Research On Service Transaction Recommendation Systems Based On Knowledge Graph Representation Learning With Graph Structure Information

With the increase of the data on service transaction applications,there is a serious problem that how to efficiently explore valuable information from these complex data for the service transaction.Today,researchers have developed recommendation systems to predict users' preferences for improving users' satisfaction with the service transaction.However,recommender systems still have bottlenecks of processing sparse data,the cold start problem or other issues.Through recommendation technology based on a knowledge graph(KG),the service transaction data can be stored with a more suitable form of the network structure,so that service transaction applications can precisely and efficiently explore these data.For this reason,we try to sufficiently learn high-dimensional features in KGs for further enhancing the practicability and extendibility of recommender systems.In this paper,we constructed a novel recommendation algorithm based on knowledge graph representation learning with graph structure information and combined graph structure features with recommender tasks so that the abundant semantic information in KGs of high connectivity and low structure can be better utilized.Firstly,we designed the knowledge graph suitable for service transaction recommendation tasks according to the service transaction data.Then,we utilized knowledge graph representation learning with graph structure information to learn related features in the KG.We studied knowledge graph representation learning methods based on the path structure and the ripple structure from the idea of deep-search way and wide-search way respectively for different situations.Finally,related features were extracted and efficiently embedded in recommender models.Service transaction recommendation systems based on knowledge graph representation learning with graph structure information can provide a better user experience for different kinds of service transaction applications in the real-world,and it can further improve the economic efficiency of the service transaction.Our major research content and contributions are introduced as follows:(1)A KG structure named collaborative recommendation knowledge graph(CRKG)was designed for service transaction recommendation tasks by analysing service transaction data.Then,we completed the process of CRKG construction and visualization on a real transaction dataset.(2)The KG-PTrans E algorithm based on path structure representation learning was proposed which embedded path structure features in recommendation models.It utilized the deep-first search idea to obtain long term dependencies of entities in KGs.We proposed and realized a KG representation learning method that can automatically extract semantic features from the path structure to obtain entities' and relations' low-dimensional vector representations,and then combined it with a recommender framework.Compared with the state-of-the-art models,experiments on two real-world service transaction datasets verify the improvement of the KG-PTrans E performance on sparse data.(3)The KG-GCN algorithm based on ripple structure representation learning was proposed which utilized the wide-first search idea to obtain k-steps connection structures of entities in KGs.We used graph convolution networks to learn nodes' information propagation of different steps,and recursively aggregated the information together to obtain the vector representations of entities and relations with ripple features and then embed them in the recommender framework.The experiments on two real-world service transaction datasets verify the consistent increase of the KG-GCN performance on dense data.