Jianxing Zheng

Neighborhood-User Profiling Based on Perception Relationship in the Micro-blog Scenario

In the micro-blog scenario, personal user profiling relying on content is limited for recommending desired diverse subjects due to its shortcomings of short text, often leading to a poor recall. Currently, many methods only utilized the personal knowledge from each individual user to represent user profile without considering the neighborhood information. However, resource information related to neighboring friends play an important role in improving the performance of recommender systems. In this paper, we present the personalized expanded user profiling for micro-blog subject recommendation via ontology semantics structure. Next, taking into account diffusion ability of followee friends, we discuss resource perception relationship (RPR) and follow perception relationship (FPR). Finally, we discuss how, by adjusting the importance of RPR and FPR, the neighborhood is selected to construct neighborhood-user profile, which can mine new relevant subjects for target user. Our experimental results demonstrate the effectiveness of our neighborhood-user profiling in comparison to the existing collaborative filtering and personal user profile recommendation approaches on Sina micro-blog platform datasets.

Method of Recommend Microblogging Based on User Model

The situations of context absence and sparse feature in short texts present a challenge to effective personalized service, especially in the recommend short texts, widened semantic gap between low-level text features representation and high-level interpretation. Meanwhile, the propagation characteristics of short texts also have an effect on the results of recommend short texts. However, the traditional methods of recommend short texts rarely take the above two aspects into account when recommending short texts to users. To solve the above problems, this paper presents a microblogging recommend method based on user model. This method maps the feature of microblogging to the semantic concept by Semantic Extension Method, then calculates the similarity of user model and semantic microblogging, furthermore calculates the factor of microbloggings forwarding and comments, and lastly comprehensively considers the similarity and factor to recommend microblogging to users. Experiments show that the method of recommend microblogging based on user model is better than traditional methods. Users are more satisfied with the recommend results by user model than by traditional methods, and have a very high appraisal of this recommend method.

Discovering Similar User Models Based on Interest Tree

With the explosive development of Internet and Social Networking Services (SNS), more and more people begin to get information from others. So how to find users, which have similar interests, is becoming an important issue. The traditional method is using a vector to calculate the similarities between user models. The similarities between user models are measured by one value. This method is very simple, but some useful details are lost. Users do not know where they are similar in detail. Particularly, the existing approaches cannot calculate the similarity between users under the different interest trees of user models. Aiming at solving these problems, a method which expresses and calculates the similarity between two user models in different granularities is proposed in this paper. Node Structure Similarity (NSS), Interest Theme Similarity (ITS), Comprehensive Interest Similarity (CIS) and Dynamical Comprehensive Interest Similarity (DCIS) are considered to describe the similarities between user models. NSS reflects to structural similarity of interest tree. ITS is the interest theme similarity between users interest trees. CIS is a comprehensive similarity which has combined NSS with ITS. DCIS is not only calculated by NSS and ITS but also considered the weight of NSS and ITS. Experimental results show that DCIS is the most reasonable one among the three methods mentioned above.

Research on life-cycle of user model in U-Business

Abstract“U-Business” is a novel type business environment, which can provide various services via many mobile devices. In order to provide personalized service to different users, user model (UM) can play an important role in U-Business. UM reflects some characteristics of users to a certain degree, which is used widely in U-Business, like personalized recommendation, social computing, information retrieval services, and so on. Currently, there are more and more researchers who focus on the building and update of UM based on the activities of people. However, as too many UM appeared, the number of UM in cyber space is increasingly large, which takes a lot of space and cost. Furthermore, after some users disappear in the physic world, their models are still working in the cyber world. This case is not reasonable obviously, but few researches take care about it. Therefore, one of important issues, the death of UM should be taken into account in the whole life-cycle of user model. This paper proposes a specific user modeling method for the Cyber Individuals (Cyber-I) in U-Business. The essential difference between this UM and traditional ones is that it has a life, that is, birth, growth, and demise, like a life-cycle of Cyber-I. Specially, the significance of UM life ending and five states of UM death are described from an organic viewpoint. In addition, there is a framework of the whole life process of UM. Finally, the proposed idea is applied to the field of personalized service.

