[PDF] SANA: Sentiment analysis on newspapers comments in Algeria

View - Download SANA: Sentiment analysis on newspapers comments in Algeria

Previous PDF

Next PDF

SANA: Sentiment analysis on newspapers comments in Algeria

Hichem Rahab

a,b,? , Abdelhafid Zitouni b , Mahieddine Djoudi c a ICOSI Laboratory, University of Khenchela, Algeria b LIRE Laboratory, University of Constantine 2, Algeria c

TechNE Laboratory, University of Poitiers, France

article info

Article history:

Received 2 February 2019

Revised 27 March 2019

Accepted 24 April 2019

Available online xxxx

Keywords:

Opinion mining

Sentiment analysis

Machine learning

K-nearest neighbors

Naïve Bayes

Support vector machines

Arabic

Comment

abstract

It is very current in today life to seek for tracking the people opinion from their interaction with occurring

events. A very common way to do that is comments in articles published in newspapers web sites dealing

with contemporary events. Sentiment analysis or opinion mining is an emergent field who's the purpose

is finding the behind phenomenon masked in opinionated texts. We are interested in our work by com- ments in Algerian newspaper websites. For this end, two corpora were used; SANA and OCA. SANA corpus is created by collection of comments from three Algerian newspapers, and annotated by two Algerian

Arabic native speakers, while OCA is a freely available corpus for sentiment analysis. For the classification

we adopt Supports vector machines, naïve Bayes and k-nearest neighbors. Obtained results are very promising and show the different effects of stemming in such domain, also k-nearest neighbors gives important improvement comparing to other classifiers unlike similar works where SVM is the most dom- inant. From this study we observe the importance of dedicated resources and methods the newspaper comments sentiment analysis which we look forward in future works.

?2019 Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access

article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Contents

1. Introduction . . ........................................................................................................ 00

2. Background. . . ........................................................................................................ 00

2.1. Matter approach . . . . . . ........................................................................................... 00

2.2. Validation method. . . . . ........................................................................................... 00

2.3. Classifiers. . . . . . . . . . . . ........................................................................................... 00

2.4. Evaluation measures . . . ........................................................................................... 00

3. Related works. ........................................................................................................ 00

4. Proposed approach. . . . . . . . . . . . . . . . ..................................................................................... 00

4.1. Model. . . . . . . . . . . . . . . ........................................................................................... 00

4.2. Annotation. . . . . . . . . . . ........................................................................................... 00

4.3. Processing . . . . . . . . . . . ........................................................................................... 00

4.4. Train and test . . . . . . . . ........................................................................................... 00

4.5. Evaluate . . . . . . . . . . . . . ........................................................................................... 00

4.6. Revise. . . . . . . . . . . . . . . ........................................................................................... 00

5. Experimental study . . . . . . . . . . . . . . . ..................................................................................... 00

1319-1578/?2019 Production and hosting by Elsevier B.V. on behalf of King Saud University.

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Corresponding author at: Laboratoire ICOSI, Faculté des Sciences et de la Technologie, Bloc D, Campus Route Oum El Bouaghi, Université de Khenchela,Khenchela 40000,

Algérie.

E-mail addresses:rahab.hichem@univ-khenchela.dz(H. Rahab),Abdelhafid.zitouni@univ-constantine2.dz(A. Zitouni),mahieddine.djoudi@univ-poitiers.fr(M. Djoudi).

Peer review under responsibility of King Saud University.Production and hosting by Elsevier Journal of King Saud University - Computer and Information Sciences xxx (xxxx) xxx

Contents lists available atScienceDirect

Journal of King Saud University -

Computer and Information Sciences

journal homepage: www.sciencedirect.com

Please cite this article as: H. Rahab, A. Zitouni and M. Djoudi, SANA: Sentiment analysis on newspapers comments in Algeria, Journal of King Saud

University - Computer and Information Sciences,https://doi.org/10.1016/j.jksuci.2019.04.012

5.1. First round. . . . . ................................................................................................. 00

5.2. Second round. . . ................................................................................................. 00

5.3. OCA corpus . . . . ................................................................................................. 00

6. Results discussion . . . . . . . . . . ........................................................................................... 00

7. Conclusion and perspectives . . ........................................................................................... 00

Conflict of interest. . . . . . . . . . ........................................................................................... 00

References ........................................................................................................... 00

1. Introduction

With the development of the web and its offered services, a huge amount of data is generated (Liu, 2012) and additional needs emerge to take benefit from this information thesaurus. Opinion mining from Political, economic and social data, is a new need to make the huge amount of available information in an easily under- stood form to decision makers in dedicated centers. Sentiment analysis vocation is to classify people opinions into specific cate- gories to facilitate understanding the behind phenomenon. A variety of classification approaches are available, some works deal only with positive vs. negatives classes (Rushdi-Saleh et al.,

2011; Atia and Shaalan, 2015; Rahab et al., 2018), others deal with

more important number of classes (Cherif et al., 2015; Ziani et al.,

2013).

