Design Principles for Competencebased Recommender Systems

Valerio Bellandi, Paolo Ceravolo, Fulvio Frati, Jonatan Maggesi Universit degli Studi di Milano, Italy Gabriela Waldhart, Isabella Seeber Leopold-Franzens University of Innsbruck, Austria

Collaborative Project FP7- ICT- 2009 - 257886

Outline
ARISTOTELE Project
Introduction on Recommender Systems

Recommendations in Competence Management Systems

Analysis of ARISTOTELE Recommender System Scenario Competence-based Management System Conclusions

IEEE DEST-CEE 2012 18-20 June 2012

ARISTOTELE Project
Modern knowledge-intensive organizations are step by step realizing that shifting the relevance from tangible to intangible assets increase competitiveness
It is important to consider complex environments that integrate models, processes and technologies with organizational aspects in a systemic approach

The FP7 ARISTOTELE research project aims at relating learning to organizational processes, as well as to the innovation process management Three kinds of processes are identified:
organizational processes (marketing & communication, HRM, business) learning processes (group training sessions) social collaboration processes (spontaneous formation of groups)
IEEE DEST-CEE 2012 18-20 June 2012 3

Recommender Systems - 1
Recommender Systems (RSs) are considered critical in ARISTOTELE approach to exploit interactions in collaborative environments and foster the innovation process
Basic RS concepts:
1. 2. 3. 4. Homophily: similarity between sources and recipients Tie strength: intensity of the relationship between the recipient and source Trust: trust relationship between recipient and source Social capital: sources reputation

Four main RS categories:

1. 2. 3. 4. Content-based: RS calculates similarity of related contents Collaborative: RS calculates similarity of user profiles basing on ratings Knowledge-based: RS considers user requirements and domain knowledge Hybrid

ARISTOTELE RS falls in the 4th category, taking care of all RS concepts

IEEE DEST-CEE 2012 18-20 June 2012 4

Recommender Systems (RSs) - 2

ARISTOTELE RS try to mix two important aspects of modern RSs
Diversity, Novelty and Serendipity
Give to users suggestions that are not limited to the closest ones, but also those suggestions that apparently are far from the users interest, but she should also like it Serendipity is the art of making an unsought finding F.P. Adams

Knowledge-sharing and Solidity

The accuracy of recommendations will be improved with a knowledge assessment service, exploiting users implicit and explicit feedbacks on suggestions
IEEE DEST-CEE 2012 18-20 June 2012

Recommendations in CompMS
ARISTOTELE RS will assist the platform Competence Management System (CompMS)
Typical CompMS process:
Identify the task to be completed Perform a gap analysis to identify available and missing competences Use the results to identify the actions to fill the learning gaps

Integration between RS and CompMS is difficult

Example: suggestions on how a user can improve his CV cannot be directly referred to users having similar CVs It requires the definition of specific design principles for the algorithms guiding the recommendation process
IEEE DEST-CEE 2012 18-20 June 2012 6

Competence-based RS Principles
The analysis started from the introduced RS concepts and ARISTOTELE platform user requirements
Four different analysis dimensions, taken from project user scenarios:
Competence Based Management Activity and Task Management Collaboration Knowledge

Result of the analysis will be in terms of preconditions and limitations, to define the boundaries of the ARISTOTELE RS
Preconditions and limitations have been exploited for the design of RS

IEEE DEST-CEE 2012 18-20 June 2012

Competence Based Management - 1

ARISTOTELE will allow the representation and reasoning on related concepts stored in the models
Contextual information, e.g. a person with specific competences working on a specific task, can be reused to express relations

Precondition 1 - Competence Profiles

Competence profiles, and the awareness on which competences are required for organizational tasks, are required Competence profiles need to be accessible and updatable by the platform RS must know to what extent a person is capable of fulfilling the task (proficiency) The knowledge base must be designed around the basic concept of competence

Limitation:
In case competence profiles levels are not available, the RS will not be able to suggest people related information

IEEE DEST-CEE 2012 18-20 June 2012

Competence Based Management - 2

Precondition 2 - Tagging/Annotating
The platform needs to provide tagging or annotation functionalities for artifacts, people, and processes Users need to have the corresponding rights and willingness to annotate available contents Content-based recommendations are efficient only with the availability of suitable tags or keywords

