By Umesha Gunasinghe
CB002789
05/01/11
Overview
Problem Background
Key Decisions Made
Design
Development
Test Results
2/27
Problem Background
Issues
 Growth of student the population
 Travelling issues
 Cost
 Time
 Counselors have to answer the general questions every
time
3/27
Proposed Solution
Web-based Virtual Counselor
 Advantages
 Accessibility
 Vast coverage of target audience
 Saves Time
 Saves Cost
 Reduces the counselors workload
4/27
Domain Analysis
Aims of the Questionnaire
 Whether the proposed system is useful
 If so what are the features that should be included
 What kind of information is needed for the system
Results(based on 76 students)
 Visited APIIT Website in search for more details 72%
 Information available on APIIT website : clear 46%, not
clear 47%
 85% said that it would be useful if there was a chat
system
5/27
Key Decisions Made
AIML language for the knowledgebase development
ALICE’s input normalization for the input
normalization
ALICE’s pattern matching algorithm
6/27
7/27
AIML Scripting
•Knowledgebase writing style
with <category>
•Rich set of AIML tags
•Recursion using <srai> tag
•Categorization with <topic> tag
•Non intuitive input and output
rules
•Complexity and possibility of
error is high
•No recursion and few set of rule
categories
•Mainly based on keyword patterns
Decisions Made-Knowledgebase
AIML Samples
 Atomic Category
<category>
<pattern>WHAT IS THE TOTAL COST OF A DEGREE PROGRAMME AT
APIIT</pattern>
<template>It is around 1.3 million.*</template>
</category>
 Default Category
<category>
<pattern>* COURSES * AVAILABLE *</pattern>
<template>
<srai>WHAT ARE THE DEGREE COURSES AVAILABLE AT APIIT</srai>
</template>
</category>
8/27
AIML Samples
 Recursive Category
<category>
<pattern>* COURSES * AVAILABLE *</pattern>
<template>
<srai>WHAT ARE THE DEGREE COURSES AVAILABLE AT APIIT</srai>
</template>
</category>
9/27
Decisions Made-Response Generation
10/27
ElizabethALICE
•Well defined normalization
process
•Simple pattern matching
algorithm depends on depth first
search
•Partitioning and combination
•Efficient
1. underscore match
2. Atomic word
3. ‘*’ key match
•Not a well defined normalization
process
•Complex process with 5 phases
•Key word matching(gives response
to the first keyword pattern
matched)
•No partitioning and combination
•Inefficient (all the input/output
patterns and keywords should be
checked)
Example
 Normalization Steps
11/27
Example
 Input Path Creation
12/27
Pattern Matching Algorithm
13/27
Example
 Pattern Matching
14/27
Special and Additional Features
Special
 Spell Check
 Similar Wordset check
 Synonyms Check
Additional
 Temporary Conversation Log
 Admin Module
15/27
System Requirements
 Core System Requirement
 The system should be able to answer the user questions
accordingly in the scope of the knowledgebase.
 Special Features
 Spell, Similar Wordset and Synonyms Check
 Additional Features
 Temporary Conversation Log
 Admin Module
16/27
Design-Overall Architecture
Reasoning
Knowledgebase
Response
Generation
Web Interface
User Query
System Output
Input query
System Response
Selection of the best
match
Best
matching
answerTemporary
Conversation Log
Spell check Synonym check
Similar wordset
check
Admin Module
Admin Controls
New Knowledge
addition
Edit the
knowledgebase
Other admin
controls
Database
Counselor Module
Request
for
synonym
check
Synonym
replaced user
input
Request
for spell
check
Spelling
checked
user input
Checked for
previous user
questions
Check
confirmation
Request
for similar
wordset
check
Similar
wordset
checked
user input
Request for the
best matching
answer
Saving the questions that could not be
answered by the system according to the
existing knowledge
New additions to the
knowledgebase
Editing the existing
knowledgebase
answers
Request for the
unanswered
user questions Unanswered
user questions
Admin request for other
necessary data
System data according
to the request
17/27
Development
Followed Incremental Methodology (10 increments)
Knowledgebase implemented using AIML
Pattern matching and response generation
 Usage of AIML interpreter in C#
Spell Check
 Usage of Google Spell Checker API
18/27
Similar Wordset Check
 Limited word set selected
Synonyms Check
 Usage of Wordnet and C# Wordnet API
 Brill Tagger API
Temporary Conversation Log
 Saves 10 recent questions and answers
Development
19/27
Counselor Module
 Integration of Spell Check, Similar wordset Check,
Synonyms Check and Conversation Log
Admin Module Functionalities
 Edit knowledgebase
 Add new knowledge
 Add words to the similar wordset
Development
20/27
Test Plan
 The unit tests were performed, to test the main 7 units
(most important methods of the classes) of the
system.
