Date post: | 15-Apr-2017 |
Category: |
Data & Analytics |
Upload: | aist |
View: | 97 times |
Download: | 0 times |
AIST 08 .04 .16YEKATERINBURG, RUSSIA
NIKOLAY KARPOVEDUARD BABKIN
ALEXANDER DEMIDOVSKIY
NATIONAL RESEARCH UNIVERSITY HIGHER SCHOOL OF ECONOMICS
Evolvable Semantic Platform for Facilitating Knowledge
Exchange
Motivation
A university undoubtedly should be a catalyst for exchanging expertise and professional knowledge in the economic cluster.
A specifically designed combination of automated text processing and ontology-based knowledge engineering may improve quality of information analysis and reduce university’s response time.
We propose to facilitate knowledge exchange by seeking relevant university experts for commenting actual information events expressed in the texts of news.
Personal ontology in InfoPort system
W3C FOAF (Friend of Friend) vocabulary specification
Researcher as a person.
Researcher as a skillful agent.
Researcher as a team member.
3
InfoPort User Interface
a) front page; b) enlarged view of personal time line
Platform implementation
In one hand we have personal ontology which includes skills of university experts
In other hand we have unstructured text of news which are expressed
We analyze semantic in the news and match it with skills of experts
For semantic matching we choose an algorithm (Momtazi and Naumann, 2013) based on a Latent Dirichlet allocation.
It is algorithmically implemented in the newly designed decision support system titled EXPERTIZE.
Matcher using Latent Dirichlet allocation
Zz
zCPdzPCdP )/()/()/( 00
We count a probability for each expert and category c and rank categories according to value.
Approach by (Momtazi and Naumann, 2013)
Interaction of EXPETIZE services with InfoPort platform
InfoPort
EXPETIZE system
InfoPort platform
Store Service
Regularoffline
services
Online services
NativeREST-
Interface
The EXPERTIZE system regularly monitors user profile sources in the Internet, performs document analysis and provide university employees with critical information about relevant events according the specific relevance matching algorithm.
Principle design of the EXPERTIZE system
REST-Interface
Crawler Service
Data Modeler
Data Store
Matcher
Temporal raw data LDA model
InfoPort Store Service
RSS Newsfeed
Online processing
REST-Interface
Offline processing
Web GUI
Graphical user interface of the EXPERTIZE system
Algorithm quality evaluation
We evaluate algorithm proposed by Momtazi and Naumann with our datacollection an queries
Score Experts CategoriesPrecision (10) 0.86 0.72Precision (5) 0.62 0.44Precision (1) 0.17 0.37MAP (10) 0.57 0.49MAP English TREC 2006 0.471 -MAP English TREC 2005 0.248 -
Conclusion
Our EXPERTIZE platform applies topic modeling to online expert recommendation using the university community as the expert pool.
We realize and evaluate an algorithm for matching news with a semantic of two indicators: experts and categories.
As a source of categories and keywords two taxonomies are used together as a machine-readable ontology of scientific areas.
The first use cases of the EXPERTIZE system show their ability to solve the task specified.