Enhancing Industrial Productivity through AI-Driven Systematic

Literature Reviews

Jaqueline Gutierri Coelho

2,5,∗ a

, Guilherme Dantas Bispo

2,5 b

, Guilherme Fay Vergara

1,2,8 c

Gabriela Mayumi Saiki

2,5 d

, Andr

e Luiz Marques Serrano

1,2,4 e

, Li Weigang

1,2,5,∗ f

Clovis Neumann

1,2,4 g

, Patricia Helena Martins

2,6 h

, Welber Santos de Oliveira

1,2,8 i

Angela Brigida Albarello

2,5 j

, Ricardo Accorsi Casonatto

2,4 k

, Patr

ıcia Missel

1,2 l

Roberto de Medeiros Junior

3,7 m

, Jefferson de Oliveira Gomes

3,7 n

, Carlos Rosano-Pe

9 o

and Caroline Cabral F. da Costa

3,7 p

Latitude, Decision Making Technologies Lab, University of Brasilia, Campus Darcy Ribeiro, Brasilia, Brazil

Projectum, Research Group for Math. Methodologies Applied to Management, University of Brasilia, Brasilia, Brazil

SENAI, National Service of Industrial Learning, Federal District, Brazil

Department of Production Engineering, Faculty of Technology, University of Bras

ılia, Federal District, Brazil

Department of Computer Science, University of Brasilia, Federal District, Brazil

Department of Economics, University of Bras

ılia, Federal District, Brazil

UNITEC, Innovation, and Technology Unit, National Service of Industrial Learning, Federal District, Brazil

Department of Electrical, University of Brasilia, Federal District, Brazil

Department of Administration, University of Brasilia, Federal District, Brazil

Keywords:

Artiﬁcial Intelligence, Automatic Classiﬁer, Innovation, Productivity in Industry, Sustainability, SLR.

Abstract:

The advent of Artiﬁcial Intelligence (AI) has opened up new possibilities for improving productivity in various

industry sectors. In this paper, we propose a novel framework aimed at optimizing systematic literature re-

views (SLRs) for industrial productivity. By combining traditional keyword selection methods with AI-driven

classiﬁcation techniques, we streamline the review process, making it more efﬁcient. Leveraging advanced

natural language processing (NLP) approaches, we identify six key sectors for optimization, thereby reducing

workload in less relevant areas and enhancing the efﬁciency of SLRs. This approach helps conserve valu-

able time and resources in scientiﬁc research. Additionally, we implemented four machine learning models

for category classiﬁcation, achieving an impressive accuracy rate of over 75%. The results of our analyses

demonstrate a promising pathway for future automation and reﬁnements to boost productivity in the industry.

https://orcid.org/0000-0002-6517-1957

https://orcid.org/0000-0002-4938-2076

https://orcid.org/0000-0002-4551-2240

https://orcid.org/0009-0008-5941-1601

https://orcid.org/0000-0001-5182-0496

https://orcid.org/0000-0003-1826-1850

https://orcid.org/0000-0003-4320-8795

https://orcid.org/0000-0002-1511-6239

https://orcid.org/0009-0002-7020-5442

https://orcid.org/0009-0002-9455-0144

https://orcid.org/0009-0001-3652-5525

https://orcid.org/0009-0009-2435-4229

https://orcid.org/0009-0002-7436-0554

https://orcid.org/0000-0002-6004-799X

https://orcid.org/0000-0002-6868-9284

1 INTRODUCTION

In a rapidly changing technological landscape, the

efﬁciency of resource allocation and the preserva-

tion of knowledge capital are critical considerations.