 The system test was performed to test the overall
performance of the system. Collected 100 user
question were used to perform this test.
 The validation test was performed for the counselor
module involving two counselors of APIIT for an
overall performance check in two stages.
21/27
Test Results
 Validation Test
23/27
0%
20%
40%
60%
80%
100%
Round 1 Round 2
No answer
Answer
Test Results
 Unit Tests (7 main methods)
 System Test
22/27
System Test
Successfully
Answered(83%)
No direct
answer(5%)
Failed(12%)
Critical Appraisal
 Advantage in the usage of special AIML tags
(<srai>,<topic>)
 Introduction of Similar Wordset Check to reduce the
knowledge that should be written.
 Usage of “*” in pattern matching sometimes increases
ambiguity
 The Order of reasoning for the efficiency in the system
response
 System improves with the increase of the knowledge
with the help of the Admin Module
24/27
Limitations and Enhancements
 Limitations
 Limited Knowledgebase
 Temporary Conversation Log (10)
 Spell Check(only 1 suggestion)
 Further Enhancements
 Persona-Type feature
 User Analysis base
 Chat Session Report
25/27
Do you want to know more?
26/27
Thank You
27/27

Virtual intelligent student counselor for apiit

  • 1.
  • 2.
    Overview Problem Background Key DecisionsMade Design Development Test Results 2/27
  • 3.
    Problem Background Issues  Growthof student the population  Travelling issues  Cost  Time  Counselors have to answer the general questions every time 3/27
  • 4.
    Proposed Solution Web-based VirtualCounselor  Advantages  Accessibility  Vast coverage of target audience  Saves Time  Saves Cost  Reduces the counselors workload 4/27
  • 5.
    Domain Analysis Aims ofthe Questionnaire  Whether the proposed system is useful  If so what are the features that should be included  What kind of information is needed for the system Results(based on 76 students)  Visited APIIT Website in search for more details 72%  Information available on APIIT website : clear 46%, not clear 47%  85% said that it would be useful if there was a chat system 5/27
  • 6.
    Key Decisions Made AIMLlanguage for the knowledgebase development ALICE’s input normalization for the input normalization ALICE’s pattern matching algorithm 6/27
  • 7.
    7/27 AIML Scripting •Knowledgebase writingstyle with <category> •Rich set of AIML tags •Recursion using <srai> tag •Categorization with <topic> tag •Non intuitive input and output rules •Complexity and possibility of error is high •No recursion and few set of rule categories •Mainly based on keyword patterns Decisions Made-Knowledgebase
  • 8.
    AIML Samples  AtomicCategory <category> <pattern>WHAT IS THE TOTAL COST OF A DEGREE PROGRAMME AT APIIT</pattern> <template>It is around 1.3 million.*</template> </category>  Default Category <category> <pattern>* COURSES * AVAILABLE *</pattern> <template> <srai>WHAT ARE THE DEGREE COURSES AVAILABLE AT APIIT</srai> </template> </category> 8/27
  • 9.
    AIML Samples  RecursiveCategory <category> <pattern>* COURSES * AVAILABLE *</pattern> <template> <srai>WHAT ARE THE DEGREE COURSES AVAILABLE AT APIIT</srai> </template> </category> 9/27
  • 10.
    Decisions Made-Response Generation 10/27 ElizabethALICE •Welldefined normalization process •Simple pattern matching algorithm depends on depth first search •Partitioning and combination •Efficient 1. underscore match 2. Atomic word 3. ‘*’ key match •Not a well defined normalization process •Complex process with 5 phases •Key word matching(gives response to the first keyword pattern matched) •No partitioning and combination •Inefficient (all the input/output patterns and keywords should be checked)
  • 11.
