Understanding Consumer Product Sentiments through Supervised Models on Cloud: Pre and Post COVID

Abstract

While a lot of work is done on extracting sentiments and opinions in unstructured text, majority of it is focused on contextual sentiment mining and features that are more focused on sentiments. The team attempted to use contextual text analytics to identify product or service features that drives the sentiment of the user. This is done through application of cosine similarity and neural networks. Customers speak about product or service feature when it is important for the them. The second stage of the analysis is focused on supervised learning, that identifies key drivers of a product or service. It helps in deriving those elements which are subconsciously being evaluated by customers but not spoken. We also test the significant difference in views of people pre and post Covid in their reviews. We found that factors related to Covid have gone up by 30% but not statistically significant. Given the volume of data, the team has analyzed these on cloud to assess the cloud computing readiness for such analysis. Feedback around the post Covid topics helps us understand the issues that need to be addressed by restaurant industry.