You are on page 1of 16

IA & Cognition project

[5 DS ]

A Conceptual graph Based Recommendation System


for monitoring RM issues : Ontology Learning

Supervised By :
Wiem Zaouga & Bilel Farah
Technologist Assistant at Esprit Engineering School

2023/2024 Esprit Tunisia 1


Outline

Project contexte & Motivation

Project outlines

The Final output: illustrations

2
Context

Project Risk Management (PRM), a knowledge-intensive process with processes


and tools, requires effective management of risk-related knowledge that can assist
in making the right decision by managers [elyamani et al., 2023].
PRMoverview.png

Only 32% of IT projects had been delivered on time, budget and with quality
requirements as initially specified [Standish Group International, 2020].

Project Failure

• Inadequate standardization of PRM terminologies [ Fitsilis et al., 2014].


• The lack of a cohesive body of knowledge with a shared conceptual framework among
project stakeholders [ Toskova, 2019, Verner & Sarwar, 2022].
• Different overlapping PM best practices, standards and guidelines [Al-Shaaby &
Almessabi, 2018].
• PM frameworks language is not computer interpretable [ Park & Park, 2020].
• Some processes, such as risk monitoring and control, are not integrated.
3
Context

Iron Triangle [PMI, 2020]


How
PRM?challenges

1 To handle the PRM terminologies problem with respect to PM practices ?

2 To formalize the content of such practice from a human readable format to a


computer
interpretable format ?
4/40 12/01/20
Context

PMBOK semantic weaks


Project Management Body Of Knowledge Why ?
(PMBOK) [ PMI, 2017]
Generic body of knowledge with rigorous 1 Lacks of agility and strategic view by
specifications of 49 Processes, 119 Tools and using
Techniques, considered as the reference all the available knowledge/tools.
practical PM Framework . 2

The continuous revisions with


Standardizes generally accepted PM translation
concepts and practices. into various languages.
Provides a common vocabulary with a
3 Disagreement and terminological
comprehensive set of definitions. ambiguities
Lacks the focus on the Knowledge Flow

Modelling the content of PMBOK via a conceptual Theory.graph


Attaches a formal semantic representation to PMBOK content (machine-
Semantic readable format).
interoperability issue
As the PMBOK 6th proposes several practices around PM with rich terminologies, it
Establishes an internal integration within processes in terms of inputs, T&T and outputs,
and is considered ideal for acting as the PM corpus in this work.
between the KAs.
Improves the PMBOK interoperability level.
5
Context: Conceptual graph_ Ontology

Ontology learning (OL)


Ontology
Text Ontology issues

The process to acquire and represent


A formal, explicit specification of ina shared 1 None of them focus on representing the
knowledge from unstructured data a format
conceptualization
machine [Gruber, 1995, Studer
RM knowledge according to best practices.
et al,1998].
understandable using NLP techniques [Al-
Aswadi et al., 2020]. OL layer cake 2
Lack of a reference ontology on RM.
Provides a consensus of knowledge.
3 Most of them are manually developed.
Ensures the semantic interoperability
among the project stakeholders. Laborious and time consuming process.
Acts as a knowledge repository to share, 4 Ontology and recommendation System
reuse and reason over it.
are implemented separately.
Improves the accuracy and precision
Reasoning or inference tools are limited
under the project implementation.

6
Motivation & Objectives
Motivation

In order to assist practitioners and researchers to better study the impact of their decisions and
evaluate the Project Risk (PR) as precisely as possible :
1 Propose an automatic building approach on Project Risk Management (PRM) to obtain a
conceptual graph
Learn from unstructured corpus “PMBOk 6 th ”: the knowledge retrieval process
Enhance the resulting ontology according to PMI standard for RM : ontology enrichment

Ontology Learning Layer Cake [ Buitilaar et al., 2005]


Considered as the cornerstone in OL [ Maimon & Browarnik, 2015], where six layers are
defined. Based on these layers, the process of OL can be divided into six tasks.