  • 12.
    Example  Input PathCreation 12/27
  • 13.
  • 14.
  • 15.
    Special and AdditionalFeatures Special  Spell Check  Similar Wordset check  Synonyms Check Additional  Temporary Conversation Log  Admin Module 15/27
  • 16.
    System Requirements  CoreSystem Requirement  The system should be able to answer the user questions accordingly in the scope of the knowledgebase.  Special Features  Spell, Similar Wordset and Synonyms Check  Additional Features  Temporary Conversation Log  Admin Module 16/27
  • 17.
    Design-Overall Architecture Reasoning Knowledgebase Response Generation Web Interface UserQuery System Output Input query System Response Selection of the best match Best matching answerTemporary Conversation Log Spell check Synonym check Similar wordset check Admin Module Admin Controls New Knowledge addition Edit the knowledgebase Other admin controls Database Counselor Module Request for synonym check Synonym replaced user input Request for spell check Spelling checked user input Checked for previous user questions Check confirmation Request for similar wordset check Similar wordset checked user input Request for the best matching answer Saving the questions that could not be answered by the system according to the existing knowledge New additions to the knowledgebase Editing the existing knowledgebase answers Request for the unanswered user questions Unanswered user questions Admin request for other necessary data System data according to the request 17/27
  • 18.
    Development Followed Incremental Methodology(10 increments) Knowledgebase implemented using AIML Pattern matching and response generation  Usage of AIML interpreter in C# Spell Check  Usage of Google Spell Checker API 18/27
  • 19.
    Similar Wordset Check Limited word set selected Synonyms Check  Usage of Wordnet and C# Wordnet API  Brill Tagger API Temporary Conversation Log  Saves 10 recent questions and answers Development 19/27
  • 20.
    Counselor Module  Integrationof Spell Check, Similar wordset Check, Synonyms Check and Conversation Log Admin Module Functionalities  Edit knowledgebase  Add new knowledge  Add words to the similar wordset Development 20/27
  • 21.
    Test Plan  Theunit tests were performed, to test the main 7 units (most important methods of the classes) of the system.  The system test was performed to test the overall performance of the system. Collected 100 user question were used to perform this test.  The validation test was performed for the counselor module involving two counselors of APIIT for an overall performance check in two stages. 21/27
  • 22.
    Test Results  ValidationTest 23/27 0% 20% 40% 60% 80% 100% Round 1 Round 2 No answer Answer
  • 23.
    Test Results  UnitTests (7 main methods)  System Test 22/27 System Test Successfully Answered(83%) No direct answer(5%) Failed(12%)
  • 24.
    Critical Appraisal  Advantagein the usage of special AIML tags (<srai>,<topic>)  Introduction of Similar Wordset Check to reduce the knowledge that should be written.  Usage of “*” in pattern matching sometimes increases ambiguity  The Order of reasoning for the efficiency in the system response  System improves with the increase of the knowledge with the help of the Admin Module 24/27
  • 25.
    Limitations and Enhancements Limitations  Limited Knowledgebase  Temporary Conversation Log (10)  Spell Check(only 1 suggestion)  Further Enhancements  Persona-Type feature  User Analysis base  Chat Session Report 25/27
  • 26.
    Do you wantto know more? 26/27
  • 27.

Editor's Notes

  • #2 The era of technological and scientific evolutionMan tries to automate the tasks done by humans inventing the new machineries and applicationsY don’t we user the similar concept to reduce the work load of the APIIT counselors and make it easy for the prospective students to retrieve necessary information regarding APIIT degree courses easilyThat’s all bout my final year project
  • #4 *Counseling plays a major role in any university*APIIT counseling two major aspects*growth of the student population is 25% per year and the information requests*handled by only 3-4 counselors*some need more clarifications some don’t*specifically come to apiit premises
  • #11 Elizabeth*Adding, modifying and deleting scripts while conversation in progress*Both stores the previous output
  • #18 A link to a PDF file will be provided for a clear view of the design