了传统用户相似度测量方法的缺点。第三节介绍了一种新的用户相似度测量方法和一种基于新用户相似度测量的协同滤波推荐算法，以下简称uicf。
剩余内容已隐藏，支付完成后下载完整资料
资料编号：[22030]，资料为PDF文档或Word文档，PDF文档可免费转换为Word
以上是毕业论文外文翻译，课题毕业论文、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。
