Learning Hierarchical User Interest Models from Web Pages

We propose an algorithm for learning hierarchical user interest models according to the Web pages users have browsed. In this algorithm, the interests of a user are represented into a tree which is called a user interest tree, the content and the structure of which can change simultaneously to adapt to the changes in a user＇s interests. This expression represents a user＇s specific and general interests as a continuurn. In some sense, specific interests correspond to shortterm interests, while general interests correspond to longterm interests. So this representation more really reflects the users＇ interests. The algorithm can automatically model a us er＇s multiple interest domains, dynamically generate the in terest models and prune a user interest tree when the number of the nodes in it exceeds given value. Finally, we show the experiment results in a Chinese Web Site.     

Personalized Recommendation Algorithm Based on Preference Features

A hybrid collaborative filtering algorithm based on the user preferences and item features is proposed.A thorough investigation of Collaborative Filtering （CF） techniques preceded the development of this algorithm.The proposed algorithm improved the user-item similarity approach by extracting the item feature and applying various item features＇ weight to the item to confirm different item features.User preferences for different item features were obtained by employing user evaluations of the items.It is expected that providing better recommendations according to preferences and features would improve the accuracy and efficiency of recommendations and also make it easier to deal with the data sparsity.In addition,it is expected that the potential semantics of the user evaluation model would be revealed.This would explain the recommendation results and increase accuracy.A portion of the MovieLens database was used to conduct a comparative experiment among the proposed algorithms,i.e.,the collaborative filtering algorithm based on the item and the collaborative filtering algorithm based on the item feature.The Mean Absolute Error （MAE） was utilized to conduct performance testing.The experimental results show that employing the proposed personalized recommendation algorithm based on the preference-feature would significantly improve the accuracy of evaluation predictions compared to two previous approaches.     

Empirical Research on User Acceptance of Mobile Searches

The mobile search, a combination of a web search engine and a mobile communication system, is viewed as the most influential application in the 3G era. Therefore, mobile search service providers are eager to know which factors most influence user acceptance of mobile searches. Based on the characteristics of mobile searches and a review of previous information technology acceptance research, this study integrates the task technology fit model and the unified theory of acceptance and use of technology model to develop a mobile search acceptance model and empirically tests this model. This study finds that, for mobile searches, the performance expectancy, social influence, and perceived cost all significantly influence use intention and the performance expectancy increases with the increasing user’s experience and higher task technology fit degree. The effort expectancy is found to not affect the use intention of mobile searches and the users’ gender does not have a significant moderating effect on the use intention. The results are then used to develop suggestions for mobile search providers to promote their application and development.     

Fuzzy clustering method for Web user based on pages classification

A new method for Web users fuzzy clustering based on analysis of user interest characteristic is proposed in this article. The method first defines page fuzzy categories according to the links on the index page of the site. then computes fuzzy degree of cross page through aggregating on data of Web tog. After that, by using fuzzy comprehensive evaluation method, the method constructs user interest vectors according to page viewing times and frequency of hits. and derives the fuzzy similarity matrix from the interest vectors for the Web users. Finally. it gets the clustering result through the fuzzy clustering method. The experimental results show the effectiveness of the method.     

A solution to the anonymity of the mobile subscriber

The user anonymity mechanism provided in GSM and UMTS network is analyzed,and a new approach to solve the anonymity of mobile subscriber is proposed in this paper.Using the ciphering algorithm with a dynamic key and a static key,the method allows mobile subscribers to connect network with user anonymity,and it can be seamlessly integrated with the existing authentication algorithms.The analyses reveal that the proposed scheme can provide the integrity protection of messages,prevent the replay attacks,and be implemented easily.     

Community-Based User Domain Model Collaborative Recommendation Algorithm

Collaborative Filtering (CF) is a commonly used technique in recommendation systems. It can promote items of interest to a target user from a large selection of available items. It is divided into two broad classes: memory-based algorithms and model-based algorithms. The latter requires some time to build a model but recommends online items quickly, while the former is time-consuming but does not require pre-building time. Considering the shortcomings of the two types of algorithms, we propose a novel Community-based User domain Collaborative Recommendation Algorithm (CUCRA). The idea comes from the fact that recommendations are usually made by users with similar preferences. The first step is to build a user-user social network based on users’ preference data. The second step is to find communities with similar user preferences using a community detective algorithm. Finally, items are recommended to users by applying collaborative filtering on communities. Because we recommend items to users in communities instead of to an entire social network, the method has perfect online performance. Applying this method to a collaborative tagging system, experimental results show that the recommendation accuracy of CUCRA is relatively good, and the online time-complexity reduces to O.n/.     

