In order to analyze the reading behavior and its meaning of readers in blockchain online reading platforms, this article conducted research on reading emotion recognition. This article utilized the characteristics of blockchain technology to analyze the reading mode of blockchain internet platforms. By using audio and image bimodal recognition methods, the recognition of readers’ reading emotions can be achieved. After feature extraction of speech and facial images, hidden Markov models (HMM) can be used for speech emotion recognition. Support vector machines (SVM) can be used for facial image emotion recognition, and decision level fusion can be used for bimodal emotion recognition. This article obtained the final emotion recognition results to analyze and predict user reading behavior. Analyzing the psychological state of readers based on emotional recognition results can achieve more intelligent reading information push. Experimental results on the effectiveness of reading bimodal emotion recognition showed that the accuracy of reading bimodal emotion recognition based on decision level fusion was much higher than that of single modal emotion recognition. The bimodal method has an average accuracy rate of over 85% in emotion recognition and has a high effect in emotion recognition. Reading bimodal emotion recognition based on audio and image can accurately identify readers’ emotions, adjust information push content in a timely manner, and achieve the regulation of readers’ emotions, which has high application value.