Sales Forecasting of Overrated Products: Fine Tuning of Customer’s Rating by Integrating Sentiment Analysis

Enhancement of the profitability of any business organization is driven by proper forecasting. However, this is challenging as many factors affect the forecasting and the degree of relevant factors changes over time. Henceforth, it is essential for any business organization to develop a reliable and consistent sales forecasting model that can drive their growth. In today’s business environment, customer ratings play a pivotal role in evaluating business performance, particularly in online retailing. These ratings provide valuable insights into the strengths and weaknesses of a product or service. The rating values are generally a set of integer values within a given range. This policy restricts users from expressing their views as they may wish to give a value that is not an integer. Hence, the system fails to capture the actual view of the customer about a certain product or service. As the intermediate values (decimal values) are not permitted, customers are generally compelled to round up their ratings, resulting overrating products. This problem can be addressed if textual reviews from the customers are recorded and these are analyzed for judging customers’ satisfaction level. In this research work, we compute customer satisfaction by analyzing the review text of each customer for a particular product by using VADER sentiment analysis tool and use this result for tuning the actual user given ratings. A novel model is proposed to consider the tuned average customer rating amalgamating with standard forecasting methods like ARIMA, SARIMA, and LSTM. The experimental results on the Amazon dataset reveal 10% to 96% improvement in forecasted values for different types of products.