Instant: Your Personal Customer Support Team

Inspiration

Most of Microsoft users would encounter products issues from time to time. It is inconvenient and troublesome to find your issues from millions of online posts or to contact services staff for help. Would it be great if there is a conversational bot to coach you on how to solve your problem? Our “Instant” can serve as your personal customer support team and help you resolve the issues instantly.

What it does

We built a customer support bot called Instant using Amazon Lex and Alexa for troubleshooting of Microsoft products, to save you hours of contacting their customer support team.

How we built it

We utilized the several services provided by AWS including Elasticsearch and Lex to build a conversational bot called "Instant", and then we loaded more than 10,000 QA pairs we scraped from Microsoft community, an online forum to troubleshoot Microsoft products. The rich QA data covers a wide range of Microsoft products, making sure “Instant” can provide an answer in most situations. In order to empower “Instant” of the ability to better match the problem asked by the user in the database, we came up with two different ways to expand the original query in. One way is using a word embedding model trained with the all the questions in the QA pairs and the other one is to add synonym extracted from an online API for all the verbs in the sentence.

Challenges we ran into

It is our first time joining a hackathon, we did not finalize our project topics until the second day, which left us not much time to evaluate the project. We are also not familiar with Amazon Lex and ran into a lot of problems when trying to connect Alexa with ”Instant”. We went through the documents in detail and tackle the issues. We also came up with two ways of query expansion in the effort of improving the results.

Accomplishments that we're proud of

We built a customer service bot called “Instant” that is empowered with more than 10,000 QA pairs about Microsoft products using Amazon Lex and Alexa. Again, this is our first hackathon and we ran to the wrong path in the beginning. But we did manage to finish the project in the remaining time.

We built three prototypes:

Instant Alpha, loaded with 10K QA pairs
Instant Beta, loaded with 4K QA pairs with shorter questions and query expansion using word embedding
Instant Gamma, loaded with 4K QA pairs with shorter questions and query expansion using verb synonyms All three prototypes can be accessed with Alexa, with invocation phrase of "Instant Alpha", "Instant Beta", "Instant Gamma".

What we learned

We got to know how to utilize Amazon Web Service including Lex and learned how to improve the conversational search results using query expansion with word embedding and synonyms.