SCHEDULING

ASSISTANT

JUNO LEE

It's like speaking with a human

Your personal assistant

Available anytime and anywhere

Siri's Response:

My Bot's Response:

How does this work?

Lots of Survey Data

Natural Language Processing

Machine Learning

Survey

Good Responses

Not so Good Responses

Random Forest Classifier

99.6% Accuracy

Intent Classifier

Tokenize, lowercase, & lemmatize messages

Raw messages & labels

Entity Extraction

Entity Extraction Challenges

Missing information (am/pm, date)

Recognizing uncommon entities

Labeling vs hacky heuristic solutions

Completed: Program that takes a message and outputs the intent & entities in the terminal

Next: Interact with the Google Calendar API

Google Calendar API

Extract entities

Creating Event

Create event with entities

Insert event in Google calendar

Displaying Events

List next 10 events in Google Calendar

Format

Get event names & start datetimes

Node.js HTTP Server

Web App Deployment

run model predictions

Client

Python HTTP Server

host web app

render web app

Python Server

Node Server

NLP data from surveys are messy

Often, data > models

Entity Recognition can be quite difficult

Get requests everywhere

Sometimes the coolest features are easiest to add

What I Learned

Juno Lee

junoleelj@gmail.com

junolee

junoleelj

SCHEDULING ASSISTANT