Object-dependent Document-level Sentiment Analysis Based on Sentence Features

In the field of natural language processing, document-level sentiment analysis is one of the tasks that are widely studied. Traditional deep-learning method is facing the challenge of long-distance dependence of document-level text. At the same time, traditional document-level sentiment analysis assumes that the article contains only the view of one object, which has some limitations. By selecting four features including object feature, sentiment feature, weight feature and discourse relation feature, we propose a document-level sentiment analysis model based on sentence features(DABS) and an object-dependent document-level sentiment analysis model based on sentence features(ODABS) to improve document-level sentiment classification. The experimental results show that the features selected in this paper are representative, and the model has better effect than the baseline model.