Figure - available from: World Wide Web
This content is subject to copyright. Terms and conditions apply.
The structure of embedding layers, which are divided into feature and field embedding layer. Field contains User, Program and Interaction. The feature embedding layer input is one-hot or multi-hot sparse vectors, and the output is low-dimension dense real-value feature embeddings, which are called feature embedding vectors. The feature embedding vectors of the same field are concatenated as the input of field embedding layer
Source publication
Users are frequently overwhelmed by their uninterested programs due to the development of smart TV and the excessive number of programs. For addressing this issue, various recommendation methods have been introduced to TV fields. In TV content recommendation, auxiliary information, such as users’ personality traits and program features, greatly inf...
Similar publications
Hotel Recommendation System Based On Hybrid Recommendation Approach for Ethiopian Cities
Citations
... They used the content of adversarial training to aware of long-term and short-term information for film recommendations, while also introducing poster information of films to supplement their feature encoder. Focusing on the role of user personality traits, program features and other auxiliary information, Zhou et al. [57] constructed a deep factorization integrated attention mechanism (DFIAM) model adopting hierarchical attention networks by joining two components, FNN and DMF. Gan et al. [9] proposed a user movie interest space (UMIS) model based on the sequential ratings of users to enhance online user behavior in the timeliness of interest. ...
With the development of online programming, the broadcast TV industry has experienced some negative impacts. The program recommendation has become crucial to boost the industry’s construction. While there have been many researches on program recommendation, there are still challenges to overcome such as alleviating the sparsity of user interaction data and improving the dynamics and generalizability of the recommendation model. In this paper, we propose a novel approach for TV program recommendation called heterogeneous information-based recommendation with graph enhanced representation (HGER). The HGER model mainly consists of two main modules. One is the program encoder which uses the program’s heterogeneous information and attention mechanism to extract personalized content and considers high-order neighbor program representations through a graph structure. The other is the user encoder which utilizes the user’s historical viewing behaviors and combines it with the graph structure to represent the high-order neighbor user. Thus, we implement an enhanced representation for both the program and user. Through extensive experiments on a real-world dataset, the results of AUC, NDCG and HR demonstrate that our approach can effectively enhance the dynamics and generalizability of the model for TV program recommendations.
