Design and Research of Personalized News Recommendation System Based on LSTM Algorithm

In order to solve the problem of low extraction accuracy caused by the lack of consideration of the impact of the following text of the central word on the topic word in LSTM network, the author proposes a personalized news recommendation system based on LSTM algorithm. Firstly, the LSTM algorithm is used to model the preceding and following information of the central word in two directions; Then, attention mechanism is introduced into the bidirectional LSTM algorithm to assign higher weights to features with higher influence; Finally, use the softmax layer to classify the words in the document into topic words or non topic words. And a two-stage model training method was also proposed, which is to pre train the model on the automatically annotated training set and then use the manually annotated dataset to train the model. The experiment conducted topic word extraction tasks on three types of news texts: sports, entertainment, and technology. The experimental results indicate that: The Att-iBi LSTM algorithm proposed by the author has improved F1 values by 13.78%, 24.31%, and 3.32% compared to SVM, TextRank, and LSTM, respectively. The Att-iBi LSTM algorithm trained in two stages has increased F1 values by 1.55% compared to the first stage training method. Conclusion: The system has achieved good extraction results in news topic word extraction tasks in different fields, indicating the generalization of this method.