Design And Implementation Of The Computer-aided Secret-level Classification System Based On Text Semantic Similarity

With the continuous development of computer and information technology, the work of guarding secrets is also facing new situations and new problems. The classification of secret-level work as the source and basic work of the guarding secrets, it's intelligent and efficient is particularly important. Currently, some of the departments and units were determined secret-level by artificial means, there will be exist some insufficiencies and problems, such as:determining secret's process is not standardized, determining secret's responsibilities are not clear; determining secret are difficult to grasp the scale accurately, and the work is hard to accumulate experience effectively, and so on. To solve the problems and deficiencies above, with the help of computer advantages such as information, intelligence and high efficiency, this paper design and implement a computer aided system to determine secrets automatically, this system will make the efficiency of the classification work effectively improved, and also can save a large number of human and material resources.Firstly, this paper expound the background of the secret classification work, and on the basis of the analysis of the prior technology, then we propose a solution for the occurrences of classification work problems, the solution is design and implement a computer aided secret-level classification system, the computer aided secret classification system is divided into two modules:1. According the description of Secrets Act, the provisions of state secrets and the security classification of the specific scope and secrets directory, we extract the rule of secret-level classification and then build the rule based on the tree structure. According to the confidential documents'business field, and through the word segmentation and word clustering to content of the document, the system retrieve and match the corresponding the determine secret rule base to determine the security classification of the document;2. If the first module can not match a rule base to set the document's secret level, proposing a Chinese text classification algorithm based on the weighted semantic similarity. This section focuses on text words' semantic concept of dimensionality reduction and disambiguation by Hownet, as far as possible to make the feature words reach orthogonal and to make the text of feature words with a smaller feature space vector representation. In the last of this part, we propose a weighted semantic similarity K-nearest neighbor algorithm to realize the secret-level classification of the text.Finally, we describe in detail how to implement the system, and then did a testing on text classification algorithm based on weighted semantic similarity, experimentation results show that our algorithm has higher Precision, Recall and F1-Measure, and the time is shorter, and this system can effectively determine the security classification of the target document.