International Journal on Natural Language Computing (IJNLC) Vol.12, No.

4, August 2023

HIGH ACCURACY LOCATION INFORMATION

EXTRACTION FROM SOCIAL NETWORK TEXTS
USING NATURAL LANGUAGE PROCESSING
Lossan Bonde1 and Severin Dembele2
1
Department of Applied Sciences, Adventist University of Africa, Nairobi, Kenya
2
Laboratoire LAMDI, Universite Nazi Boni, Bobo-Dioulasso, Burkina Faso

ABSTRACT
Terrorism has become a worldwide plague with severe consequences for the development of nations.
Besides killing innocent people daily and preventing educational activities from taking place, terrorism is
also hindering economic growth. Machine Learning (ML) and Natural Language Processing (NLP) can
contribute to fighting terrorism by predicting in real-time future terrorist attacks if accurate data is
available. This paper is part of a research project that uses text from social networks to extract necessary
information to build an adequate dataset for terrorist attack prediction. We collected a set of 3000 social
network texts about terrorism in Burkina Faso and used a subset to experiment with existing NLP
solutions. The experiment reveals that existing solutions have poor accuracy for location recognition,
which our solution resolves. We will extend the solution to extract dates and action information to achieve
the project's goal.

KEYWORDS
Dataset for Terrorist Attacks, Social Network Texts, Information Extraction, Named Entity Recognition

1. INTRODUCTION
Over the past decades, the world has faced terrorist threats that have shaken the foundations of
national security and global stability. Among these attacks, the deadly September 11, 2001, attack
on the Twin Towers of the World Trade Center in New York remains ingrained in the memory of
everyone due to the scale of the tragedy and its global impact. In response to this situation,
governments worldwide have taken measures to strengthen security and fight terrorism,
mobilising all available resources, including advanced technology. Scientific researchers have
played a vital role in this context, especially in computer science. They realised that machine
learning could help detect terrorist attacks by analysing data transmitted over the internet [1]. It is
undeniable that terrorists are increasingly using social networks, forums, and instant messaging to
communicate, plan attacks, and recruit new members [2]. This internet use is not reserved for
terrorists, as the general public also uses the web to exchange information about terrorist
activities [3]. Analysis of this data could help detect suspicious activities and anticipate potential
attacks.

However, analysing these textual data presents a major challenge due to their heterogeneity and
lack of integration. Raw textual data cannot be directly used to train machine learning models;
many transformations are required to identify and extract useful information and then put this
information in formats and structures fit for machine learning. This research project proposes a
system that extracts relevant information from the internet and social network texts and organises

DOI: 10.5121/ijnlc.2023.12401 1
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
it into structured data for machine learning. This goal cannot be achieved in a single paper; we
have divided the project into three phases. The first phase, the object of this paper, is to
specifically address the question of extracting location information with high accuracy.
Subsequent phases will address the extraction of other types of information, with the last phase
dealing with the automatic collection of texts from the internet and social network sources. In the
project's current phase, we have developed a highly accurate (98% of accuracy) location
recognition system, outperforming all the selected existing solutions with which it has been
compared.

This work contributes towards constructing a real-time system for detecting terrorist attacks,
using supervised machine learning. The final product of the whole project will be able to analyse
the necessary data from various sources, including a mobile or web platform accessible to the
public, as well as the internet in general and social networks in particular. The collected
information will then be used to train machine learning models to detect suspicious activities and
anticipate potential attacks.

Though the proposed solution is designed in the context of Burkina Faso, the system can be
easily adapted for any country by changing the location names database and the language if need
be.

The remaining part of the paper is organised into four sections. Section 2 explores the related
works that enable us to identify existing gaps and provide foundations upon which we have built.
Section 3 introduces the methodology of the research and gives details on the work that was
completed in this work. Then in Section 4, the results of this work are presented and compared to
others. Finally, the last section concludes the paper and provides directions for future related
research works.

