The customer, a leading banking firm in, is embarking on a strategic journey to support their growth objectives. The customer wants to build a sentiment analysis model to make informed business decisions through actionable insights. Blazeclan proposed the customer with the implementation of a cloud-based sentiment analysis model.
The Need for A Sentiment Analysis Model
The customer was looking to build a sentiment analysis model so that they could address any client queries with quick actions. The client queries can be from multiple touch points like Social Media, email, Survey, Product reviews, etc.
With Sentiment analysis model, the customer seeks to get actionable insights for improving the customer experience, and making better, informed decisions. The requirement is focused on setting up a ML Pipeline for Data engineering and building Sentiments and Topic Models.
Blazeclan’s suggested Product model
Blazeclan proposed a product centric approach for a holistic solution:
- Data Sources can be from Voice based Call centers, text based reviews and social media data, question-answers from Survey data, etc.
- All of these can be captured, processed and analyzed in a centralized platform for Customer Feedback and Sentiments.
- There can be real time actionable insights to keep customers informed and followed up with agent actions to resolve.
- All the above can be captured and centrally analyzed as “Voice of Customer.
Blazeclan proposes the following Architecture for implementation of Sentiment Analysis model.
- Data ingestion of various sources using Queues, Streaming messages, Databases, SFTP, etc.
- Cleansing/masking of data removing PII and other personal information.
- Encryption of data using KMS
- Performing Translation service using Open Source libraries.
- Perform Data Engineering processes – Tokenization, Stemming, Lemmatization, etc.
- Sentiment Detection using AI based Cloud APIs.
- Extracting Key Phrases and Topic Detection using Supervised and Unsupervised Learning Algorithms.
- Capturing insights of Sentiments and Topics in BI Dashboards.
Benefits Achieved by the Customer
Automated Model Training and Inference: With the ML Pipeline, the topic detection models are trained periodically on new data. This keeps the process upto date and capture the right topics based on verbiage/comments.
Reuse of trained models in AWS: Amazon Comprehend service runs on pre-trained datasets that AWS has already makes use of to build Sentiment Detection models. This makes it easy for customers to not build models from scratch and saves time by just leveraging the APIs.
Actionable Insights: The cloud-hosted sentiment analysis model enabled the customer to get actionable insights on complaints for making better decisions and bringing overall improvement in end-user experience.
|Amazon Comprehend||AWS Lambda||Amazon S3|
|Amazon Sagemaker||Amazon SFTP||AWS Athena|