Within this context, productivity serves as a vital in-

dicator, assessed primarily through labor productivity

(LP) and total factor productivity (TFP). LP measures

wealth generated per worker by dividing GDP by an-

nual labor hours, while TFP gauges the comprehen-

sive impact of inputs like labor development and cap-

ital assets in the production process. These measures

are fundamental to assessing economic progress and

https://orcid.org/0000-0003-2005-3823

472

Coelho, J., Bispo, G., Vergara, G., Saiki, G., Serrano, A., Weigang, L., Neumann, C., Martins, P., Santos de Oliveira, W., Albarello, A., Casonatto, R., Missel, P., Medeiros Junior, R., Gomes,

J., Rosano-Peña, C. and F. da Costa, C.

Enhancing Industrial Productivity Through AI-Driven Systematic Literature Reviews.

DOI: 10.5220/0012235000003584

In Proceedings of the 19th International Conference on Web Information Systems and Technologies (WEBIST 2023), pages 472-479

ISBN: 978-989-758-672-9; ISSN: 2184-3252

competitiveness related to the evolution of the tech-

nology (Mundial, 2018).

Figure 1: Framework of AI-Driven Systematic Literature

Reviews.

To explore this direction further, the current study

presents an AI-Driven Systematic Literature Reviews

(SLR) framework, depicted in Figure 1, along with an

optimized four-step method to yield the desired out-

comes. The research relies on a publicly accessible

web research database named Web of Science (WoS)

to source the most relevant scientiﬁc articles across

six predeﬁned categories.

In the initial step, datasets are obtained from WoS

using six distinct queries for each category. Subse-

quently, the combination of these datasets creates a

labeled database that facilitates the application of var-

ious artiﬁcial intelligence algorithms to achieve the

desired outcomes. The research not only utilizes these

databases but also emphasizes the impact of each

method employed. It conducts a critical and compar-

ative analysis of these methodologies and discusses

their practical implications. The primary objective is

to gain insights into how different methods model the

SLR process, the underlying phases involved, the as-

sumptions guiding their application, and other con-

tributing factors to this process. By exploring these

aspects, the study seeks to contribute valuable knowl-

edge and understanding enhancing the efﬁcacy of sys-

tematic literature reviews in the context of labor pro-

ductivity research.

2 LITERATURE REVIEW

This section aims to present how the literature pro-

vides the necessary information for the review of ex-

isting literature and practices on productivity. In re-

cent years, studies on the evolution of productivity

have increasingly gained ground in the economic de-

bate in Brazil. This can be seen by the number of

studies such as those by Torezani (2018), Santos et al.

(2019), Borracini (2021), among others. Thus, under-

standing the pattern of productivity evolution is justi-

ﬁed by the need to assimilate a country’s competitive-

ness, either to maintain a space in the international

scenario or to allow economic growth.

2.1 Systematic Literature Review (SLR)

Systematic literature review (SLR), a long-standing

practice in health care, was consolidated in the late

1980s with the publication of the book Effective care

during pregnancy and childbirth. In this context, it has

been used as a way to overcome weaknesses and min-

imize biases left by narrative reviews, whose guiding

threads can follow the multiple domains of knowledge

arising from it. The following steps need to be fol-

lowed: Subject search sources, Strategies for research

bias, Evaluation of the studies and literature selected

for SLR, Synthesis of results, and Study report.

It is noteworthy that one of the main strengths of

the SLR is the focus on a speciﬁc search, the clarity in

retrieving articles for review, objective and quantita-

tive summaries, and evidence-based inferences Ades-

ope et al. (2010). Galv

ao et al. (2017) summarize four

types of mixed reviews: the quantitative convergence

mixed review, the qualitative convergence mixed re-

view, the exploratory sequential hybrid review, and

the explanatory sequential mixed review. The quan-

titative convergence mixed review translates qualita-

tive, quantitative, and mixed studies into qualitative

ﬁndings. The exploratory sequential hybrid review

occurs in two phases: transformation of results into

qualitative ﬁndings and tabulation and comparison of

quantitative results. The exploratory sequential mixed

review measures the effects of an action and explains

their differences.