2. RELATED WORKS
Social networks have become a source of important and huge amounts of information that, if
adequately mined, can be useful to valuable applications that can help monitor and control
various events, processes, and natural phenomena. The information from social networks is made
of text which is unstructured by nature, and applications often will need to extract from the text
structured data. Virmani et al. in [4, pp. 626, 627] have identified six challenges NLP
applications face in extracting information from social network text. Out of the six challenges,
one is of great interest to this study: information about entities. For the specific work of this
paper, Named Entity Recognition (NER) is the focal technology considered.

Since the release of ChatGPT in November 2022, it has been used in several domains, including
NER applications. This research considers it worth exploring the ChatGPT capabilities in relation
to the problem at hand. Consequently, we organised the literature review of this paper into two
sections. The first explores NER approaches not based on ChatGPT, and the second section
revolves around solutions based on ChatGPT. Each section summarises how named entity
recognition has been addressed and identifies possible gaps.

2.1. Non-ChatGPT Approaches to Named Entity Recognition

Information extraction is one of the successful applications of Natural Language Processing

(NLP). According to Khurana et al., "extracting entities such as names, places, events, dates,
times, and prices is a powerful way of summarising the information relevant to a user's needs"
[4]. Named Entity Recognition is a well-known approach to performing information extraction of

2
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
that nature. In [5], Pinto et al. made a performance comparison between various NER solutions
and established that, in general, NLP solutions tend to lose performance when applied on social
network texts. Their study concluded that OpenNLP was the best tool for formal texts like
newspaper and web pages, but TwitterNLP was identified as the best solution for social media
text.

[6] introduced an interesting study which built a "Thesaurus-based Named Entity Recognition
System for detecting spatio-temporal crime events in Spanish language from Twitter" system,
which is specific to the Spanish language. In exploring the various studies conducted on NER
applications, we observe that the solutions are either language-specific ([7], [8], [9]) or problem-
specific ([10], [3], [11]).

2.2. ChatGPT for Named Entity Recognition

Since its release in November 2022, ChatGPT has been applied in many fields, including NER. A
Google Scholar search with the key phrase "ChatGPT for Named Entity Recognition" on June 09,
2023, returns a list of 1290 entries. It is, therefore, apparent that ChatGPT has already been
explored for NER systems.

We have also observed that ChatGPT is used for NER implementations in context-specific
applications. Below are some of these applications:

 NER in clinical studies [12], [13]

 NER in historical documents [14]
 NER in financial text analysis [15]
 NER in the military sector [16]
 NER in the legal sector [17]
 And NER in many other areas of applications [18]

Following the trend observed in the literature review and the result of the experiment conducted,
we concluded that we needed to build a specific solution for the problem of extracting location
information in the context of social network texts on terrorism for the specific case of Burkina
Faso.

3. METHODOLOGY
This research follows a three-step process to produce the desired outcome: data collection,
literature review and experiment and, finally, the design and implementation of a new solution.

3.1. Data Collection

As stated above, the research project's final aim is to extract relevant information from social
network text and structure it into an adequate format for machine learning algorithms to learn and
predict terrorist attacks. In the project's first phase, the focus is on extracting location information
from the social network texts. Subsequent phases will address the extraction of other required
information. Figure 1 portrays an example of a social network text such as it is acquired. The text
is in French, the language used in the social network texts involved in the study. A set of 3000
social network texts of this nature have been collected.

3
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023

Figure 1. Example of social network text

The text in Figure 1 is a message that describes an incident that occurred on 6 August 2021,
where some terrorists stopped a truck going from Tanwalbougou to Ougarou and forced the
driver to take them on board. Fortunately, the driver was able to escape and raise the alert. The
processing of such text should produce a list of all the locations (region, province, county, or
city/village) in the text.

3.2. Literature Review and Experiment

With a clear knowledge of the data involved and the desired output, we did a thorough literature
review to identify existing information extraction techniques. Various NER solutions were
explored and, among them, ChatGPT [19], Stanford CoreNLP [20], and Spacy [21] were selected
for further consideration. We conducted an experiment to assess the efficiency of these solutions.
We randomly selected 20 texts out of the set of 3000 and tested each solution to determine how
accurately each of them could identify location information in the texts. The details of the results
are presented in Section 4. In general, the detection rates were low with the best score being 54
%. Hence, the experiment revealed that none of these existing solutions can be directly used to
resolve our challenge. It was therefore necessary to design and implement a better solution,
which constitutes the last step of this process.

