Post on 15-Jul-2015
ENTER 2015 Research Track Slide Number 1
OpeNER: Open Tools to Perform Natural Language Processing on
Accommodation ReviewsAitor García-Pablos, Montse Cuadros, María Teresa Linaza
Vicomtech-IK4, Spainagarciap,mcuadros,mtlinaza@vicomtech.org
http://www.vicomtech.org
ENTER 2015 Research Track Slide Number 2
Summary
• Introduction• The OpeNER project
– Objective– Architecture
• An example• Some results• Conclusions
ENTER 2015 Research Track Slide Number 3
Introduction
• Web 2.0 and Social Networks have changed the way the customer information flows
• These new channels generate high amounts of information related to the following issues– Preferences of potential customers– Requests/complaints from current customers– Feedback from past customers
ENTER 2015 Research Track Slide Number 4
Introduction (2)
• However, it is not feasible to manage this information manually– Too time consuming– Large investment to track all sources from the
Web
• Computers can help processing texts – Detection of mentions of certain entities– Classification of the reviews regarding their
polarity
ENTER 2015 Research Track Slide Number 5
Introduction (3)
• The so-called “opinion mining” and “sentiment analysis” tools offer these type of services
• Main current limitations– Most of them are not free– They are too complex– It is not obvious how to integrate them into a
real service
ENTER 2015 Research Track Slide Number 6
OpeNER project
• OpeNER is a 7th Framework Programme European project which aims at providing a set of Open Source tools to perform text processing tasks– Named Entity Recognition, sentiment analysis, etc.– For six languages– Free and Open Source– Modular and easy to integrate
www.opener-project.eu
ENTER 2015 Research Track Slide Number 7
OpeNER project (2)
• Basic tools to allow end users and/or SMEs building a customized products or services with textual content analysis– Free tool– Easy to integrate to ease building upon it– Open Source to customize the code
www.opener-project.eu
https://github.com/opener-projecthttps://github.com/opener-project
ENTER 2015 Research Track Slide Number 8
OpeNER architecture
KAF (Bosma et al. 2009)
ENTER 2015 Research Track Slide Number 9
A practical example
“I have been at Albergo Acquarello hotel at Lugano and I liked the beautiful decoration. The rooms were very comfortable. On the other hand, the restaurant was really expensive.”
An hypothetic customer review
ENTER 2015 Research Track Slide Number 10
A practical example (2)
Named Entity Recognition, Classification and Linking:
ENTER 2015 Research Track Slide Number 11
A practical example (3)
Sentiment/Polarity detection:
ENTER 2015 Research Track Slide Number 12
A practical example (4)
Opinion detection using:
ENTER 2015 Research Track Slide Number 13
Some evaluation resultsTool Language Precision Recall F-Score Method Dataset
Opinion detector
en 85,52% 58,45% 69,44%CRF + SVM
OpeNER manual hotel annotations
Opinion detector
nl 82,8% 51,77% 63,71%CRF + SVM
OpeNER manual hotel annotations
Opinion detector
de 75,64% 48,88% 59,38%CRF + SVM
OpeNER manual hotel annotations
Opinion detector
es 74,41% 46,55% 57,27%CRF + SVM
OpeNER manual hotel annotations
Opinion detector
it 65,47% 40,39% 49,96%CRF + SVM
OpeNER manual hotel annotations
Opinion detector
fr 70,94% 46,28% 56,02%CRF + SVM
OpeNER manual hotel annotations
ENTER 2015 Research Track Slide Number 14
Tour-pedia
Concept application: Tour-pedia,Developed at the CNR Pisa, within the OpeNER project
www.tour-pedia.org/gui/demo/
ENTER 2015 Research Track Slide Number 16
Conclusions
• Web 2.0 enables a valuable customer communications channels that require technology to be efficiently processed
• There are some tools already in the market and in the academia, but they are usually difficult or expensive to use
• OpeNER provides with some of these technologies, free, open source, and easy to use and integrate
ENTER 2015 Research Track Slide Number 17
Thank you for your attention!Any question?
www.opener-project.euhttp://www.vicomtech.org