Research On The Classification Method Of LinkedIn Account Based On Network Embedding

As the largest professional social media platform in the world,LinkedIn allows users to update their personal career files,find career development opportunities and interact with other users.The classification of LinkedIn user accounts helps the platform to place advertisements and recommend services for similar users more accurately.It can also help LinkedIn to analyze the composition of users and grasp the relationship between the supply and demand of talents in the labor market.For users,the recommendation service on the LinkedIn platform can expand their social networks and obtain more career development opportunities.Existing research shows that user attribute information and user characteristics in the relationship network can effectively improve the accuracy of user classification.However,there are few interactions between LinkedIn users,and it is not possible to effectively build a user relationship network.LinkedIn user attributes are rich in information and more authentic and reliable.Therefore,this thesis constructs feature word co-occurrence networks based on LinkedIn account attribute data and proposes a account classification method based on network embedding.The main work and contributions of this thesis are as follows:Firstly,aiming at the problem that there are few interactions among LinkedIn accounts and it is difficult to classify accounts based on the connections between accounts,this thesis constructs co-occurrence networks of attribute feature words to mine LinkedIn accounts' similarities.Taking user accounts as nodes and co-occurrence relationship of attribute feature words among user accounts as edges,co-occurrence networks of attribute feature words are constructed in this thesis.The similarity of LinkedIn accounts is mined through the analysis of the complex network,which lays the foundation for subsequent classification of accounts.Secondly,aiming at the problem of incomplete representation of user accounts in traditional account classification,this thesis proposes a new feature selection method to represent user accounts.At first,a network embedded representation method based on spectral graph wavelet is introduced to mine the structural features of account nodes in the feature words co-occurrence network,and then aggregate with the account's inherent attribute features,that is,the account's word features,text features and network structure features as a new feature to characterize the account.The experimental results verify the effectiveness of the method.Thirdly,in order to comprehensively consider the impact of attribute information on user account nodes in the network embedded representation,this thesis proposes a network embedded representation method that fuses user attribute information and applies it to LinkedIn account classification research.At first,different edge weight calculation methods are proposed for different feature word co-occurrence networks,then merged the user account attributes with the feature words co-occurrence networks,and adopted a random walk algorithm that is suitable for the feature words co-occurrence network to capture the network structure of nodes.Finally,the Skip-gram model is used to obtain vectorized representations of nodes fused with the accounts' attribute information to mine the user accounts' network structure features.The experimental results on the LinkedIn account dataset show that the method can effectively mine user account characteristics and improve the accuracy of LinkedIn accounts classification.