Docear - User Modeling Inc.

advertisement
Utilizing Mind-Maps for
Information Retrieval
and User Modelling
Joeran Beel, Stefan Langer, Marcel Genzmehr, Bela Gipp
www.docear.org
Docear – The Academic Literature Suite
1
Agenda
1.
Introduction to mind-maps
2.
Ideas for utilizing mind-maps beyond
their original purpose
3.
Prototype for mind-map-based user
modeling
www.docear.org
Docear – The Academic Literature Suite
2
Docear Team
www.docear.org
Docear – The Academic Literature Suite
3
Docear (www.docear.org)
www.docear.org
Docear – The Academic Literature Suite
4
1. Introduction to
Mind-Maps
www.docear.org
Docear – The Academic Literature Suite
5
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
6
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
7
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
8
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
9
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
10
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
11
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
12
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
13
Example Mind-Map: Paper Draft
www.docear.org
Docear – The Academic Literature Suite
14
Example: Conference & Journal
Overview
www.docear.org
Docear – The Academic Literature Suite
15
Example: Career Planning
www.docear.org
Docear – The Academic Literature Suite
16
Research Question
 How
to utilize mind-maps beyond their
original purpose?
Original Purpose
Utilized for
Emails
Communication
User modeling &
personalized advertisement
Social Tags
Personal document
organisation
Website indexing
Research Articles
Publishing research results
Impact analysis
Mind-Maps
Information management
???
www.docear.org
Docear – The Academic Literature Suite
17
2. Ideas for Mind-Map based
IR Applications
And An Analysis of the Feasibility
www.docear.org
Docear – The Academic Literature Suite
18
Ideas, Overview
 Search
Engines for Mind-Maps
 Document Indexing / Anchor Text Analysis
 Document Relatedness
 Document Summarization
 Impact Analysis
 Trend Analysis
 Semantic Analysis
 User Modelling
www.docear.org
Docear – The Academic Literature Suite
19
Illustration of some ideas
 Anchor
Text Analysis / Website Indexing
 Document Relatedness / Distance Analysis
 Semantic Analysis
 User Modeling
www.docear.org
Docear – The Academic Literature Suite
20
60%
50%
40%
30%
20%
10%
0%
2010
Public Mind-Maps 67,167
MindMeister
23%
2011
2012
2013
2014
350
300
250
200
150
100
50
-
Thousands
Number of (Public) Mind-Maps & Users
88,624 119,778 195,087 303,084
29%
36%
46%
50%
Mindomo
28%
24%
21%
23%
20%
XMind
37%
37%
35%
23%
16%
Mindjet
0%
0%
1%
4%
10%
Others
12%
10%
7%
5%
4%
 Dozens
of mind-mapping tools
 2 million active mind-mapping users
 5 million new mind-maps every year
 300,000+ public mind-maps
www.docear.org
Docear – The Academic Literature Suite
21
Example of a Mind-Map Gallery
www.docear.org
Docear – The Academic Literature Suite
22
Content Analysis
 Analysis
of 19,379 mind-maps
 Number of nodes per mind-map


Average = a few dozen
Maximum = a few thousand
 63.88%
contain no links,
 Those who contain links, contain typically
only few
 Ideas
requiring links are less feasible
 Text-based ideas are feasible
www.docear.org
Docear – The Academic Literature Suite
23
Users’ Acceptance
 Up
to 61% acceptance for user modeling
and recommendations
 Around 10% acceptance for other ideas
 User
modeling is the most feasible idea
www.docear.org
Docear – The Academic Literature Suite
24
3. User Modeling
Prototype
A Research Paper Recommender System
www.docear.org
Docear – The Academic Literature Suite
25
How to do the user modeling?
www.docear.org
Docear – The Academic Literature Suite
26
User Modeling Approaches
 Stereotype
Approach
Recommend books that we assume to be
relevant for researchers
 Content Based Filtering




Single Node I:
Terms of the last modified node (Recs on each
modification)
Single Node II:
Terms of the last modified node (Recs every few
days)
Single Mind-Map:
All terms of the last modified mind-map
All Mind-Maps:
All terms of all mind-maps
www.docear.org
Docear – The Academic Literature Suite
27
Recommender System
www.docear.org
Docear – The Academic Literature Suite
28
Results I
differences depending on the
approach
 Overall, reasonable CTR, despite the
trivial approaches
CTR
 Strong
10%
5%
0%
6.12%
0.28%
4.99%
All MindMaps
Stereotype
1.17%
Single Node Single Node Current
(I)
(II)
Mind-Map
www.docear.org
6.24%
Docear – The Academic Literature Suite
29
CTR
Results II
3.16%
4.00%
[1;9]
[10;
49]
5.94%
6.51%
7.47%
6.28%
[50;
99]
[100;
499]
[500;
999]
1000+
Node Count
 Strong
differences depending on the
specific parameters (for the „All MindMaps“ approach)
www.docear.org
Docear – The Academic Literature Suite
30
Questions?
Download Docear: http://www.docear.org
More research: http://www.docear.org/docear/our-publications/
www.docear.org
Docear – The Academic Literature Suite
31
Download