Regarding the quality of reviews, the Oxford Clas-

siﬁcation of Level of Evidence established by Howick

(2011), understands that systematic reviews of cross-

sectional studies, systematic reviews of cohort stud-

ies, and systematic reviews of randomized controlled

trials have level 1 evidence on a scale of 1 to 5, with

1 being the highest level of evidence. Therefore, this

classiﬁcation highlights the importance of systematic

reviews for scientiﬁc progress and different decision-

making.

On the other hand, Cohen et al. (2006), point out

that the automatic classiﬁcation of documents is an

effective practice that brings clarity to the system-

atic literature review. Following this line, de Melo

et al. (2022) demonstrate a change in the 2006 met-

ric, now called Adjusted Work Saved over Sam-

Enhancing Industrial Productivity Through AI-Driven Systematic Literature Reviews

473

pling (AWSS@R), such metric brings the possibility

of comparison between different domains, especially

with regard to the automated citation screening step.

3 CATEGORY OF

PRODUCTIVITY IN INDUSTRY

In our study, we conducted a systematic literature re-

view by the traditional method of manual keyword

selection and ﬁltering. We then used the Category

Classiﬁcation of Scientiﬁc Articles and the Relevance

Classiﬁcation of Scientiﬁc Articles as a fair workload

measure to compare and demonstrate its efﬁciency.

To conduct a comprehensive bibliometric review

on the topic of Productivity in Industry, we will adopt

a rigorous approach by applying speciﬁc ﬁlters to en-

sure the relevance and timeliness of the included stud-

ies. In order to cover a signiﬁcant time range, we will

analyze articles published in the period 2014 to 2023,

considering several reliable sources.

Within the context of productivity in industry,

there are six main axes that play essential roles in the

advancement and transformation of the sector. These

axes are innovation, sustainability, industry 4.0, in-

dustry 5.0, efﬁciency, and artiﬁcial intelligence. We

will explore each of them below:

Innovation improves industrial productivity by in-

troducing new ideas, technologies, and processes,

optimizing operations, and boosting efﬁciency. It

also allows for adaptation to market changes and the

search for creative solutions to industrial challenges.

Sustainability is key in the industry, with practices

that reduce resource consumption, minimize environ-

mental impact and promote energy efﬁciency. Com-

panies adopt sustainable strategies to improve produc-

tivity and reduce their environmental impact.

Industry 4.0 integrates advanced digital technolo-

gies such as IoT, big data, and artiﬁcial intelligence

to create smart factories. This results in automation,

improved quality control, and higher productivity.

Industry 5.0 integrates advanced technologies,

such as collaborative robotics and artiﬁcial intelli-

gence, with human labor to improve productivity. Hu-

mans and machines work together, leveraging their

unique abilities, combining efﬁciency and creativity.

Efﬁciency Is fundamental to industrial productiv-

ity, involving process optimization, waste reduction,

and quality improvement. It is achieved through con-

tinuous analysis and improvement of processes, aim-

ing for more effective operational performance.

Artiﬁcial Intelligence (Weigang et al., 2022)

drives industrial productivity through automation,

supply chain optimization, and data-driven decision-

making, resulting in greater efﬁciency and quality.

These six axes interact with each other, forming

a set of approaches and technologies that contribute

to boosting productivity in the industry. The adop-

tion and integration of these elements into industrial

strategies and processes can result in signiﬁcant gains

in efﬁciency, competitiveness, and sustainability.

4 SLR BY WEB OF SCIENCE

Web of Science by Clarivate Analytics is a widely rec-

ognized and extensively utilized platform among re-

searchers, providing valuable access to academic re-

sources, including articles and conferences, as well as

a comprehensive citation index. Its robust features,

such as research proﬁles and citation alerts, facilitate

the exploration of relevant scholarly works and en-

able the monitoring of publication impact. Undoubt-

edly, this tool plays a crucial role in supporting sys-

tematic reviews, owing to its expansive coverage and

advanced ﬁltering capabilities. Consequently, Web of

Science signiﬁcantly contributes to the advancement

of academic knowledge.