3.3. Design and Implementation of a New Solution

A more accurate solution is required since the existing NLP solutions offer a poor rate of location
name recognition. The approach has been to use the best of existing solutions and make some
extensions that improve the recognition rate in the specific context of social network texts. To
that respect, we selected Stanford CoreNLP to serve as the base NER solution. The pipeline of
the proposed solution is shown in Figure 2, where two extensions have been added to the normal
Stanford CoreNLP NER pipeline to consider some specific issues related to social network texts.

4
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023

Figure 2. NER Pipeline of the Proposed Solution

3.3.1. CoreNLP NER Pipeline

The proposed solution uses the CoreNLP NER pipeline to split the text into tokens which are
then processed by the "Location Recognizer" to identify with high accuracy location named
entities. The normal CoreNLP pipeline presented in Figure 3 has been simplified because our
approach only needs tokens; we stopped the pipeline just after the tokenisation step. This pipeline
reduction speeds up the process.

Figure 3. Stanford CoreNLP Pipeline (source:[22])

3.3.2. Extensions

Two extensions have been added to the normal CoreNLP pipeline: the pre-processing and the
location recognizer, as shown in Figure 2 and described in the subsections below.

3.3.2.1. Pre-processing

In the pre-processing extension, two types of transformations are done on the initial text:
removing special symbols and normalising multi-word location names.

 Removal of special symbols: Social network texts, especially tweets, contain specific
symbols and characters such as the hashtag (#) and the at symbol (@). The presence of those
symbols can hinder the recognition of a location. The pre-processing extension removes
those special symbols from the initial text before the CoreNLP process starts.
 Normalisation of multi-word location names: Some location names like "Bobo Dioulasso"
and "Boucle du Mouhoun" are composed of multiple words, each of which will be

5
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
recognised separately by the CoreNLP pipeline. During the pre-processing phase, we identify
such names, and their corresponding words are combined using hyphens to make them one
token in the transformed text. For example, the name "Boucle du Mouhoun" will change to
"Boucle-du-Mouhoun."

After the two types of transformations, the output text called "clean text" (in the pipeline
presented in Figure 2) is ready for the NER operations.

3.3.2.2. Location Recognizer

The location recognizer is the heart of the proposed solution; it takes each token from the
CoreNLP pipeline, looks up the database of location names, and determines if the token (the
token's text) matches a known location name. If so, that name is returned as output; otherwise,
the token does not correspond to a location name.

The matching system is based on the Generalized Levenshtein Distance (GLD), which measures
the difference between two strings. As raised by [23], social network texts are often written with
spelling errors and non-standard abbreviations. Any good NLP tool dealing with this type of text
must use a string similarity algorithm to handle the misspelling occurrences. Among the multiple
existing algorithms to solve this problem, Yujian and Bo [24] recommend the GLD as a better
solution, which we also adopted in this research. Using the GLD algorithm and the database of
the location names, the matching algorithm is depicted in Figure 4.

Figure 4. Matching Algorithm

In the flow diagram of Figure 4, the variables and functions are described as follows:

 Variables:
o word: corresponds to the current token that the Stanford CoreNLP returns.
o dic_word: is the last word read from the database of names.

6
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
o d: is the distance (based on the GLD) between word and dic_word.
o current: represents the name from the database that is the closest to word.
o min: represents the minimum distance found so far between word and the names in
the database.
 Functions:
o getNameFromDB ( ): this function reads the next name in the database.
o GLD (str1, str2): represents the GLD which determines how close or similar the two
strings passed in arguments are.

In summary, this paper has proposed a new solution to the location named entities recognition
problem in the context of a country where all names of regions, provinces, counties, cities, and
villages are recorded in a database, and the GLD is used to avoid sensitivity to names that are
misspelt. In the next section, we shall compare the solution's performance against previous ones.

