As the college students have been a most active user group in various social media, it remains significant to make effective sentiment analysis for college public opinions. Capturing the direction of public opinion in the student community in a timely manner and guiding students to develop the right values can help in the ideological management of universities. Universally, the recurrent neural networks have been the mainstream technology in terms of sentiment analysis. Nevertheless, the existing research works more emphasized semantic characteristics in vertical direction, yet failing to capture sematic characteristics in horizonal direction. In other words, it is supposed to increase more balance into sentiment analysis models. To remedy such gap, this paper presents a novel sentiment analysis method based on multi-scale deep learning for college public opinions. To fit for bidirectional semantic characteristics, a typical sequential neural network with two propagation paths is selected as the backbone. It is then extended with more layers in horizonal direction. Such design is able to balance both model depth and model breadth. At last, some experiments on a real-world social media dataset are conducted for evaluation, well acknowledging efficiency of the proposed analysis model.
Keywords: multi-scale deep learning; public opinions; semantic characteristics; sentiment analysis.