A timely personalized comments generation assistant based on LSTM-SNP

Posting personalized comments on an upcoming hot topic in time is very meaningful, which not only likely to attract more users to participate, but also affects other users’ point of view. LSTM-SNP is a variant of long short-term memory (LSTM) inspired by the nonlinear spiking mechanism in nonlinear spiking neural systems. In order to improve the efficiency and diversity of user-edited comments, we design a novel assistant based on the LSTM-SNP model. The assistant consists of two modules, one for predicting topic hotness and the other for generating comments with personalized expression features based on blog post and user information. Experimental results show that, this novel assistant not only predicts the upcoming hot topics accurately, but also outperforms the baseline model in terms of automatic evaluation and human discernment of comment generation. More importantly, the generated comments excel in terms of timeliness and personalization.