4. RESULTS AND DISCUSSIONS

As indicated in the methodology section, we have performed an experiment to test some existing
solutions on a set of 20 randomly selected texts out of the 3000 collected. The same experiment
has also been done on the proposed solution and, in this section, we present and discuss the
results of the experiment.

4.1. Results

The experiment was carried out on ChatGPT, Spacy, and Stanford CoreNLP. The results are
summarised in Table 1, where the column Expected shows the exact list of location names
contained in the text and the number of these names. Then, for each of the tools, the column
Detected lists the locations the tool was able to recognise, and the column Rate gives the number
of the locations recognised over the number of locations expected to be recognised.

Table 1. Experiment Results

#Text Expected ChatGPT Spacy Stanford CoreNLP

Values Number Detected Rate Detected Rate Detected Rate

1 Komandjari 2 Gayérie 1/2 Komandjari 2/2 Komandjari 2/2
Gayérie Gayérie Gayérie

2 Oudalan 3 Zigberi 2/3 0/3 Oudalan 3/3

Zigberi Markoye Zigberi
Markoye Markoye

3 Seno 3 0/3 Seno 1/3 Seno 3/3

Bilakoka Bilakoka
Gorgadji Gorgadji

4 Oudalan 2 Oudalan 2/2 Gorom 1/2 Oudalan 2/2

Gorom Gorom Gorom

5 Poni 2 La grande 0/2 Poni 2/2 Poni 2/2

Djigouè mosque de Djigouè Djigouè
Djigouè

7
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
#Text Expected ChatGPT Spacy Stanford CoreNLP

Values Number Detected Rate Detected Rate Detected Rate

6 Oudalan 2 2 1/2 0/2 Oudalan 2/2
Deou Deou

7 Soum 2 kelbo 1/2 kelbo 1/2 kelbo 1/2

kelbo

8 Tuy 2 Bereba 1/2 0/2 0/2

Bereba

9 Loroum 3 Bouna (12 km 0/3 Loroum 2/3 Loroum 2/3

Bouna de Titao) Titao Bouna
Titao

10 Bam 2 Bam 2/2 Bourzanga 1/2 0/2

Bourzanga Bourzanga

11 Toboulé 4 Les villages de 0/4 Toboulé 3/4 Toboulé 4/4

Damba Toboulé, Damba Damba
Soboulé Damba et Nassoumbou Soboulé
Nassoumbou Soboulé Nassoumbou
(commune de
Nassoumbou)

12 Tapoa 2 Partiaga 1/2 0/2 0/2

Partiaga

13 Bam 4 0/4 Komsilga 1/4 0/4

Komsilga
Minima
Zimtenga

14 Tapoa 3 Boungou 2/3 0/3 0/3

Boungou Nadiabondi
Nadiabondi

15 Banwa 2 Solenzo 1/2 0/2 0/2

Solenzo

16 Tanwalbougou 2 Tanwalbougou 2/2 0/2 Tanwalbougou 1/2

Ougarou Ougarou

17 Kossi 4 Bourasso 1/4 0/4 0/4

Bourasso Axe
Dedougou Dedougou
Nouna Nouna

18 Tapoa 2 marché de 0/2 Tapoa 1/2 Tapoa 1/2

Sambalgou Sambalgou

19 Gourma 2 Nagré 1/2 Gourma 1/2 Gourma 1/2

Nagré

20 Kénédougou 2 0/2 0/2 Kénédougou 2/2

N_Dorola N_Dorola

8
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
#Text Expected ChatGPT Spacy Stanford CoreNLP

Values Number Detected Rate Detected Rate Detected Rate

Average Rate 18/50 Average 16/50 Average Rate 27/50
Rate

As seen from Table 1, the best of the three tools is the Stanford CoreNLP with the average
recognition rate of 27/50 (accuracy of 54%). However, this accuracy is low for the target type of
application. The proposed solution is based on the best of the three, the Stanford CoreNLP, to
which extensions have been made, as presented in the previous section. Table 2 shows the results
of the experiment of the proposed solution over the same set of twenty texts.

