Entity perception of Two-Step-Matching framework for public opinions

•A Two-Step-Matching method is designed to identify the precise target of Chinese entity from ambiguous user comments of public opinions.•BiLSTM-CRF model is used to extract potential entity and TF-IDF model is used to extract characteristic words from user comments.•Jaro–Winkler distance algorithm is used in the first matching process, where a business directory is built according to entity registration details.•In the second matching process, an industry-characteristic dictionary is introduced to identify precise target entity if ambiguity exists.•Associated rate and accuracy rate are used to evaluate the effect of the method. And two indicators are improved remarkably.