A Model to Personalized Text Summarization Generation Based on Eye Tracking

In order to meet the personalized demands of various users for summarization generation, a gaze-based key information guide network (GKGN) is proposed. Firstly, it uses eye-movement data, such as fixation and saccade in the reading process, to extract key eye-movement information. The key eye-movement information is encoded by long-short term memory (LSTM) networks. Secondly, the hidden layer state of the key eye-movement information is fused with mechanisms of attention. Apart from that, intra-attention model and pointer generator network (PGN) are fused to generate the personalized text summarization. The ADEGBTS dataset, which contains users’ eye-movement data on reading Chinese news, is used for evaluating the GKGN model. The results show that, compared with existing text summarization models, the GKGN model scored 48.68% higher on the ROUGE. In addition, a text summarization system is designed and developed, and the user test results show that the system can efficiently generate the personalized summarization.