S2SA-BiLSTM:面向法律纠纷智能问答系统的深度学习模型

S2SA-BiLSTM:a Deep Learning Model Towards Intelligent Q&A System for Legal Disputes

利用司法领域的网络资源对用户所提的法律纠纷问题自动提供有价值的参考解决方案能在很大程度上降低人工成本和社会资源.为了智能化解决用户的法律纠纷问题,论文提出一种seq2seq(Sequence to Sequence,序列到序列)融合注意力模型和双向长短时记忆网络的法律纠纷自动问答深度学习模型——S2SA-Bi-LSTM.该模型从大规模法律纠纷问答对出发,利用BiLSTM获取输入序列的上下文信息,结合注意力机制对序列权重进行更新,通过编码得到输入序列的向量化表示.并在此基础上,本文修改了Bi-LSTM中输入门和遗忘门的参数并计算输出序列,以得到与输入序列相对应的输出序列.实验证明,所提模型在真实的数据集上生成的答案具有较高的准确率,MAP值和MRR值也优于已有研究.

The use of network resources in the judicial field to automatically provide solutions to the legal disputes raised by users reduces the labor cost.In order to solve the legal disputes of users intelligently,this paper proposes an automatic question answering model of seq2 seq(Sequence to Sequence)fusion attention model and bidirectional long short term memory——S2 SA-Bi-LSTM.The model starts with a large-scale legal dispute Q&A pair,uses Bi-LSTM to obtain the context information of the input sequence,combines the attention mechanism to update the sequence weight,and obtains a vectorized representation of the input sequence through coding.Based on this,this paper modified the parameters of the input gate and the forgotten gate in Bi-LSTM and calculated the output sequence to obtain the output sequence corresponding to the input sequence.Experiments show that the proposed model has a higher accuracy in the answers generated on real data sets,and the MAPvalue and MRR value are also superior to existing studies.