Video recommendation method based on multi-attention mechanism and heterogeneous information network

Abstract

The online learning platform has rich video resources. To recommend video resources suitable for learners' knowledge background and learning interests, it is necessary to know the details of each entity object related to learners. To this end, a video recommendation method based on multi-attention mechanism and heterogeneous information network is proposed. The heterogeneous information network learning method is used to obtain the representation of node feature vectors, metapath and attribute embedding; Implicit feedback matrix is used to obtain the embedding of nodes and neighbors; Use convolutional neural network to embed neighbor nodes and aggregate meta-path pairs; Multi-attention mechanism is used to modify node embedding and meta-path embedding; The above embedding is spliced and input into the fully connected neural network for learning and recommendation. The experiment was carried out on the MOOCCube and MOOCdata dataset, and the evaluation model was carried out using indicators such as accuracy, recall rate, average reciprocal ranking, and area under the working characteristic curve of subjects. Experimental results show that the proposed method achieves better recommendation performance than common recommendation methods.