基于改进BERT模型的吸毒人员聊天文本挖掘

TEXT MINING OF DRUG USERS'CHAT BASED ON IMPROVED BERT MODEL

对涉毒人员聊天文本进行语义分析,可快速精准地挖掘出海量复杂网络中涉毒人员并及时追踪调查。利用带有方言特色和特定语境的吸毒信息采集平台的吸毒人员聊天文本数据,采用改进BERT模型训练涉毒人员聊天文本,通过学习上下文语境,对聊天文本的语义挖掘效果显著,在准确率、召回率和F1值均优于贝叶斯模型,对涉毒文本正确分类准确率达到90%。对具有方言特色暗号的聊天文本数据,BERT模型可以高效地挖掘潜在涉毒人员,为禁毒部门对涉毒人员管控提供决策辅助。

Semantic analysis of the chat texts of drug related personnel can quickly and accurately excavate drug related personnel in massive and complex networks and follow up and investigate in time.Anti-drug mobile phone collection platforms with dialect characteristics and specific context were used to collect chat text data of drug related persons,and improved BERT model was used to train chat texts of drug related persons.The semantic mining effect of chat text was remarkable by learning the context.The accuracy rate,recall rate,and F1 value were better than the Bayesian model,and the accuracy rate of correct classification of poisonous texts reached 90%.For the chat text data with dialect characteristic codes,BERT model could efficiently mine the potential drug related personnel,and provide decision support for the drug control department to control the drug related personnel.