Efficient Domain-Specific News Push Service Using Deep Learning Based Text Regression without Users’ Information

While there is an ever increasing collection of domain-specific news, it is difficult for people to identify the important ones from massive information. The challenge lies in that users’ information, e.g., comments and CTR (Click-Through-Rate), may not be available in those professional news articles. In this paper, we develop a deep learning model for important news push service, referred to as HMA, which consists of a Hierarchical attention network and a Multi-head Attention mechanism for non-linear text regression. We conduct the experiments with a dataset of marine industry news to evaluate the performance of the proposed deep learning model. The experiment results show that HMA outperforms other alternative deep learning models due to the hierarchical structure and multi-head attention mechanism. Moreover, the execution time of inference by HMA is less than the computation of TF-IDF when adding the news articles into the news repository. Therefore, the proposed method has the potential for efficiently pushing the important domain-specific news articles without users’ information.