The extraction of food public opinion entity relationship is the key technology for constructing the map of food sensation knowledge, and it is also an important research topic in the field of information extraction. Aiming at the entity-to-many relationship problem often appearing in food grievances, the location-aware domain-based semantic attention mechanism is introduced to the convolutional neural network (CNN) and location-aware semantic roles attention mechanism is introduced to the bidirectional long short-term memory network (BLSTM). Then, model was constructed based on CNN-BLSTM for food sentiment entity relationship extraction. In this paper, a comparative experiment was carried out on the food moisture dataset. The experimental results showed that the accuracy of the entity relationship extraction model of food public opinion based on CNN-BLSTM was higher than that of the commonly used deep neural network models. From 8.7% to 13.94%, the rationality and effectiveness of the proposed model were verified.