Natural language processing techniques


Techniques parse a formal representation and a meaning of a text written in the form of
natural language, using transforms models (S_BERT/ TOPIC BERT/ robeta models)

7
Motivation & Objectives

Objectives

2 The resulting conceptual graph will be enriched with the extracted knowledge from the PMI
standard for RM.

3 Implementing a recommender engine into web application that infers personalized


recommendation from the proposed conceptual graph.

8
system architecture

9
Project Outlines: Steps to be followed

The automatic knowledge retrieval process from unstructured text which is the
PMBOK 6th 41 pages [from 309 to 354 page]) and PMI’s standard for PRM (116
pages) using NLP techniques using transformer models.

This retrieved knowledge will be parsed following OL layer cake into concepts ,
Object/Data properties, axioms ,rules for building conceptual graph such as
“ontology” by applying machine learning techniques and deep learning
architectures to structure the ontological elements.

Based on graph repository, Recommendations will inferred using the


recommendation techniques.

Implementing an conceptual graph based recommendation system as a web


application to infer a real-time personalized recommendation according to project
stakeholder’s requests.

10
Project Outlines: Steps to be followed

knowledge retrieval module (1): aims to extract the relevant knowledge (concepts,
instances, relationships) from PMI’s corpus which is the reference best practice on
PM.
Conceptual graph building module (2): With respect to the OL process, we
automatically structure and represent the PM knowledge into OWL, RDFS, RDF
format. Then, a recommendation engine will be developed to encapsulate the
interdependencies between PM concepts (e.g. “IF PR has probability value between
0.8 and 1 and impact rating value 100, THEN we obtain very high exposure and
score value 100”).
The recommendation module (3) aims to infer personalized recommendation, advices
on whether additional information is needed and suggests ways to obtain such
information using recommendation techniques and approaches.
EXAMPLE:
For each process, the related tools and the deliverables (documents, lesson learned)
are inferred based on user’s search query and profile.

 A web-based system is implemented to support practitioners (e.g. project manager,


risk owner, risk action owner) activities and then make the right decisions.
11
The Final output: illustrations

The steps followed for building PRM-Onto

Step 6:
Step4:
3:
2:
5: Non-Taxonomic
1:Weighting
Concepts
Data preprocessing
Concepts
Synonyms and
conceptsrelations
properties
hierarchy
extraction pertinence
discovery
The knowledge retrieval process
Segmentation is to recognize
Identify the concepts the boundaries
list (NN) and propertiesof sentences
(VB) [from 309 tothe
list from 354].
set of
1
Term Prepare
stemmed.
Through Spacy and
Rule analyze
Frequency-Inverse
Based the PMBOk
Document
Matching using 6 th
corpus
Frequency
the for
pattern the semantic
(TF-IDF)
based annotation
extraction approach tasks.
, the
Wu-Palmer
Tokenization Score [ Wang
splits a et al., 2020]
character sequence up into (words/phrases) called tokens.
Calculates the frequency of occurrence of each term x within document y.
Computes
Chunking
2 the relationships
Identifytask
the that between
segments
concepts and twolabels
and terms by
properties. considering
multi-token the depthThe
sequence. of two
resulted
concepts hierarchy
part-of-speech
synsets is extracted
tagging using
is that Subject-Is
assign for a/is
eachpart of
token - Object
its patterns.
related partof-speech
groupinofwordNet
tokens istaxonomy and the Least Common Subsumer (LCS) depth.
called “chunk.”
[noun (NN), verb (VB), Determiner (DT),etc], using NLTK.Tag.
3 Weight the pertinence of each candidate term.
Stop word removal removes some stop words from the text which have no
4 Extractrelevance
significant synonyms. [adjective ADJ or adverb ADV, DT].
Identify
Stemming
5 the concept
removes word hierarchy.
to the stem (root) of derived words.
ThisGenerate
step3.2
6SpaCy the non-taxonomic
is pipeline
achieved means ofrelations.
byused.
is NLTKit !

12/ 12/01/20 12
The Final output: illustrations

Ontology illustration

13
The Final output: illustrations

Rule-Based recommendation engine


The reasoning process

14
The Final output: illustrations

System illustration

2022/2023 15
16

You might also like