The paper addresses the challenges of distant supervision in relation extraction, particularly the wrong label problem and the reliance on ad hoc features. It proposes a novel model called Piecewise Convolutional Neural Networks (PCNNs) with multi-instance learning to tackle these issues. PCNNs automatically learn relevant features without complex NLP preprocessing, using a convolutional architecture with piecewise max pooling to capture structural information between entities. The multi-instance learning approach helps mitigate the wrong label problem by treating bags of instances rather than individual instances. Experimental results on a widely used dataset show that PCNNs outperform several competitive baseline methods, demonstrating the effectiveness of the proposed approach in distant supervised relation extraction.The paper addresses the challenges of distant supervision in relation extraction, particularly the wrong label problem and the reliance on ad hoc features. It proposes a novel model called Piecewise Convolutional Neural Networks (PCNNs) with multi-instance learning to tackle these issues. PCNNs automatically learn relevant features without complex NLP preprocessing, using a convolutional architecture with piecewise max pooling to capture structural information between entities. The multi-instance learning approach helps mitigate the wrong label problem by treating bags of instances rather than individual instances. Experimental results on a widely used dataset show that PCNNs outperform several competitive baseline methods, demonstrating the effectiveness of the proposed approach in distant supervised relation extraction.