0% found this document useful (0 votes)

228 views

Pilucka Biznesa

The paper proposes a multi-agent approach called CLOE for simultaneous enrichment of ontologies in different languages from input text. CLOE uses agents that can learn through being taught a pre-built ontology and translation table or by communicating with neighboring agents to query concepts. It was tested on English, German and Arabic text and evaluated using satisfaction questionnaires and by comparing to standard datasets, finding CLOE outperformed other approaches. The reviews were positive and recommended the paper be accepted with minor revisions.

Uploaded by

Mohamed Aly

Available Formats

Download as DOCX, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

228 views

Pilucka Biznesa

Uploaded by

Mohamed Aly

Available Formats

Download as DOCX, PDF, TXT or read online on Scribd

You are on page 1/ 6

Fwd: Semantics 2018 notification for paper 45

Inbox
Wed, Jun 6, 2018 at 11:49 PM

Said Fathalla

---------- Forwarded message ---------

From: Said Fathalla <[email protected]>
Date: Wed, Jun 6, 2018, 11:48 PM
Subject: Fwd: Semantics 2018 notification for paper 45
To: Said Fathalla <[email protected]>

On Jun 6, 2018, at 11:01 PM, Semantics 2018 <[email protected]> wrote:

Dear Said Fathalla,

We are pleased to inform you that your Cross-Lingual Ontology Enrichment

Based on Multi-Agent Architecture has been accepted as a FULL paper for the
Research & Innovation Track of SEMANTiCS 2018.

Congratulations! Enclosed are the reviews.

Please consider the reviews carefully and modify your paper to address issues
raised by the reviewers.

Your camera-ready version must be submitted until August 6, 2018 (11:59 pm,
Hawaii time). In order to publish the paper in the proceedings at least one
author has to register and pay the regular fee for the conference until
August 6, 2018. Please use the following link for registration:
http://2018.semantics.cc/registration.

Please also note that an accepted paper in this track can be accompanied
with a submission to the Poster & Demo track. We especially recommend
it if your paper is supported by an interesting tool that you developed.
Such an additional submission must cite the accepted paper and
explicitly discuss the added value compared to the conference paper,
where the added value could include a) extended results and experiments
not presented in the conference paper for space reasons, or b) a
demonstration of a supporting prototype implementation.

Such a submission will undergo a shortened review process (check against the
review comments implementation and Poster and Demo track compliance by the
Poster and Demo track chairs) towards being accepted at the Poster & Demos
track. The submission system will remain open until 25 June, 2018, and the
Call for Papers of the Poster and Demo track can be found at:
http://2018.semantics.cc/calls.
Best regards, and looking forwards to seeing you in Vienna for SEMANTiCS
2018,

Anna Fensel and Victor de Boer

SEMANTiCS 2018 Research & Innovation Track Chairs

----------------------- REVIEW 1 ---------------------

PAPER: 45
TITLE: Cross-Lingual Ontology Enrichment Based on Multi-Agent Architecture
AUTHORS: Mohamed Ali, Said Fathalla, Shimaa Ibrahim, Mohamed Kholief and
Yasser Hassan

Reviewer's confidence: 4 ((high))

APPROPRIATENESS: 4 (good)
IMPACT OF IDEAS AND RESULTS: 4 (good)
CLARITY AND QUALITY OF WRITING: 4 (good)
RELATED WORK: 3 (fair)
IMPLEMENTATION AND SOUNDNESS: 3 (fair)
EVALUATION: 3 (fair)
ORIGINALITY / INNOVATIVENESS: 3 (fair)
OVERALL EVALUATION: 2 (accept)
Vocabulary/Ontology: no
Potential Poster or Demo: no
Potential Industry Paper: no
Informative question on DBpedia: no

----------- Paper Summary -----------

The paper presents an approach for cross-lingual ontology enrichment. The
authors propose a multi-agent solution thus enabling cross-ontology and
simultaneous enrichment. With the growing volumes of digital heterogeneous
data, having suitable ontologies in order to be able to understand the data,
is crucial. Ontology learning and enrichment are one solution to this need.
Therefore, the paper presents a current and an important research problem.

