The minimize the response time and human involvement in responding to general user queries.
What it does
User records a query message and receives an automated reply to the query generated by the information retrieval system (of FAQs) in less than a minute.
How we built it
Twilio’s TwiML APIs were used to process the user query.
User’s voice_response recording is transcribed using TwiML. This transcribed_text is then searched in the FAQs DB and the relevant response is retrieved. An outgoing call is then made to the user which reads out this response using TwiML(response.say).
When a user calls the Twilio registered Wordpress number, he/she is prompted to record the question as a voice message. This voice message is transcribed by the TwiML API and passed to the Information Retrieval System. The answer returned from the IRS is then sent to the user with a callback. Only when a question is not
Information Retrieval System:
We scraped the Wordpress FAQ page to get the question and their respective answers which were stored in a data structure as key-value pairs. The transcribed_recording i.e the input query is cleaned of its stopwords(the topically irrelevant terms with the highest frequency eg: the, a, as etc.). The query is then searched in our data structure and all the keys(questions) are ranked according to their cosine similarity with the query. The value of key with the highest cosine similarity is returned.
The application then makes an outbound call to the user and reads out the returned message.
Challenges we ran into
Using Twilio API, ranking the questions in IRS, running the app on aws.
Accomplishments that we're proud of
We were able to think of a useful idea and creation of a modular prototype of the proposed approach.
What we learned
Working of Twilio APIs, usage of AWS and ssh, also explored chalise.
What's next for QuickFAQs
- It can be useful for visually impaired customers.
- People who don't have access to the internet can use this feature.
- It can be used for large-scale employee training purposes, which would result in major decrease in the training costs.