摘要： 现有的效应概念图匹配方法多数存在匹配容错性差的问题。为此,从大数据的角度提出一种新的挖掘专利与效应对应关系的方法。利用长短期记忆网络(LSTM)与基于attention的双向LSTM相结合形成模型训练专利语料,通过Softmax分类模型进行分类,得到专利所属的效应。实验结果表明,该方法利用Bi-LSTM-ATT模型进行训练对判定专利所属效应具有一定的可用性,准确率可以达到70%以上。

关键词: 深度学习, attention机制, Bi-LSTM-ATT模型, 专利分类, 产品创新

Abstract: The existing concept graph matching methods have the problem of poor matching fault tolerance.Therefore,this paper proposes a new method to explore the relationship between patent and effect.Long Short-term Memory(LSTM) is used to combine with bidirectional LSTM which based on the attention mechanism to train the patent corpus.The Softmax classification model is used to classify the patent and the effect of the patent is obtained.Experimental results show that the method uses the Bi-LSTM-ATT model to train has certain validity in determining the effect of patents,and the accuracy can reach more than 70%.

Key words: depth learning, attention mechanism, Bi-LSTM-ATT model, patent classification, product innovation

中图分类号: 

• TP18