A Recommendation Mechanism for Web Publishing Based on Sentiment Analysis of Microblog

Microblog is a social platform with huge user community and mass data. We propose a semantic recommendation mechanism based on sentiment analysis for microblog. Firstly, the keywords and sensibility words in this mechanism are extracted by natural language processing including segmentation, lexical analysis and strategy selection. Then, we query the background knowledge base based on linked open data(LOD) with the basic information of users. The experiment result shows that the accuracy of recommendation is within the range of 70%-89% with sentiment analysis and semantic query. Compared with traditional recommendation method, this method can satisfy users’ requirement greatly.     

Research on Bayesian Network Based User＇s Interest Model

It has very realistic significance for improving the quality of users＇ accessing information to filter and selectively retrieve the large number of information on the Internet. On the basis of analyzing the existing users＇ interest models and some basic questions of users＇ interest （representation, derivation and identification of users＇ interest）, a Bayesian network based users＇ interest model is given. In this model, the users＇ interest reduction algorithm based on Markov Blanket model is used to reduce the interest noise, and then users＇ interested and not interested documents are used to train the Bayesian network. Compared to the simple model, this model has the following advantages like small space requirements, simple reasoning method and high recognition rate. The experiment result shows this model can more appropriately reflect the user＇s interest, and has higher performance and good usability.     

An approach for user’s trajectory prediction based on the trajectory cluster

The existing user’s trajectory prediction methods considered little about the interrelation among users and would fail if the user historical trajectory data were lack. This paper presents a user’s trajectory prediction model and corresponding algorithms by the historical trajectories of users based on the trajectory cluster. The experimental results on MDC dataset show that the proposed method has great improvement in efficiency, accuracy, and scalability comparing with the traditional methods, and it also be applied to the situation where user’s historical trajectory data are lacked.     

News Recommendation System Based on Topic Embedding and Knowledge Embedding

News recommendation system is designed to deal with massive news and provide personalized recommendations for users. Accurately capturing user preferences and modeling news and users is the key to news recommendation. In this paper, we propose a new framework, news recommendation system based on topic embedding and knowledge embedding(NRTK). NRTK handle news titles that users have clicked on from two perspectives to obtain news and user representation embedding :1) extracting explicit and latent...     

基于双向情感分析的实时性音乐推荐系统设计

目前的音乐推荐系统,一般采用基于个体兴趣的推荐方法,这种方式虽然能满足大部分情景下的用户需求,但无法感知到用户实时性的心情变化。考虑到不同情绪状态下用户对于音乐的需求往往也会发生改变,提出一种基于双向情感分析的算法并构建了实际系统,实时分析用户的情感需求来进行音乐推荐。一方面基于音乐在频域的梅尔倒谱系数构建特征分类器完成歌曲的情感分类;另一方面通过获取用户在社交网络中的实时文本信息,基于自然语言理解分析出用户当前的情感需求,最终为用户产生音乐推荐列表,实现基于情境感知的实时音乐推荐。实验表明,使用该个性化推荐算法具有更高的准确性,用户群体可以获得更为满意的用户体验。     

基于图神经网络特征交叉的协同过滤算法

学习用户和项目有效的向量表示是推荐系统的核心目标,现有的推荐模型大多通过深度神经网络或专门设计的特征交叉,来学习用户-项目间的特征交叉生成用户(项目)向量表示,但并未将用户(项目)特征间的交叉信息编码到嵌入向量中充分利用特征交叉信息,且多个特征交叉信息对于生成最终的用户(项目)向量表示的影响不同.基于此,构建两个图神经网络模块,学习用户(项目)特征间的交叉信息、用户-项目之间的特征交叉信息,并通过计算注意力分数对特征交叉信息进行加权,得到用户(项目)的特征信息;然后通过门控循环神经网络(GRU)聚合原始的特征信息和网络层学习到的特征交叉信息,得到最终的用户(项目)向量表达;最后通过用户向量与项目向量的元素积得到最终的推荐结果.在数据集MovieLens 1M、Book-Crossing和Taobao上验证了模型的有效性.     

Trust-based context-aware mobile social network service recommendation

The service recommendation mechanism as a key enabling technology that provides users with more proactive and personalized service is one of the important research topics in mobile social network (MSN). Meanwhile, MSN is susceptible to various types of anonymous information or hacker actions. Trust can reduce the risk of interaction with unknown entities and prevent malicious attacks. In our paper, we present a trust-based service recommendation algorithm in MSN that considers users’ similarity and friends’ familiarity when computing trustworthy neighbors of target users. Firstly, we use the context information and the number of co-rated items to define users’ similarity. Then, motivated by the theory of six degrees of space, the friend familiarity is derived by graph-based method. Thus the proposed methods are further enhanced by considering users’ context in the recommendation phase. Finally, a set of simulations are conducted to evaluate the accuracy of the algorithm. The results show that the friend familiarity and user similarity can effectively improve the recommendation performance, and the friend familiarity contributes more than the user similarity.     

