Abstract: China Agricultural Technology Promotion Information Platform（NJTG） Q ＆ A community can help users interact with agricultural experts, so as to obtain accurate answers to questions to solve agricultural scenarios. In the platform Q ＆ A community, thousands of questions about rice are added every day, and the detection of the same semantic questions is the key technical link of agricultural intelligent Q ＆ A. To solve this problem, Word2 Vec at character level is used to represent the initial question representation, and Siamese neural network is used as the basic model framework to learn the semantic features of sentences and obtain the context information. Then BiLSTM long and short term neural network is used to extract semantic temporal features. Finally, a cosine function containing semantic information is used to calculate the similarity of question at the semantic level, and a comparative experiment is conducted with other semantic matching models. A similarity pair dataset of 7 820 pairs of rice questions is constructed to optimize and train the important parameters of the model. The experimental results show that the BiLSTM-CNN model proposed in this paper can extract features with different granularity effectively and improve the matching effect of rice question similarity. In the constructed data set, BiLSTM-CNN model has higher accuracy and F1 value than other text matching models, reaching 98.2% and 88.75%. At the same time, the accuracy of the proposed model is better than that of other comparison models in six different categories of rice question pairs. In the case of small amount of data, the accuracy is still high, which proves that the proposed model has good robustness.