Research on the Construction of College Student Employment Prediction Model Based on Deep Learning

The development of the Internet promotes the diversification of the forms of dissemination of Marxist theory, and the comments of social media users contain rich emotional tendencies, which is an important basis for evaluating the effectiveness of dissemination. In this paper, we propose a hybrid sentiment analysis model based on BERT pre-trained language model and Bi-directional Long-Short-Term Memory Network (Bi-LSTM). The model uses the deep bidirectional semantic understanding ability of BERT to capture the contextualized meaning of theoretical terms, and combines the long-distance temporal dependence modeling advantage of Bi-LSTM to resolve complex emotional patterns (such as negative + commendatory word combinations). By collecting and strictly preprocessing user comment text data from authoritative platforms such as Learning Power and People's Network Theory Channel, an end-to-end architecture including input layer, two-way semantic coding layer (BERT), context enhancement layer (Bi-LSTM) and sentiment classification layer is constructed. The empirical analysis shows that the hybrid model is significantly better than single BERT, Bi-LSTM and random forest model in the accuracy (87. 3%), macro F1 value (86. 8%) and other key indicators, especially in the recognition of composite political expressions and the resistance to semantic noise. It provides an effective technical tool for accurately and quantitatively evaluating the emotional feedback of the dissemination of Marxist theory.