Table 2. Results of the Experiment on the New Solution

# Text Expected Detected Rate

1 Komandjari Komandjari 2/2
Gayérie Gayérie
2 Oudalan Oudalan 3/3
Zigberi Zigberi
Markoye Markoye
3 Seno Seno 2/2
Bilakoka Bilakoka
Gorgadji Gorgadji
4 Oudalan Oudalan 2/2
Gorom Gorom
5 Poni Poni 2/2
Djigouè Djigouè
6 Oudalan Oudalan 2/2
Deou Deou
7 Soum Soum 2/2
kelbo kelbo
8 Tuy Tuy 2/2
Bereba Bereba
9 Loroum Loroum 3/3
Bouna Bouna
Titao Titao
10 Bam Bam 2/2
Bourzanga Bourzanga
11 Toboulé Toboulé 4/4
Damba Damba
Soboulé Soboulé
Nassoumbou Nassoumbou
12 Tapoa Tapoa 2/2
Partiaga Partiaga
13 Bam Bam 4/4
Komsilga Komsilga
Minima Minima
Zimtenga Zimtenga
14 Tapoa Tapoa 3/3
Boungou Boungou
Nadiabondi Nadiabondi

9
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
# Text Expected Detected Rate
15 Banwa Banwa 2/2
Solenzo Solenzo
16 Tanwalbougou Tanwalbougou 2/2
Ougarou Ougarou
17 Kossi Kossi 4/4
Bourasso Bourasso
Dedougou Dedougou
Nouna Nouna
18 Tapoa Tapoa 2/2
Sambalgou Sambalgou
19 Gourma Gourma 2/2
Nagré Nagré
20 Kénédougou Kénédougou 1/2
N_Dorola
Average Rate 49/50

The new solution has a recognition rate of 49/50 (accuracy of 98%).

4.2. Discussions

The accuracy of ChatGPT, Spacy, Stanford CoreNLP, and the proposed solution on the test set is
given in Figure 5. In terms of accuracy, the proposed solution is outstanding, giving more
confidence to users who want to extract information from social network texts.

Figure 5. Accuracy of the Solutions

The low performance of NER tools compared to our proposed solution can be explained by two
reasons. Firstly, the French language is known to have less publicly available labelled data that
participate in the training of the NER tools [25]. Secondly, the social network texts are often
poorly formulated, with grammatical and syntax errors which make the context difficult to
understand by the tools and thus failing to recognise location entities. Besides the accuracy, we
have also compared the speed (execution time) of the new solution to the others. The execution
time on a Core i5 CPU @2.3 GHZ, 12 GB RAM computer was respectively 55 seconds, 3
seconds, 59 seconds, and 293 seconds. Despite this higher execution time of the proposed
solution, we still recommend it because the gain in accuracy outweighs the speed deficit.

10
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023

5. CONCLUSION
The objective of this first phase of the research project, which was to build a high-accuracy
location name recognition system, has been achieved. The proposed solution has an accuracy of
98%, which no other tool, according to our knowledge, has been able to reach. The new solution
is both an extension and a simplification of the Stanford CoreNLP: a simplification because we
have reduced the pipeline to the tokenisation phase and an extension because we have introduced
the pre-processing and location recognizer steps. It is also important to note that the gain in
accuracy did result in significant overhead in execution time. However, for most applications,
execution time is not a crucial factor.

While the result of this first phase is outstanding, the proposed solution is of little interest if the
project's subsequent phases are not accomplished. Other information to extract from the internet
and social network texts include dates and terrorist actions. In the project's next phase, we will
address the extraction of these types of information to provide a complete solution.

ACKNOWLEDGEMENTS

This research has not received any specific support to be acknowledged.