----------- Reasons to Accept -----------

+ The evaluation results show a clear advantage in comparison to similar
approaches
+ The combination of slave and master agents is new and brings direct
advantages
+ The work is nicely described, presents solid research and addresses a
current problem that is word investigating

----------- Reasons to Reject -----------

- The approach is not really novel. Both agent-based and cross-language
approaches have been around for a while.
- The related work section is quite short, even though, there is a lot of
existing research in the field.
- There is no link to the implementation/ the repository with the code

----------- General Comments to the Author(s) -----------

The paper is nicely written and easy to follow. The logical structure and
line of argumentation is understandable and well-presented.

Detailed comments:
- Section 3: The way that the different phases are listed and described can
be improved. It would be easier if Figure 1 is moved in the beginning of
section 3 and each phase is highlighted (eg. with bold)
- Section 3.3.1 these are actually not hypothesis, they are assumptions or
fundamentals that you take as given. Hypothesis is not the correct way to
call these.
- It would be interesting to explore if the approach can be improved if
instead of using google translate, the concepts are translated based on
language entries in dbpedia. There are quite a few cross-lingual approaches
that have shown to work better than using google translate
- Section 4. "Finally, experts edit and validate them before adding them to
the ontology” Is this before or after the precision and recall measurements
in the evaluation section? If you evaluate after a human has corrected the
concepts, then it makes no sense to do precision and recall. If the precision
and recall are as high as given in the evaluation, why do you need a human
expert. This part is a bit confusing.
- Section 5. For the evaluation it is important to say what text you
consider, size of corpus, average length, etc. The link to the data corpus
was not working on 24.05
- Algorithm 1 and especially Algorithm 2 do not add much to that paper. Alg.
2 is described in the text…
- The four state-of-the-art approaches from the literature that were used for
the evaluation, did they use exactly the same dataset? You need to make clear
what evaluation parameters were the same and what were different. Otherwise
the evaluation results would not make much sense in terms of comparison.

----------------------- REVIEW 2 ---------------------

PAPER: 45
TITLE: Cross-Lingual Ontology Enrichment Based on Multi-Agent Architecture
AUTHORS: Mohamed Ali, Said Fathalla, Shimaa Ibrahim, Mohamed Kholief and
Yasser Hassan

Reviewer's confidence: 4 ((high))

APPROPRIATENESS: 5 (excellent)
IMPACT OF IDEAS AND RESULTS: 5 (excellent)
CLARITY AND QUALITY OF WRITING: 5 (excellent)
RELATED WORK: 4 (good)
IMPLEMENTATION AND SOUNDNESS: 4 (good)
EVALUATION: 4 (good)
ORIGINALITY / INNOVATIVENESS: 5 (excellent)
OVERALL EVALUATION: 2 (accept)
Vocabulary/Ontology: no
Potential Poster or Demo: no
Potential Industry Paper: no
Informative question on DBpedia: no

----------- Paper Summary -----------

The paper presents a novel agent-based approach (CLOE) for simultaneous
enrichment of several ontologies (in different languages from the language of
the input text).
The system is based on a set of agents that can learn using two ways: 1)
users
can teach them by supplying a pre-built ontology and its translation table,
and 2) an agent can learn a new experience through its communications with
its neighbors by querying another agent for a certain concept.
The system is tested with a multilingual text, containing English, German and
Arabic. The performance of the system is evaluated following two strategies:
satisfaction questionnaire evaluation and Gold standard-based evaluation
(standard datasets used).

----------- Reasons to Accept -----------

The paper is well-written and easy to follow. An evaluation of the system is
proposed and the evaluation results are promising (CLOE outperforms all other
systems in terms of lexical precision and lexical recall, and consequently F-
measure).
The authors:
- introduced a novel approach for simultaneous ontology enrichment,
- demonstrated that agents could learn from each other, using a predefined
communication scheme (based on already-learned concepts).

