Harshadeep

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

HarshaDeep

1. Develop a Virtual Agent that can provide details about a project.


2. Virtual Agent integrated with Skype for Business
Use case example:
For instance, if a manager would like to access any alerts or issues from a
project, in the conventional scenario, the manager was supposed to access the
details through a portal by entering the required fields - has manual effort and
time taking - unlike this process, the manager can directly interact with the
virtual agent available on skype and have the details right on the chat box.
1. Identifying user intentions
2. Extracting the Enitity from the sentence/query.
3. Project names, intents, and department names are compound words.
4. User journey
5. Retrieving the information/details for the query.
1. How do you keep track of the user intention?
2. what information did the user provided and what else do we require?
3. how do we ask for the unavailable details?
4. did the user change his intention?
5. if intention is changed what do we do about the previously unanswered
Questions?
6. how to check if the question asked is relevant to our use case . ex: how are
you? is not a relavant question
7. how to answer generic questions?
8. how to handle spelling mistakes?
Database

Development of Building AI
USER JOURNEY Engine

Virtual
Agent on
Skype
https://www.dropbox.com/s/j97zhrn2c8b4101/SFB%20Demo.mp4?dl=0
1 . Total 10 lakh samples
2. 1-2 lakhs samples were manually labelled and the rest all were labelled
through an automated process.

Examples of a labelled sentence:


HarshaDeep
Our goal here, firstly is to compose a database aka knowledge base by pushing
the standard contents or quality information from the mail and secondly build a
chatbot that could efficiently answer user queries by reading from the database
There are three steps in which the entire task is achieved:
1. Compose the database aka knowledge base – typically paragraphs with less than
1000 words.
2. Train a SVM model – Mapping Model – with the all the paragraphs from the
database. In other words, we can understand this as the domain/document
classification problem.
3.Train a Deep Learning Model – Question answering model – Once we have the
question from the user. We send it to the Mapping model to understand the domain
the question belongs to. Once we have the class/domain of the question. We pull the
respective paragraph from the database aka knowledge base and along with the user
Question we send these as inputs to the Question answering model. The Deep
Learning model outputs an appropriate answer extracted from the paragraph.
https://www.dropbox.com/s/7176rds965f4eui/KB%20Demo.mp4?dl=0

You might also like