| Learning the profile of each user based on their posts in different markets allows for better cross-market aggregation of user information,which helps identify anonymous users across sites.Therefore,the crosssite association method of dark web users has important application value.After recent years of research,dark web users cross-site association method has achieved some research results.However,due to the strong anonymity of dark web data,this method still has many key problems to solve.Different from the open web,the user label and attribute information of the dark Web are extremely lacking,and the explicit association between users is also lacking,which limits the applicability of traditional methods and increases the difficulty of analyzing the users of the dark web.Based on the actual needs of crime governance in the dark web,this paper takes post information published by network users as the core to carry out research on cross-site dark-web user multi-account association methods,which mainly includes three aspects:user representation model,user multiaccount association method and dark-web map,so as to more effectively associate and describe users engaged in illegal activities in the dark web.It provides some method support for law enforcement departments to deal with dark net crimes.The main research work and achievements of this paper are as follows:1.A dark web user representation model based on text implicit features is proposed.Firstly,in order to solve the problem of frequent changes in the length of text posted by users in the dark web market,the attention mechanism and convolutional neural network are used to mine the text features from different perspectives of global text and local text,and the adaptive gate mechanism is used to eliminate the positive noise produced by two processing of text data.Then,we extract the time features of users’ posts and introduce heterogeneous information network to extract the behavioral features of users.Finally,the features extracted from the user information are integrated into the set,which is used as the representation model of dark web users.2.This paper proposes a method of cross-site dark web user multiaccount association based on temporal context features.First of all,since the existing methods do not take into account the temporal correlation between posts published by users in the dark web market,this paper uses sequential convolution to extract temporal context features of posts published by users based on the dark Web user representation model from the previous research point,so as to enhance the user representation capability of the model.Then,based on the obtained user representation vector,dark network users are associated by calculating the cosine distance between different user representation vectors.Finally,this paper conducts joint training on four different dark web market data sets,so as to complete cross-website dark-web user multi-account association.This paper verifies the effectiveness of the association method on four publicly available dark Web market data sets.Compared with the mainstream approach,the Mean Reciprocal Rank(MRR)and standard recall rate(Recall@10)were respectively increased by 23.5%and 25.6%.3.A dark net atlas analysis system is designed and implemented.First of all,due to the large scale,strong temporal dynamics and sparse discrete problems of the dark web atlas,this paper builds a knowledge representation model and ontology model based on the results of cross-site user multi-account association from the previous research point,which is based on time characteristics and uncertainty characteristics.Secondly,knowledge extraction is carried out on the dark Web data based on ontology model,and the dark Web data is stored in ElasticSearch database based on the improved knowledge representation model.Finally,the front-end display interface of the dark web map is constructed,and the display and sample analysis of the dark web knowledge map are conducted.The effectiveness of the dark Web atlas analysis system is verified on the dark Web market data set. |