REFERENCES

[1] M. DeRosa, Data Mining and Data Analysis for Counterterrorism, Washington D. C: CSIS Press,
2004. Available: https://cdt.org/wp-content/uploads/security/usapatriot/20040300csis.pdf
[2] S. Ressler, “Social Network Analysis as an Approach to Combat Terrorism : Past , Present , and
Future Research,” New York, vol. 2, no. 2, pp. 1–10, 1973, Accessed: Jun. 07, 2023. [Online].
Available: https://www.academia.edu/download/35997070/2.2.8_1.pdf
[3] N. Chetty, S. Alathur, “Hate speech review in the context of online social networks,” Aggression
and Violent Behaviour, vol. 40, pp. 108-118, 2018, doi: 10.1016/j.avb.2018.05.003.
[4] D. Khurana, A. Koli, K. Khatter, and S. Singh, “Natural language processing: state of the art,
current trends and challenges,” Multimed. Tools Appl., vol. 82, no. 3, pp. 3713–3744, 2023, doi:
10.1007/s11042-022-13428-4.
[5] A. Pinto, H. G. Oliveira, and A. O. Alves, “Comparing the Performance of Different NLP Toolkits
in Formal and Social Media Text,” DROPS-IDN/6008, vol. 51, no. 3, pp. 31–316, Jun. 2016, doi:
10.4230/OASICS.SLATE.2016.3.
[6] M. Sotomayor and F. Veloz, “Thesaurus-based named entity recognition system for detecting
spatio-temporal crime events in Spanish language from Twitter,” in 2017 IEEE 2nd Ecuador
Technical Chapters Meeting, ETCM 2017, Jan. 2018, vol. 2017-Janua, pp. 1–5. doi:
10.1109/ETCM.2017.8247537.
[7] W. Khan, A. Daud, K. Shahzad, T. Amjad, A. Banjar, and H. Fasihuddin, “Urdu Named Entity
Recognition Using Conditional Random Fields,” Appl. Sci., vol. 12, no. 13, p. 6391, Jun. 2022, doi:
10.3390/app12136391.
[8] A. Purwarianti, A. Andhika, A. F. Wicaksono, I. Afif, and F. Ferdian, “InaNLP: Indonesia natural
language processing toolkit, case study: Complaint tweet classification,” 4th IGNITE Conf. 2016
Int. Conf. Adv. Informatics Concepts, Theory Appl. ICAICTA 2016, Dec. 2016, doi:
10.1109/ICAICTA.2016.7803103.
[9] A. S. Wibawa and A. Purwarianti, “Indonesian Named-entity Recognition for 15 Classes Using
Ensemble Supervised Learning,” Procedia Comput. Sci., vol. 81, pp. 221–228, 2016, doi:
10.1016/J.PROCS.2016.04.053.
[10] D. Dandeniya, “An Automatic e-news Article Content Extraction and Classification,” Jan. 2019, pp.
196–202. doi: 10.1109/icter.2018.8615480.
[11] P. H. Luz de Araujo, T. E. de Campos, R. R. R. de Oliveira, M. Stauffer, S. Couto, and P. Bermejo,
“LeNER-Br: A Dataset for Named Entity Recognition in Brazilian Legal Text,” in Lecture Notes in
Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in

