Entity relation extraction is a crucial part of knowledge base construction. Among many methods of relationship extraction

the method of distant supervision combined with neural network model is satisfactory in terms of accuracy and other performance. However

there is often a large amount of noise data in the labeled corpus obtained by distant supervision

which has a great impact on the training of relationship extraction model. In this paper

we propose an entity relationship extraction model of convolutional neural network based on improved attention mechanism. Aiming at the sentence set containing the same entity pair

this model tries to find out all the positive instances that embody the relationship between the entity pair

construct the combined sentence vector

and discard the possible noise sentences

so as to minimize the impact of noise sentences and make full use of the semantic information of positive instances. Experimental results show that the accuracy of the proposed relation extraction model is better than that of the comparative relation extraction model.