4.1 Database

Therefore, the Web of Science is essential for struc-

turing the data of the articles for the analysis of each

axis. To structure the article data for analysis in each

category, the process involves:

1. Selection of categories: researchers begin by se-

lecting the relevant categories pertinent to their

analysis;

2. Keyword search in WoS: utilizing speciﬁc key-

words, such as ”industry”, ”productivity”, and the

chosen category names, it was possible to perform

complete searches in Web of Science;

3. Creation of new column for category label: upon

retrieving relevant data from the searches, re-

searchers create new columns in their dataset to

label each selected category;

4. Replication of the process for other categories:

repeat the search and data retrieval process for

the remaining ﬁve categories, creating dedicated

columns for each;

5. Compilation of databases: the data obtained from

the six categories are then compiled into a uniﬁed

database, referred to as the ”Labeled Database”;

6. Filtering of Replications: to avoid redundancies,

WEBIST 2023 - 19th International Conference on Web Information Systems and Technologies

474

duplicates were removed ensuring that each arti-

cle is linked to only one category.

After completing the data extraction stage, a

uniﬁed database was obtained, structured in nine

columns: Title, Abstract, Author, Journal, Country,

Keywords, Citation (ordered by frequency, with the

most frequent citations at the top), DOI, and Cate-

gory. This database will play a key role in the training

process of the machine learning models used by the

classiﬁcation of categories, as shown in 1.

4.2 Analysis of the Database

Bibliometrix is a computational tool for bibliometric

review analysis, it is one of the RStudio packages.

The information regarding the database compiled and

with the ﬁlters resulted in 7781 articles in the period

from 2014 to 2023, with a growth rate of 12.18% of

publications per year, about 16,141 authors, 29.97%

with international co-authorship, 284,418 references,

and 20,055 keywords.

In the period 2014 to 2023, the country that pub-

lished the most was China, with 8564 articles pub-

lished, followed by the USA with 2624 articles and

the UK with 1289 articles. Brazil has about 729 ar-

ticles published, ranking eighth among countries that

have published the most on the subject among the six

categories.

Regarding the main topics covered in articles, be-

tween the period 2014 to 2016, the main topics cov-

ered in the articles are industry, construction, produc-

tivity, design, optimization. From 2017 to 2019, the

main topics addressed are performance, design, inno-

vation, data envelopment analysis, productivity, and

fermentation. From 2020 to 2023 the main topics cov-

ered are productivity, systems, innovation, biomass,

management, and optimization. Among the cate-

gories covered in the survey, only innovation ap-

peared as one of the main themes of the 7,781 arti-

cles. Among the two basic keywords of the research,

industry, and productivity, it was found that between

2014 and 2016 industry is a more prominent theme

and during the period from 2017 to 2023 productivity

is the most addressed theme.

In terms of the country of the corresponding au-

thors for the articles, China stands out with the high-

est number of authors, totaling 2361, followed by the

USA with 674, and India with 361 authors. Addi-

tionally, China also leads in terms of the number of

citations received, boasting an impressive 39853 cita-

tions, which is more than double the number of cita-

tions secured by the USA, ranking second with 15486

citations. These statistics demonstrate China’s sig-

niﬁcant dominance not only in authorship and arti-

cle production but also in the relevance and produc-

tivity within the industry, particularly across the six

research categories. The data clearly indicates that

China’s research efforts in the focused theme surpass

that of other countries, cementing its leading position

in the ﬁeld.

5 CATEGORY CLASSIFIER OF

SCIENTIFIC ARTICLES

In this chapter, the process adopted to train and evalu-

ate four machine learning models (Multinomial Naive

Bayes Classiﬁcation, Stochastic Gradient Descent,

Support Vector Machines and Decision Tree Classi-

ﬁer) for classifying scientiﬁc articles in six categories

will be presented: Innovation, Sustainability, Indus-

try 4.0, Industry 5.0, Efﬁciency and Artiﬁcial Intelli-

gence.