11
 International Journal on Natural Language Computing (IJNLC) Vol.12, No.4, August 2023
Bioinformatics), 2018, vol. 11122 LNAI, pp. 313–323. doi: 10.1007/978-3-319-99722-3_32.
[12] Y. Hu et al., “Zero-shot Clinical Entity Recognition using ChatGPT,” Mar. 2023, Accessed: Jun. 10,
2023. [Online]. Available: https://arxiv.org/abs/2303.16416v2
[13] R. Tang, X. Han, X. Jiang, and X. Hu, “Does Synthetic Data Generation of LLMs Help Clinical
Text Mining?,” Mar. 2023, Accessed: Jun. 10, 2023. [Online]. Available:
https://arxiv.org/abs/2303.04360v2
[14] C.-E. González-Gallardo, E. Boros, N. Girdhar, A. Hamdi, J. G. Moreno, and A. Doucet, “Yes but..
Can ChatGPT Identify Entities in Historical Documents?”, Mar. 2023, Accessed: Jun. 10, 2023.
[Online]. Available: https://arxiv.org/abs/2303.17322v1
[15] X. Li, X. Zhu, Z. Ma, X. Liu, and S. Shah, “Are ChatGPT and GPT-4 General-Purpose Solvers for
Financial Text Analytics? An Examination on Several Typical Tasks,” May 2023, Accessed: Jun.
10, 2023. [Online]. Available: https://arxiv.org/abs/2305.05862v1
[16] S. Biswas, “Prospective Role of Chat GPT in the Military: According to ChatGPT,” Qeios, Feb.
2023, doi: 10.32388/8WYYOD.
[17] K. Iu, V. W.-A. at SSRN, and undefined 2023, “ChatGPT by OpenAI: The End of Litigation
Lawyers?,” papers.ssrn.com, Accessed: Jun. 10, 2023. [Online]. Available:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4339839
[18] D. Kalla, N. Carolina, N. Smith, and D. Candidate, “Study and Analysis of Chat GPT and its
Impact on Different Fields of Study,” Int. J. Innov. Sci. Res. Technol., vol. 8, no. 3, pp. 827–833,
Mar. 2023, Accessed: Jul. 16, 2023. [Online]. Available: https://papers.ssrn.com/abstract=4402499
[19] M. Abdullah, A. Madain, and Y. Jararweh, “ChatGPT: Fundamentals, Applications and Social
Impacts,” pp. 1–8, Mar. 2023, doi: 10.1109/SNAMS58071.2022.10062688.
[20] C. D. Manning, M. Surdeanu, J. Bauer, J. Finkel, S. J. Bethard, and D. McClosky, “The stanford
CoreNLP natural language processing toolkit,” Proc. Annu. Meet. Assoc. Comput. Linguist., vol.
2014-June, pp. 55–60, 2014, doi: 10.3115/V1/P14-5010.
[21] Y. Vasiliev, Natural language processing with Python and spaCy: A practical introduction. 2020.
Accessed: Jun. 10, 2023. [Online]. Available:
https://books.google.com/books?hl=en&lr=&id=lVv6DwAAQBAJ&oi=fnd&pg=PR15&dq=related
:8-gnxC3TQusJ:scholar.google.com/&ots=6XOoAQRRFD&sig=v24g_-
byTZwR7qfTv_SAgqQebtY
[22] CoreNLP, “Overview - CoreNLP.” 2020. Accessed: Jun. 10, 2023. [Online]. Available:
https://stanfordnlp.github.io/CoreNLP/index.html#download
[23] C. Virmani, A. Pillai, and D. Juneja, “Extracting Information from Social Network using NLP,” Int.
J. Comput. Intell. Res., vol. 13, no. 4, pp. 621–630, 2017, Accessed: May 16, 2023. [Online].
Available: https://www.ripublication.com/ijcir17/ijcirv13n4_15.pdf
[24] L. Yujian and L. Bo, “A normalized Levenshtein distance metric,” IEEE Trans. Pattern Anal.
Mach. Intell., vol. 29, no. 6, pp. 1091–1095, Jun. 2007, doi: 10.1109/TPAMI.2007.1078.
[25] A. Choudhry et al., “Transformer-Based Named Entity Recognition for French Using Adversarial
Adaptation to Similar Domain Corpora,” ArXiv, 2022, doi: 10.48550/ARXIV.2212.03692.

AUTHORS
Lossan Bonde is a PhD holder since 2006 from the University of Science and
Technologies of Lille, France. He is currently assistant professor of computer science at
the Adventist University of Africa, Nairobi, Kenya. His research interests are in Artificial
Intelligence and the Internet of Things with special focus on building NLP solutions for
real life problems.

Severin Dembele has completed a master’s degree in computer Science with

specialisation in Decision Support Systems at the Public University, Nazi Boni, Bobo-
Dioulasso, Burkina Faso. He is in the process of starting his PhD studies in the field of
NLP which is also his current research interest.

12