09. Recommendations
Recommendations:
After evaluating reasons behind customer behaviour of customers in different segment, we can try to develop personalized marketing strategies for customer in different segments.
Customer Review Sentiment Analysis: As the features are fresh, we can continuously make changes to our system, from data collection to model improvements. Note that our recommendations below are part of an unexhaustive list of recommendations.
A proposed integration into our system with adjustments includes:
-
Model Improvements: Search for better pre-trained models than our current ones or build our own model using this dataset as our training data, to increase accuracy. Another way would be to change the way we collect reviews from customers; getting them to give their review in a specific format suitable for our models.
-
Create data pipelines: For large e-commerce businesses in the world today, many hire data engineers to handle data flow across pipelines, turning raw data into usable form for our analysis. To keep up to date with a significant inflow of real-time reviews, we need to use data engineering tools such as Apache, Snowflake and Amazon Web Services. While our current project does not use these, we hope to integrate them into the data science workflow in our e-commerce platform.
-
Further Model Improvements: Consider improving our model to handle reviews of different languages, so we can expand our business to other countries. We can also conduct deeper-scale analysis on reviews, such as demographic and income groups, getting insights into each group’s interests and tuning our marketing/product handling methods to fit their needs.
-
Pre-launch: Implementation of systems to handle large numbers of incoming reviews and firewall to prevent hackers from getting unauthorized access to our systems and customers’ sensitive data. This phase ensures that system failures are avoided, and if there are, a prompt response is needed.
-
Post-launch: If possible, consider improving the model. Future plans include the implementation of new systems to cater to customer groups’ activities, based on insights gained from different customer groups’ reviews (such as category of products bought or customer characteristics like income, age or race).
Below is an image of our preferred roadmap to integrate the Large Language Model system for aspect-based sentiment analysis to guide you through each recommendation.
We can follow a similar integration path for our general sentiment analysis and extraction of issues feature.