ai chatbot service framework based on backpropagation network for predicting student's...

20
AI Chatbot Service Framework Based on Backpropagation Network for Predicting Student’s Performance James Hsieh P96044168 Benny Suryajaya P96057022

Upload: -

Post on 08-Jan-2017

134 views

Category:

Engineering


1 download

TRANSCRIPT

Page 1: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

AI Chatbot Service Framework Based on Backpropagation Network for Predicting Student’s Performance

James Hsieh P96044168Benny Suryajaya P96057022

Page 2: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

52%parentsworried about their children’s progress

Page 3: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

70%studentsnot sure about their grade

Page 4: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

AI Chatbot Service Framework Based on Backpropagation Network for Predicting Student’s Performance

Page 5: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Mr. Wang Features

Grade prediction by analyzing student’s behavior and their living environment

Prediction comes in 3 options: Poor, Average, Good

Fast reply via Facebook message

Demo video: https://youtu.be/_3xyxJ-ACxM

Page 6: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Process Flow of the System

Cloud Platform

User

via Facebook API

Send Message

Response Message

Receive Message

Response Message

via Facebook API

Page 7: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

System Framework on Cloud Platform

Function Model

Neural Network Algorithm Module

Semantic Analysis Crawler

Core

Node.js Mongo DB

Cloud Server ( Ubuntu OS)

Service

Application 1

Application n…

Page 8: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Crawler: Get Any Data from Internet

Page 9: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Semantic Analysis : Analyze the Best Response for User

Page 10: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Neural Network Algorithm Module

Provide neural network algorithm for applications on this system

Get the training data and target data from database, for training model

Calculate the result of input data using NN algorithm

Page 11: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Data Collection

Student Performance Data Set taken from UCI Machine Learning Repository (https://archive.ics.uci.edu/ml/datasets/Student+Performance)

Source: Paulo Cortez, University of Minho, Guimarães, Portugal

Contains student achievement in secondary education in Portuguese school.

Subjects: Mathematics (395 students) & Portuguese (649 students)

Each data consists of 30 data of behaviour & environment, and 3 grade results

Data is in literal description, enumeration needed

Page 12: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Data Collection

Sex Age Address Family Size Parent Status … G1 G2 G3

1 M 17 Urban <= 3 Divorced 9 10 11

2 F 18 Rural > 3 Together 15 14 12

… … … … … … … … … …

649 M 18 Urban > 3 Divorced 16 17 15

Portuguese Dataset

Page 13: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Methodology

Method used: Backpropagation Neural Network

Reason No information in the form of function f(x) Dataset available with sample of inputs and outputs Problem is a forecasting/prediction, not an optimization

Training data Inputs: column 1 – 30 (student behavior and living environment) Targets: column 31 (grade of student in first period)

AI Chatbot Facebook-based, made with Node.js

Page 14: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Methodology

Method used: Backpropagation Network

Number of hidden layers: 2 Layer 1: 3 nodes Layer 2: 4 nodes

Learning rate: 0.6

Page 15: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Performance

Page 16: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Training State

Page 17: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Regression

Page 18: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Limitation

Limited model accuracy

Number of training data too few

Model parameter

Too many input parameters

Questions to be asked by chatbot also become a lot

Less interactive chatbot

Too many questions to be asked

Page 19: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Further Improvements

Add more training data Current training data might be too small & not representative enough

Reduce the amount of input parameters Purpose is to shorten the amount of questions asked Principal Component Analysis may be used

Make full use of the crawler Using the crawler to get translation data, Mr. Wang can be made to speak in multiple

languages

Use another algorithm In case NN is proven to be not the best choice, we can use another algorithm to

enhance the robot ability for solving other kind of problem

Page 20: AI Chatbot Service Framework based on Backpropagation Network for Predicting Student's Performance

Conclusion

Parents & students are worried about the grade

Mr. Wang can predict how students would fare in his study

Mr. Wang uses a BPN model to predict student performance

Mr. Wang is not perfect and there are still rooms for improvement