5.1 Data Preparation

In this study, we curated the Web of Science (WoS)

database by excluding articles lacking essential in-

formation and ensuring that each article belonged to

only one category to prevent duplication. This data

cleanup process maintained the integrity of our train-

ing and validation sets, as depicted in Figure 2.

Figure 2: Distribution by category without duplicates.

The dataset was then randomly divided into 70%

for training and 30% for validation, ensuring that both

datasets were representative of the six categories.

For the textual representation of the articles, a

corpus was built using the TF-IDF approach (Term

Frequency-Inverse Document Frequency) to convert

the text into numerical vectors. Furthermore, stem-

ming was applied to reduce the words to their stem,

which helped to reduce the dimensionality of the fea-

ture space and improve the performance of the mod-

els.

Enhancing Industrial Productivity Through AI-Driven Systematic Literature Reviews

475

5.2 Machine Learning Models

With the training set prepared, the four selected ma-

chine learning models were trained using the training

data.

1. Multinomial Naive Bayes Classiﬁcation (MNBC)

According to Chebil et al. (2023), as MNBC uses

the Frequency Estimate parameter to highlight the

frequencies of the available date.

2. Stochastic Gradient Descent (SGD) is observed

as a classiﬁer that learns functions of increas-

ing complexity and generalizes over parameter-

ized data (Nakkiran et al., 2019).

3. Support Vector Machines (SVM) Overall, SVM

takes a non-linear input set and converts it to lin-

ear with the help of a kernel function (Leong et al.,

2021).

4. Decision Tree Classiﬁer (DTC) Wang et al. (2020)

deﬁne DTC as a hierarchical classiﬁer, as it pro-

vides multi-level classiﬁcation and provides the

speciﬁc pattern to which each data belongs, in

addition to allowing ﬂexibility against binary and

multi-class classiﬁcations.

For each model, a hyper-parameter adjustment

was performed using the GridSearchCV technique,

which consists of deﬁning a grid of possible values

for the hyper-parameters and performing an exhaus-

tive search for the optimal combinations.

5.3 Method Validation Metrics

Model validation is crucial in machine learning de-

velopment, assessing performance on unseen data

through metrics like recall, precision, accuracy, F-

measure, and confusion matrices. Careful analysis

of these aspects helps select the best model and ﬁne-

tune parameters for optimal performance, considering

problem-speciﬁc requirements and class distinctions.

6 RELEVANCE CLASSIFIER OF

SCIENTIFIC ARTICLES

Top articles were identiﬁed using citation count, and

outlier detection techniques like z-scores and standard

deviation were employed for selection.

7 DISCUSSION OF RESULTS

This section presents the results of the research by

presenting and analyzing the performance of four dif-

Figure 3: Relevance Classiﬁer of Scientiﬁc Articles.

ferent classiﬁcation models applied to the task of cat-

egorizing scientiﬁc articles.

7.1 Category Classiﬁer

The models evaluated are Multinomial Naive Bayes

(MNB), SGD Classiﬁer, Support Vector Machine

(SVM), and Decision Tree. Each model was run with

different hyper-parameter settings and their perfor-

mance was evaluated using metrics such as accuracy,

precision, recall and F-measure. The aim is to iden-

tify the best model that ﬁts the proposed problem of

subject categorization. For this analysis, data that has

been pre-processed and transformed into vectors has

been used. The corpus consists of documents labeled

with different subject categories, and the objective is

to train the models to correctly classify new docu-

ments into their respective categories.

Table 1: Web of Science results.

Variables

MultinomialNB

(Naive Bayes)

SGDClassiﬁer SVM

Decision

Tree

Accuracy 0.6748 0.7380 0.7494 0.7554

Precision 0.6542 0.7270 0.7384 0.7378

Recall 0.6748 0.7380 0.7494 0.7554

F-measure 0.6606 0.7134 0.7367 0.7441

Table 1 shows the values of the metrics for each of

the models. It is also possible to see the performance

for model, especially, the accuracy rate arrive 75% by

Decision Tree.