Limitation:
If users cannot tag or annotate contents or resources, it is likely that recommender results will not reach high accuracy

IEEE DEST-CEE 2012 18-20 June 2012

Activity and Task Management - 1

ARISTOTELE supplies a comprehensive workflow management, where people can define, refine, monitor, and assign objects
Workers daily tasks can be improved with the help of accurate resource recommendations

The RS should support community work by recommending appropriate groups that hold necessary expertise or resources
The platform needs to be able to merge information coming from various sources, e.g. social networks, ERPs, CRMs

IEEE DEST-CEE 2012 18-20 June 2012

10

Activity and Task Management - 2

Precondition 3 - Activity/Task Monitoring
RS needs to interface with the platform to extract necessary information about users tasks/activities from enterprise systems
These systems need to provide endpoints to extract necessary information about users tasks/activities

Users need to enrich tasks description with set of metadata that will help RS to provide reliable suggestions

Limitation
If activity/task information are not available for recommendation, ARISTOTELE methodologies related to it might be restricted
IEEE DEST-CEE 2012 18-20 June 2012 11

Collaboration
User scenarios included possible communication channels: instant messaging, private message, discussion boards, shared workspaces, forums, meetings arrangement, workflow management systems,
Collaboration should result in annotations of users contents by resource rating, tagging, or giving feedback or comments on documents Precondition 4) Willingness to change communication practices
It is necessary that the organizational culture, and therefore people, accept and are willing to use ARISTOTELE communication channels It is important that the RS will be able to access collaborative resources and contents and give suggestions based on them
IEEE DEST-CEE 2012 18-20 June 2012 12

Knowledge - 1
ARISTOTELE core is the handling of explicit/codified knowledge, giving the appropriate support of people to exchange implicit knowledge successfully
The platform needs to support activities to update and share organizational knowledge
Knowledge might also be found or stored outside the organizational boundaries It is important to find mechanisms to determine trustable and justifiable knowledge sources The platform needs to reach a critical mass of users and contents to give reliable suggestions
IEEE DEST-CEE 2012 18-20 June 2012 13

Knowledge - 2
Precondition 5) Migration of existing knowledge base
There is a need that the existing knowledge base gets migrated into ARISTOTELE
Ontology matching techniques could achieve that same things from diverse systems will be described as one thing RS must rely on a common metamodel to be able to derive suggestions from different sources

Limitation
If a knowledge base is not available, the platform could have problems in solving the start-up problem and create sufficient positive network effects
IEEE DEST-CEE 2012 18-20 June 2012

14

Knowledge - 3
Precondition 6) Context Information
To derive contextual information ARISTOTELE needs to have interfaces to other enterprise systems (CRM, HRM, ERP,) Information needs to be mapped among the different sources if they deliver the same semantic information (e.g., worker, client, user, etc.)
It is important to have a common metamodel to uniform all the data

Limitation
If ARISTOTELE cannot draw on information stored elsewhere in the organization, content-based recommendation will be highly limited
IEEE DEST-CEE 2012 18-20 June 2012 15

ARISTOTELE Competence-based RS - 1
Stated preconditions and limitations were inputs for the development of the ARISTOTELE RS
It differs from other common RS by taking as central concept the competences and working experiences of all members of an organizations Designed to be triggered by a specific stimulus
Giving suggestions on activities and learning plans correlated to the subject Suggesting a set of alternative objects that, at a first sight, could seem completely unrelated (serendipity)

RS can introduce novel and unexpected knowledge fields in the ordinary business process
IEEE DEST-CEE 2012 18-20 June 2012 16

ARISTOTELE Competence-based RS - 2

IEEE DEST-CEE 2012 18-20 June 2012

17

Conclusions
We investigated important preconditions and limitations that are considered to have impacts on the design and success of a competence-based Recommender System
Preconditions can be directly used as input factors for the RS Preconditions can inform the design of RS from a theoretical point of view with the goal to improve the acceptance of recommender results Currently we are working on:
The implementation of the RS within the ARISTOTELE platform Validation of the approach through independent experiments
Results of the validation presented during the Innovation Adoption Forum
IEEE DEST-CEE 2012 18-20 June 2012 18

Thank you
Any questions?

info@aristotele-ip.eu

ADDITIONAL SLIDES

RS Metamodel CR2S

IEEE DEST-CEE 2012 18-20 June 2012

21