Research on case preprocessing based on deep learning.

Considering the problem of missing fields in the criminal case system, this article proposes a deep learning algorithm to extract the features of the case description and fill in the missing value. Due to Chinese expressions and characteristics of criminal cases, we make both character vectors and word vectors to present text embedding. Character vectors are from bert model. Word vector is trained by long short‐term memory model with attention. The experiment uses 13,890 data totally. This work is an extension of our short conference proceeding paper. The results show that the combination of characters and words can effectively improve the accuracy of the conference paper by 9%. This is the first time to cascade the character and word dimensions on the criminal case information preprocess and it can provide higher quality data especially for the crime data mining. [ABSTRACT FROM AUTHOR]