----------- Reasons to Reject -----------

----------- General Comments to the Author(s) -----------

Paper can be further improved by
- adding a paragraph on motivation - how important is the approach for Arabic
resources on the Web in general and understanding Arabic texts by english-
speaking community
- explaining the selection of languages in the experiments - English, Arabic
and German
- adding a paragraph on problems of Pre-processing and Concept Extraction
from Arabic texts.Are the problems the same as in other languages ?
- discussing the reliability of Google API for e.g. Arabic texts

----------------------- REVIEW 3 ---------------------

PAPER: 45
TITLE: Cross-Lingual Ontology Enrichment Based on Multi-Agent Architecture
AUTHORS: Mohamed Ali, Said Fathalla, Shimaa Ibrahim, Mohamed Kholief and
Yasser Hassan

Reviewer's confidence: 4 ((high))

APPROPRIATENESS: 4 (good)
IMPACT OF IDEAS AND RESULTS: 3 (fair)
CLARITY AND QUALITY OF WRITING: 4 (good)
RELATED WORK: 4 (good)
IMPLEMENTATION AND SOUNDNESS: 4 (good)
EVALUATION: 4 (good)
ORIGINALITY / INNOVATIVENESS: 4 (good)
OVERALL EVALUATION: 1 (weak accept)
Vocabulary/Ontology: no
Potential Poster or Demo: no
Potential Industry Paper: no
Informative question on DBpedia: no

----------- Paper Summary -----------

The paper addresses ontology enrichment cross-linguistically using a multi-
agent architecture. The agents interact with language-specific ontologies and
text resources, in order to help automate ontology learning from one language
to another. Candidate texts in a particular language are preprocessed using
standard NLP piping technology (from language identification to tokenization
to part-of-speech tagging) and then used as starting resources by the agents.
English is used as an intermediate language between, e.g., Arabic and German.
A given set of language agents will communicate with other-language agents to
see if their text exerpts can be translated into the other language, and if
so, the latter agent responds with a semantic concept description vector back
to the former agent. Each language agent maintains a translation table of
concepts it knows about and a mapping from the concept to the concept in the
associated ontology. Potential new concepts and relations for the !
given ontology are then reviewed and validated by a human expert. Over time,
multiple language-specific ontologies could be enriched using this
architecture.

----------- Reasons to Accept -----------

1) The paper presents an interesting extension to ontology enrichment using
multi-lingual ontologies.
2) The paper is well-written and relatively well-structured.
3) The primary case study and evaluation seems sound, is well-described, and
the evaluation indicates better results, as judged against some previous
studies.

----------- Reasons to Reject -----------

1) There are some minimal English errors that could be corrected.
E.g., p. 3, "-Language Identification" => "Language Identification".
p. 5, "the master agent reply" => "the master agent replies".
p. 6, "which lead to guide and refine the learning process." => "which are
used to guide and refine the learning process."
Etc.
2) Figures and Tables could be moved closer to their citations in the text.
This is largely an issue in space management.
3) The research is a modest extension of existing methods.

----------- General Comments to the Author(s) -----------

I would give this paper a "weak accept". It is interesting and a useful, if
non-revolutionary extension to existing methods. There are some (minimal)
English and presentation errors. However, the paper is generally well-written
and well-presented, with good textual and graphic descriptions of the
important contributions. Finally, a sound evaluation is given, that seems to
indicate value in the approach, given other reported evaluations in the
literature.

Quick Reply
To: Said Fathalla
<[email protected]>
Include
quoted
text
with
reply

« Back to Inbox

1 of about 91 Older ›
Get Gmail on your mobile phone at http://mail.google.com using your phone's web
browser. Learn more
You are currently using 2307 MB (15%) of your 15360 MB
Last account activity: 0 minutes ago at this IP (156.196.216.32). Details
Gmail view: standard | basic HTML Learn more
Terms - Privacy - Gmail Blog - Google Home