In e-commerce applications, user reviews considered as significant information to increase the customer and revenue of the product items. Product or item recommendation to user relies on analysis of the online review and clustering of the user similarity. Conventional model towards recommendation on analysis of the review is carried out using the machine learning model. However machine learning model suffers from analysing the fake online review due to curse of dimensionality and data sparsity issues. In this paper, an improved convolution Neural Network is designed to analyse and detect the fake review to the products. Improved convolution Neural network is established on incorporation of the sentiment analysis process in the fully connected layer of the convolution neural network. Initially preprocessing is carried out to remove the stop words and to carry out the tokenization and stemming process. Preprocessed data is employed to the latent Dirichlet allocation mechanism to capture of the intention of the user on the review and latent features of the user is gathered on user profiling. Feature Embedding is carried out and resultant feature is applied to the improved convolution neural network. Improved convolution neural network process the embedded feature in convolution layer, max pooling layer and fully connected layer to identify the fake reviews. Experimental analysis is carried out on the Amazon review dataset. Performance analysis of the proposed model is evaluated against the conventional model on the measures of the accuracy. It proves that proposed model achieves the 99.25 percent accuracy compared to state of ar approaches.