In the field of public opinion monitoring in minor languages, it is difficult to obtain annotated corpus in minor languages, resulting in poor practice of deep learning. It is difficult to extract the core opinions of the information published by the private and the media for further analysis of public opinion. Taking Vietnamese as an example, a method for extracting Chinese-Vietnamese news opinion sentences that incorporated shared topic features was proposed. The problem of scarcity of small language resources was solved with the help of sufficient Chinese annotation corpus, and the topic information was used to shared between bilingual comparable corpora. It could optimize the extraction effect, and then enhance the public opinion monitoring effect. First, the topics of Chinese-Vietnamese comparable news were extracted separately to construct shared topic features through LDA topic modeling; then, the bilingual word embedding model was trained to achieve shared semantic spatial representation of Chinese and Vietnamese; Finally, the features were integrated with word vectors for combining semantic information with topics, emotion and location information to enhance the effect of extracting consequent. The experimental results in the Chinese Vietnamese comparable news dataset show that the integration of shared topic features can improve the extraction of Chinese Vietnamese news opinion sentences, the value of F₁ is 0.721, which supports the monitoring of public opinion in minor languages.