A very important amount of useful information is available in the comments of newspapers websites visitors around the world and in different languages. A lot of works in this era deal with Eng- lish, and other European languages, but works treating Arabic lan- guage still in their beginning (Alotaibi and Anderson, 2016). Arabic is a Semitic language spoken by about 300 million of people in 22 Arab countries. And the importance of Arabic is also that it is the language of the holy Quran (Cherif et al., 2015) the book of 1.5 billion Muslim in the world. We can find three forms of Arabic language, Classical Arabic, Modern Standard Arabic, and Dialectal Arabic. Classical Arabic is the original form of the lan- guage preserved from centuries by the Islamic literature and espe- cially the holy Quran. For Modern Standard Arabic, it takes the role of the official language in almost all Arabic administrations. The effective spoken languages in daily conversations are Arabic dia- lects, which are spoken languages without a standardized writing form. They can be classified into: Levantine (spoken in Palestine, Jordan, Syrian and Lebanon) Egyptian (in Egypt and Sudan), Magh- rebi (spoken in the Arab Maghreb) and Iraqi (Jarrar et al., 2017), this later one may be also divided into Iraqi versus Gulf classes (Zaidan and Callison-burch, 2011). In these Dialect families, we will find also sub-families. In the case of the Algerian dialect, the work of (Harrat et al., 2016) classify Algerian dialects in 4 groups: 1) the dialect of Algiers and its out- skirts, 2) the dialect of the east in Annaba and its outskirts, 3) the dialect Oran and the west of Algeria, and 4) the dialect of the

Algerian Sahara.

Even the newspaper content is written in MSA and comments follow generally this style, we find some visitors that use Algerian Dialects words in their comments. For example the Arabic Hu faqat fi Alduwal almutaxalifa 1 (things like this occur only in retarded cir fy Ald?uwal almutaxalifa. Also, we found in several cases the use ofﺩd, instead ofﺫð, which is a characteristic of the Dialect of Algiers the capital of Alge-

ria (Harrat et al., 2016), as the case in the commentﺷﻜﺮﺍﻳﺎﺣﻔﻴﻆﻫﺪﺍﻫﻮ

Almasw

taqidminȂjlȂn yantaqid wa yuTab?il wa yudafi

11an Alz?aman

Al?adi mar wa lakin hunaka rijAl yaSna

1un Almajd bitaHad?iyhim

AlwaqaAA

1(Thank you hafid this is the state of the responsible to

whom is affected a mission, and he fails, so he become critic for critic, and he defends the earlier time but there are men making the glory by confronting the realities). We are interested by comments in the Arabic Algerian online press, in the goal of developing an approach to classify these com- ments into positive and negative classes. The paper is organized as follows. In theSection 2a background of adopted methodology and used parameters are given. In theSec- tion 3, a literature review is presented.Section 4is dedicated to the proposed approach. An experimental study is explained and obtained results are in theSection 5.InSection 6the achieved results are discussed. We finish by conclusion and perspectives to future works.

2. Background

2.1. Matter approach

MATTER is a cyclic approach for natural language texts annota- tion, the approach is based on several iterations to achieve the annotation process (Pustejovsky and Stubbs, 2012). The MATTER approach consists on a cycle of six steps. The model of the phe- nomenon may be revised for further train and test steps (Ide and

Pustejovsky, 2017).

Model: in the first step the studied phenomenon will be modeled. Annotate: an annotation can be seen as a metadata (Matthew and Jessica, 2010). This metadata will be added to our corpus for data classification into predefined classes like positive, neg- ative, neutral, etc. The annotation may be integrated in the doc- ument to annotate, in a manner, that when the document is moved, the metadata still integrated, for example the addition of a distinction word in the file name. It can also take the form of a folder in which the data files are grouped, in this case a file extracted out of this folder will lose this metadata (Matthew and Jessica, 2010).

The annotation can be done at several levels.

oDocument level:the whole document take the same label, such as: positive/negative (Rushdi-Saleh et al., 2011) or subjective/ objective,...etc. oSentence level:in this level each sentence in the document may have an independent tag, an example of this level is the tweet's classification (Brahimi et al., 2016) that the tweet cannot exceed 140 words. oWord level: Also known as Part Of speech tagging POS (Tunga,

2010), where each word is tagged according to its position

in the text (e.g. noun, verb, and pronoun) (Jarrar et al.,

2017).

1 For transliteration we follow in this work the scheme developed byHabash et al. (2007).

2H. Rahab et al./Journal of King Saud University - Computer and Information Sciences xxx (xxxx) xxx

Please cite this article as: H. Rahab, A. Zitouni and M. Djoudi, SANA: Sentiment analysis on newspapers comments in Algeria, Journal of King Saud

University - Computer and Information Sciences,https://doi.org/10.1016/j.jksuci.2019.04.012 We can find several ways to achieve annotation with. Annota- tion by 2-5 persons having some specified skills (Alotaibi and Anderson, 2016)(Pustejovsky and Stubbs, 2012), Crowdsourcing where the annotation is done by an important number of annota- tors without specific skills (Bougrine et al., 2017), or Annotation based on rating systems offered by opinion sites (Rushdi-Saleh et al., 2011). The final version of the annotated data called the gold standard is the corpus to be used in the classification step (Pustejovsky and

Stubbs, 2012).

Train: a part of the data with their true classes is used to train the classifier. Test: the rest of data (which is not used for training) is submit- ted to classifier for test. Evaluate: evaluation metrics are calculated, to measure the annotation and classification performances.quotesdbs_dbs48.pdfusesText_48

[PDF] algeria wikipedia

[PDF] algerie 1 togo 0 2017

[PDF] algerie 1982

[PDF] algerie 1982 almond mache complet

[PDF] algerie 1985

[PDF] algerie 1988

[PDF] algerie 1988 youtube

[PDF] algerie 1990

[PDF] algerie 1992

[PDF] algerie 1992 gia

[PDF] algerie 1993

[PDF] algerie 3

[PDF] algerie 3 streaming

[PDF] algerie 7 tanzanie 0

[PDF] algerie 7 vs 1 ethiopie 25/03/2016