The paper introduces a neural network model for document-level sentiment classification, addressing the challenge of encoding intrinsic relations between sentences in a document's semantic meaning. The model first learns sentence representations using convolutional neural networks (CNN) or long short-term memory (LSTM), and then adaptively encodes sentence semantics and their relations into document representations using gated recurrent neural networks (GRNN). Experimental results on four large-scale review datasets from IMDB and Yelp Dataset Challenge show that the proposed model outperforms several state-of-the-art algorithms, demonstrating superior performance in sentiment classification. The study also highlights the effectiveness of GRNN in capturing complex linguistic relations compared to standard recurrent neural networks (RNN).The paper introduces a neural network model for document-level sentiment classification, addressing the challenge of encoding intrinsic relations between sentences in a document's semantic meaning. The model first learns sentence representations using convolutional neural networks (CNN) or long short-term memory (LSTM), and then adaptively encodes sentence semantics and their relations into document representations using gated recurrent neural networks (GRNN). Experimental results on four large-scale review datasets from IMDB and Yelp Dataset Challenge show that the proposed model outperforms several state-of-the-art algorithms, demonstrating superior performance in sentiment classification. The study also highlights the effectiveness of GRNN in capturing complex linguistic relations compared to standard recurrent neural networks (RNN).