Multi-granularity Recommendation Based on Ontology User Model

The traditional personalized recommendation system supplies the target user with top k items in fixed interest subject. However, the recommended items cover the coarse subject level and the accuracy performance is poor. Taking into account ontology structure of subject, users actual interests can distribute in multiple sub-subject structures. In this paper, multi-granularity recommendation mechanism relying on multi-granularity similarity is proposed to fit users actual detail demands. Specially, a personalized ontology user model is learned to represent users multi-granularity interests. According to ontology structure, the multi-granularity similarity method is implemented by combing content closeness and semantic closeness between user models at different grained subjects. Lastly, recommendation method distributed in multi-granularity subjects is achieved to compare against traditional single subjects recommendation for their performances. The experimental results show that the proposed mechanism is more successful.

MaaS: Model as a Service in Cloud Computing and Cyber-I Space

In recent years, there has been an increasing interest in user modeling, because of its extremely wide usage in real world applications, such as recommendation systems and personalized services. However, a large number of researchers have been devoted to focusing on developing a user model that can only facilitate their own applications, which has limited the rapid advances in user modeling research, so that ubiquitous modeling theory and its applications have become an emerging and open issue to solve. To handle this challenging task, this paper investigates the existing user modeling techniques and potentially applicable areas, including cloud computing and Cyber-Individual (Cyber-I), which are on the way of gradually evolving as important computing paradigms. This paper first proposes a novel user modeling conception and profile from Service-Oriented Architecture (SOA) point of view, called Model as a Service (MaaS). Then, based on the widely used three-layer abstraction of cloud computing, a new MaaS-based four-layer new architecture of cloud computing is proposed, allowing users and model developers to participate in cloud activities for the purpose of supporting personalized services. Finally, a novel MaaS-based Cyber-I framework is proposed on the basis of the architecture of Cyber-I oriented platform, where user modeling and MaaS-related operations center on the functionalities, providing the services of creation, evolution and retrieval in Cyber-I space.

Service composition and user modeling for personalized recommendation in cloud computing

In recent years, cloud computing is gradually evolving as a popular computing paradigm, which offers a uniform platform for service providers to publish their applications as cloud services. In many cases, however, single cloud service cannot satisfy a service request due to its simple functionality. Furthermore, current service composition systems have seldom taken into account user interests for personalized recommendation. In this paper, we propose a novel framework for personalized service recommendation in cloud computing platform by Web service composition and user modeling. The proposed framework first models cloud services together with a service request as a Web service composition problem, called cloud service recommendation (CSR) planning problem. It is fed into our self-developed service planner to compose a cloud service with complex business workflow. Second, our framework also applies user modeling for checking whether the generated composite cloud service can be matched with the interests of service consumer. To validate the feasibility of CSR framework, we have designed and implemented two prototype systems, QoS-aware service composition system and service platform based on user model.

Diversification recommendation of popular articles in micro-blog scenario

With the information overload in web services, micro-blog has been increasingly providing as a media for end-users to express their opinions. The notable feature of micro-blog articles is prone to be a burst of popularity during a short period. In addition, diverse interests make users bored in redundant items in most recommender systems. Therefore, providing users with diverse popular micro-blogs that suit their interesting topics is an important issue. In this paper, depending on forwarding number and comment number of micro-blogs, an effective model for popularity prediction is proposed to discover popular topics. Then, a MaxMin diversity algorithm based on content distance and popularity density is proposed to discover top k micro-blogs. Finally, we design a diverse personalized popularity attention (DPPA) recommendation approach for target user. We conduct extensive experiments on large scale micro-blog datasets. The experimental results show that our proposed approach can satisfy users requirements with a higher recall than personal attention methods.

Adaptive Updating Algorithm of User Model Based on Interest Cycle

User model is the base of supporting personalized recommendation. It is not only able to represent users interests, but also could reflect the change of users interests over time. However, existing user model updating approaches have several disadvantages, such as unified interest update mode, the same interest category without considering interest cycle, which reduce the accuracy of user model representing and updating. To solve these problems, this paper introduces a method to discover interest cycle by cutting valid time of interest existing in the curve of interest changing, and divides the interests into three kinds of categories which are long-term interest, medium-term interest and short-term interest. Based on the different kinds of interest, the disparate interest update strategies are presented to make change of interest adaptive, especially adaptive forgetting functions with the factor of interest forgotten and adaptive incenting functions with the factor of interest incented are constructed, both of them could reflect the actual intention of interest changing over time. It is shown by the experiment that the application of the method presented in this paper is more effective and practicable than the method with adopting the unique forgetting and incenting function for all interests.