This paper presents a convolutional deep neural network (DNN) for relation classification, aiming to improve the performance of traditional methods that rely heavily on pre-existing natural language processing (NLP) systems. The proposed method extracts lexical and sentence-level features from word tokens without complex preprocessing, using word embeddings and a convolutional approach. Lexical features are derived from marked nouns, while sentence-level features are learned through a convolutional network. These features are concatenated and fed into a softmax classifier to predict the relationship between two marked nouns. The experimental results on the SemEval-2010 Task 8 dataset demonstrate that the proposed method significantly outperforms state-of-the-art methods, highlighting the effectiveness of the learned features and the proposed position features (PF) in capturing the relative distances between target noun pairs.This paper presents a convolutional deep neural network (DNN) for relation classification, aiming to improve the performance of traditional methods that rely heavily on pre-existing natural language processing (NLP) systems. The proposed method extracts lexical and sentence-level features from word tokens without complex preprocessing, using word embeddings and a convolutional approach. Lexical features are derived from marked nouns, while sentence-level features are learned through a convolutional network. These features are concatenated and fed into a softmax classifier to predict the relationship between two marked nouns. The experimental results on the SemEval-2010 Task 8 dataset demonstrate that the proposed method significantly outperforms state-of-the-art methods, highlighting the effectiveness of the learned features and the proposed position features (PF) in capturing the relative distances between target noun pairs.