基于深度神经网络和概率矩阵分解的混合推荐算法

针对个性化推荐中用户和项目描述信息未充分利用,用户评分矩阵数据集极端稀疏的情况,提出了基于深度神经网络和概率矩阵分解(PMF)的混合推荐算法.首先,对用户和项目描述信息进行预处理,形成包含用户偏好特征的用户和项目特征集,再将各特征输入深度神经网络模型中进行训练.同时,利用概率矩阵分解模型,根据用户评分矩阵通过最大后验估计优化得到潜在特征向量;然后,通过对概率矩阵分解模型的用户和项目潜在特征向量以及深度神经网络模型的真实特征向量进行迭代更新,收敛得到融合用户和项目真实信息的潜在特征向量;最后,利用该特征向量对用户进行个性化推荐.实验证明,本文算法较经典推荐算法以及前人算法在均方误差与平均绝对误差指标上均有改善,说明本文算法的有效性.     

一种基于商品基因的个性化推荐模型

个性化推荐技术在电子商务系统中得到了广泛应用．但现有的个性化推荐模型不能反映出商品的某些特殊属性对用户购买行为的影响及用户购买某商品的真正原因．引入生物界中基因的概念，提出了基于商品基因的个性化推荐模型，发现用户钟爱商品基因，并将钟爱商品基因遗传到用户选择的商品．该模型能更好地发现用户的购买动机，从而进一步提高个性化推荐精度和用户满意度．     

基于万有引力的个性化推荐算法

本文把物理学中的万有引力定律引入推荐系统,提出一种个性化推荐算法,即基于万有引力的个性化推荐算法。算法把用户使用的标签看作用户喜欢物体的组成颗粒,标注项目的标签被看作项目物体的组成颗粒,社会标签的类型就是颗粒的类型,由此构建了用户喜好物体模型和项目物体模型。喜好物体和项目物体间存在着万有引力,并且引力大小遵循万有引力定律。计算喜好物体和项目物体间的万有引力,并把该引力大小作为二者的相似度度量,引力越大,二者的相似度就越高,对应的项目物体就越有可能被用户喜欢。实验结果证明本文提出的算法可以获得好的推荐性能。     

云计算环境下旅游信息适应性自组织推送算法研究

针对当前推送旅游信息质量不高,而无法满足使用者需求的问题,提出一种旅游信息适应性自组织推送方法。通过求出目标使用者和其他使用者之间的相似度,找到目标使用者的最近邻集合。利用模糊遗传算法确定加权函数,通过加权函数实现使用者对旅游项目评分的更新,以适应使用者变化的旅游兴趣。求出目标使用者对某旅游项目兴趣度预测值,将预测值最高的旅游项目推送给使用者。实验结果表明,所提算法推送旅游信息质量高,可满足使用者需求。     

一种新的基于LDA-MURE模型的音乐个性化推荐算法

针对基于音乐作品信息的音乐作品个性化推荐及协同过滤方法的不足,通过分析音乐作品需求者的音乐试听数据及下载数据,并结合LDA(latent Dirichlet allocation)主题挖掘模型,提出一种基于LDA-MURE模型的推荐算法.实验结果表明,与基于音乐作品需求者的协同过滤算法和基于音乐属性项目的协同过滤算法相比,LDA-MURE算法可更高效地向音乐作品需求者推荐感兴趣的音乐作品.     

结合项目类别信息的协同过滤推荐算法

针对个性化推荐系统中协同过滤算法面临的数据稀疏问题以及用户相似性度量的不准确,提出了一种结合类别信息的协同过滤推荐算法。该算法利用用户评分数据计算用户之间对类别关注的相似性,并将用户对类别关注的相似性和用户评分相似性进行组合,得到用户综合相似性,从而提高了最近邻居搜索的准确度,缓解了数据稀疏性问题。实验结果表明,该方法能够有效地避免传统相似性度量方法存在的问题,使得数据稀疏性对最终推荐结果的负面影响变小,在一定程度上提高系统的推荐精度。     

结合项目类别信息的协同过滤推荐算法

针对个性化推荐系统中协同过滤算法面临的数据稀疏问题以及用户相似性度量的不准确，提出了一种结合类别信息的协同过滤推荐算法。该算法利用用户评分数据计算用户之间对类别关注的相似性，并将用户对类别关注的相似性和用户评分相似性进行组合，得到用户综合相似性，从而提高了最近邻居搜索的准确度，缓解了数据稀疏性问题。实验结果表明，该方法能够有效地避免传统相似性度量方法存在的问题，使得数据稀疏性对最终推荐结果的负面影响变小，在一定程度上提高系统的推荐精度。