【Purpose/significance】In order to address the problems of small scale and incomplete classification of sensitive words in ex⁃

isting sensitive dictionaries, a more complete and reasonable sensitive dictionary is introduced based on sensitive semantics and com⁃pound co-occurrence.【Method/process】Firstly, the initial word set, which is constructed on the basis of the sensitive words collected from multiple social platforms, is divided into the basic sensitive dictionary and the candidate sensitive word set through content analy⁃sis and category annotation. Secondly, a sensitive semantic extension word set is obtained by integrating the sensitive prior probability,sensitive semantic correlation and compound co-occurrence to extend the sensitive semantic information of the candidate sensitive words. Finally, according to the defined comprehensive sensitivity calculation method, the candidate sensitive words are filtered by measuring the comprehensive sensitivity of candidate sensitive words and sensitive semantic extended word set, and the construction of the extended sensitive dictionary is completed.【Result/conclusion】Compared with the existing sensitive dictionary, the accuracy, re⁃call and F1 value of sensitive information recognition is respectively improved by 17%, 24% and 22% based on the expanded sensitive dictionary proposed in this paper, and the comprehensive performance of sensitive information recognition can be effectively improved.【Innovation/limitation】This paper starts with the important basic resources for sensitive information recognition, and constructs a basic sensitive dictionary. The effective extension words are filtered by measuring the comprehensive sensitivity, and the extension of the ba⁃sic sensitive dictionary is implemented. The disadvantage is that the influence index mining of word sensitivity is not enough.