The recommender system plays a vital role in the field of online business. To improve the quality of the recommender system, in this research work we focus on enhancingthesentiment analysis process to generate the most efficient recommendations to the users. E-commerce website has given consumers a venue to share their opinion on different entities. These opinions have a high impact on other customers to decide whether to purchase a product or not. By taking advantage of this, some users follow unethical ways to promote their products. Analyzing these opinions manually is difficult, and thus Attention-based Bi-LSTM ConvolutionNeural Network (ABC) with RMSprop optimizer is proposed to extract the genuine opinion level of reviews automatically. The contributions are: (i) Development of the ABC model for text classification with fake review detection, (ii) Evaluation of the results in cross-domain datasets, (iii)Analyzing the impact of different optimization algorithms in the proposed neural networkthat can improve the classification task, (iv) Finally, sentiment analysis module is passed to recommender module to produce Top N recommendations.Extensive experiments are carried out on three different domain datasets and the results clearly indicate ABCwith RMSprop optimizer outperforms better than state-of-art text classification methods. © 2019, Institute of Advanced Scientific Research, Inc.. All rights reserved.