Abstract:-Web mining plays crucial role in modern technological world. It helps in automating different mining procedures in order to gain useful insights. Since web mining involves textual contents most of the time, NLP plays vital role in in understanding the data. It does mean that NLP and Machine Learning (ML) go hand in hand in processing such data. Many existing methods that are used for web mining through ML and NLP suffer from mediocre performance. To overcome this problem, in this paper, we proposed a framework known as Web Mining for Sentiment Analysis Framework (WMSAF).It has strong pre-processing methodology that exploits multiple procedures including NLP and ML based approach for machine learning towards useful analysis of data. We considered e-commerce case study dealing with automatic data collection from Amazon product reviews web URLs and perform pre-processing, NLP and machine learning towards automatic sentiment analysis. Our methodology is based on the BERT model which is found efficient when compared with other models. We proposed an algorithm named Web Mining and Sentiment Analysis of Amazon Product Reviews (WMSA-APR) to realize the proposed framework. From the results it is observed that the proposed BERT based method showed highest accuracy with 96% when compared with existing models such as UniLM, Reformer and XLNet.