Dynamic Heterogeneous Information Network Representation Learning

The rapid development of network representation learning(a.k.a.,network embedding)provides novel ideas for network analysis and effectively improves the performance of data mining tasks.Traditional methods mostly focus on static,homogeneous networks,that is,assuming that the network is static and has a single node and edge type.However,in the real world,networks are usually dynamic,with various node and edge types,so-called dynamic,heterogeneous information networks,which makes traditional methods unable to effectively model the properties of the network.The learned node representations are difficult to apply to real application scenarios.Therefore,this research work is based on studying dynamic and heterogeneous information network embedding methods.Specifically,this paper first studies the dynamics of the network and proposes a dynamic network embedding method M2DNE that combines micro and macro dynamics.Secondly,this paper studies the heterogeneity of the network,and proposes a relational structure-aware heterogeneous information network embedding method RHINE.Then,this paper studies the dynamics and heterogeneity of the network,and proposes a dynamic heterogeneous information network embedding method DyHNE.Finally,in the article recommendation scenario of WeChat's "Top Stories",this paper studies the friend-enhanced recommendation and proposes a social influence attentive neural network SIAN to verify the effectiveness of heterogeneous information network embedding in real scenarios.In the real world,dynamic networks which usually evolve over time in terms of microscopic and macroscopic dynamics,are ubiquitous.The micro dynamics describe the formation process of network structures in a detailed manner,while the macro-dynamics refer to the evolution pattern of the network scale.Both micro-and macro-dynamics are the key factors to network evolution;however,how to elegantly capture both of them for temporal network embedding,especially macro-dynamics,has not yet been well studied.In this paper,we propose a novel dynamic Temporal Network Embedding with Micro-and Macro-Dynamics,named M2DNE.Specifically,for micro-dynamics,we propose a temporal attention point process to capture the formation process of network structures in a fine-grained manner.For macro-dynamics,we define a general dynamics equation parameterized with network embeddings to capture the inherent evolution pattern and impose constraints in a higher structural level on network embeddings.Mutual evolutions of micro-and macro-dynamics in a temporal network alternately affect the process of learning node embeddings.Extensive experiments on three real-world temporal networks demonstrate that M2DNE significantly outperforms the state-of-the-arts.On the other hand,information networks in the real world are often heterogeneous,that is,the network contains multiple types of nodes and relations,called Heterogeneous Information Networks(HIN).Most existing methods usually employ one single model for all relations without distinction,which inevitably restricts the capability of HIN embedding.In this paper,we argue that heterogeneous relations have different structural characteristics,and propose a novel Relation Structure-aware HIN Embedding model,called RHINE.By exploring four real-world networks with thorough analysis,we present two structure-related measures which consistently distinguish heterogeneous relations into two categories:Affiliation Relations(ARs)and Interaction Relations(IRs).To respect the distinctive structural characteristics of relations,in RHINE,we propose different models specifically tailored to handle ARs and IRs,which can better capture the structures in HINs.Finally,we combine and optimize these models in a unified manner.Furthermore,the dynamic nature and heterogeneity of networks usually coexist,so-called dynamic heterogeneous information networks.In practice,a real HIN usually evolves over time with the addition(deletion)of multiple types of nodes and edges,and even a tiny change can influence the entire HIN structure and semantic information.In order to capture the dynamic evolution of the HIN,the conventional HIN embedding methods need to be retrained to get the updated embeddings,which is time-consuming and unrealistic.In this paper,we investigate the problem of dynamic HIN embedding and propose a novel Dynamic HIN Embedding model(DyHNE)with meta-path based proximity.As the HIN evolves over time,we naturally capture changes with the perturbation of meta-path augmented adjacency matrices.Thereafter,we learn node embeddings by solving generalized eigenvalue problem effectively and employ eigenvalue perturbation to derive the updated embeddings efficiently without retraining.Experiments on three real-world datasets show that DyHNE outperforms the state-of-the-arts in terms of effectiveness and efficiency.The powerful modeling capabilities of HIN representation learning have also promoted the development of real application scenarios in the industry.Therefore,this work further studies the friend-enhanced recommendation in the article recommendation scenario of WeChat's "Top Stories",and proposes a Social Influence Attentive Neural network,called SIAN.In order to fuse rich heterogeneous information,SIAN models recommendation scenarios from the perspective of heterogeneous social networks.In order to fuse rich heterogeneous information,the attentive feature aggregator in SIAN is designed to learn user and item embeddings at both node-and type-levels.More importantly,a social influence coupler is put forward to capture the influence of the friend referral circle in an attentive manner.Experimental results demonstrate that SIAN outperforms several state-of-the-art baselines on two real-world datasets.