It’s been an exciting time with a whole stack of new features coming out of the 2017 Amazon re:Invent conference. Today I thought I’d focus on one of the newly released Machine Learning services, which really excites me – Amazon Comprehend. It’s a powerful tool to help facilitate your decisions and understanding.
The AWS Website describes it like this:
Amazon Comprehend is a natural language processing (NLP) service that uses machine learning to find insights and relationships in text. Amazon Comprehend identifies the language of the text; extracts key phrases, places, people, brands, or events; understands how positive or negative the text is; and automatically organizes a collection of text files by topic.
Essentially what it means is that you can feed Amazon Comprehend qualitative data for processing. For example, Online Store reviews, Customer Surveys, to perform an analysis and understanding at scale!
Take the following scenario: You’re the CEO of a fast food Mexican restaurant chain. You’ve had successful year-on-year growth, but this year, revenue has declined. You’ve got sales data, wholesale supplier order data, and membership data, and you’ve had business analysts look over the data but nothing stands out. What’s happened?
To get an understanding of what is happening, you might want to survey your customers.
But there’s a problem, you have membership details for 100,000 customers. How could you and your team possibly go through potentially 100,000 customer responses? Especially if you provide a space for a text based response. Previously, organisations typically approached this problem in a couple of ways:
- Rather than getting 100,000 customer responses, you send out a survey to a smaller subset of customers (for example: 100 people) or;
- You send out the survey to all 100,000 customers, inclusive of quantitative responses (0=Bad, 5=Average,10=Excellent), but no space for qualitative responses (free-form text response).
The problem with the first way is that you’re missing a substantial amount of customers – how do you ensure that the right cross-section of customers are targeted? Do they accurately represent the whole client base?
The problem with the second way is that you’re restricting yourself to responses that you’ve thought about. What if the issue is something so unique or abstract that you couldn’t possibly have included it with a sliding scale of sentiment?
Wouldn’t it be great if you could truly understand and not limit your investigation? With Amazon Comprehend, you can feed potentially thousands of text based responses and have it provide meaningful insights. Entity identification, Key phrases identification, Language identification and even Sentiment analysis can be extracted. Topic Modelling extracts common themes and discovers what your customers are talking about.
Turns out that the due to seasonal changes, customers in certain warmer-climate areas were more frequently ordering cooler meals – like salads. To cater for the demand and to reduce costs, some hot and spicy meals were removed from the menu. This resulted in reduced choice and during the cooler season, the hot & spicy meals were not re-added to the menu. This resulted in fewer customers returning to your restaurant.
Need some help with understanding your data, or using Natural Language Processing (NLP) or leveraging Machine Learning? Idea 11 can help you plan and implement an analysis across your data. Contact us if you’d like a discussion on how we can help.