Domain-aware reputable service recommendation in heterogeneous manufacturing service ecosystem

Abstract

Networked manufacturing becomes an important manufacturing method for modern manufacturing enterprises. With the wide adoption of service-oriented architecture and cloud manufacturing, manufacturing enterprises and organisations publish their manufacturing capability, such as resources, processes and knowledge as manufacturing services. A rapidly growing manufacturing service ecosystem can be observed nowadays, which brings the information overload problem for the service selection. Thus, how to organise these services and how to recommend the reputable services become two important issues for the manufacturing service retrieval and reuse. In this paper, the service cluster method WSTPCluster, which develops the topic model based on Latent Dirichlet Allocation, is employed to cluster services into specific domains. As the service ecosystem can be modelled as a heterogeneous service network, a unified reputation propagation method URPM is proposed to calculate the reputation of services and to distinguish the reputable services in each domain. Combining WSTPCluster and URPM, domain-aware reputable service recommendation method is introduced to recommend the high reputable services in each domain for the consumers. Experiments show that this method brings a better performance both in the recommendation accuracy and in the long tail recommendation.

Research on cloud manufacturing service recommendation based on graph neural network

“…For example, Fan et al [ 13 ] conducted service recommendation research from the perspective of manufacturing service clustering, developed a service clustering method by using a Latent Dirichlet Allocation-based topic model to cluster CMfg services into specific domains, and then introduced a domain-aware reputation service recommendation method to recommend highly reputable services in each domain for users. Zhang et al [ 41 ] developed a CF method based on hybrid social networks for recommending personalized manufacturing services.…”

“…Accordingly, manufacturing service recommendation systems have gained an extensive attention recently [ 11 , 12 ]. These systems aim at automatically recommending a list of manufacturing service resources that may meet users’ demands [ 13 ]. Meanwhile, as an important means of information filtering, the recommendation system is one of the most effective methods to solve the information overloading problem at present [ 14 , 15 ], which helps to alleviate the serious information overloading problem occurring in CMfg service systems and hence improves the efficiency of CMfg platforms.…”

“…For example, there are some studies that use customer transaction data to obtain the interactions between users and manufacturing service resources, and then use collaborative filtering algorithms to achieve manufacturing service recommendation [ 18 ]. In addition, there are also some other studies that use clustering algorithms to cluster and analyse users with different manufacturing service demands from the perspective of user clustering as a way to conduct research on the recommendation of manufacturing service resources [ 13 ]. However, traditional recommendation algorithms cannot effectively exploit the hidden features among resources, and can only learn low-dimensional and linear features from past interactions between users and services.…”

The study on the architecture of crowd system supporting platform

“…In terms of architecture, it mainly explores how service individuals aggregate to form complex networks and the impact of individual behavior on the overall network. Tsinghua University has done a lot of research work in the basic theory of service internet, structural model, and behavioral characteristics (Li et al, 2011;Liu et al, 2013;Fan et al, 2015). In terms of basic theory, it mainly studies the characteristics of the openness, dynamics and adaptability of the service internet and analyzes the behavioral characteristics and dynamic evolution characteristics of the service network.…”

Extracting relevant terms from mashup descriptions for service recommendation

“…Wu et al [20] proposed a dynamic weight formula to calculate service reputation from historical ratings and unfair ratings were removed leveraging the idea of olfactory fatigue phenomenon. Fan et al [21] employed LDA to cluster services into different domains and recommended services with the highest reputation in each domain. QoS-based recommendation centers on non-functional properties of web services and is different from our focus.…”