1) Multinomial Naive Bayes (MNB) is a classi-

ﬁer based on Bayes’ theorem and is widely known

for its simplicity and computational efﬁciency. When

we evaluated the performance of MNB on our sub-

ject categorization task, we found that both accuracy

and recall were the same, around 67.48%. These re-

sults indicate that the model correctly classiﬁed ap-

proximately 67.48% of the instances into their correct

categories. Although this is a reasonable result, we

cannot consider it exceptional.

When analyzing the speciﬁc accuracy of MNB,

we conﬁrm that it is 67.48%, and the F-measure is

66.06%.

WEBIST 2023 - 19th International Conference on Web Information Systems and Technologies

476

Despite its advantages in terms of simplicity

and computational efﬁciency, the results suggest that

MNB may not be the best choice for our speciﬁc task

of subject categorization. In this context, it is essen-

tial to consider alternative models that have obtained

more promising results, in order to select the best

model to meet the requirements and objectives of our

project.

2) The Stochastic Gradient Descent (SGD) classi-

ﬁer is a linear classiﬁer that uses stochastic gradient

descent computation for optimization. When eval-

uating its performance on the subject categorization

task, our results show that the SGD classiﬁer achieved

an accuracy of 73.80%, a precision of 72.71% and

a recall of 73.80%. In addition, the F-measure was

71.34%. These results indicate that the model outper-

formed MNB, which can be attributed to its ability to

better handle more complex classiﬁcation problems.

The recall of 73.80% indicates that the SGD clas-

siﬁer was able to retrieve approximately 73.80% of

the categories of each item. The precision of 72.71%

means that approximately 72.71% of the classiﬁed

items actually belonged to that category, i.e. there was

a signiﬁcant proportion of true positives.

These superior results in terms of accuracy, recall

and precision compared to MNB can be attributed to

the SGD classiﬁer’s properties and ability to deal with

linear relationships between problem features, which

can be particularly useful in more complex classiﬁ-

cation tasks. Therefore, the results indicate that the

SGD classiﬁer is a more promising option for the

task of subject categorization compared to MNB, but

we should still consider the other models evaluated

to identify the best choice that meets the speciﬁc re-

quirements of this study.

3) Support Vector Machines (SVM) is a widely

used classiﬁer in classiﬁcation problems and is known

for its effectiveness in high-dimensional spaces. Our

results show that the SVM achieved an accuracy

of 74.94%, a precision of 73.84% and a recall of

74.94%. In addition, the F-measure was 73.67%.

These results are promising and show that SVM out-

performed both MNB and SGD Classiﬁers.

The recall of 74.94% indicates that the SVM was

able to recover approximately 74.94% of the clas-

siﬁed categories. In turn, the precision of 73.84%

shows the percentage of classiﬁed categories that ac-

tually belonged to that category, demonstrating a sig-

niﬁcant proportion of true positives.

These superior results in terms of accuracy, recall,

and precision compared to MNB and SGD Classiﬁer

conﬁrm the effectiveness of SVM for the task of sub-

ject categorization, especially in scenarios with high

data dimensionality and classiﬁcation complexity.

Therefore, the results highlight SVM as a highly

competitive and successful option in the subject cate-

gorization task. However, it is still crucial to consider

other factors, such as runtime and interpretability, in

order to select the most suitable model for the speciﬁc

needs and requirements of our project.

4) Decision Tree is a decision rule-based model

known for its interpretability. Our results show that

the Decision Tree achieved an accuracy of 75.54%, a

precision of 73.78%, and a recall of 75.54%. In ad-

dition, the F-measure was 74.41%. These results are

the best of the models tested, indicating that Decision

Tree excelled at the task of subject categorization.

The recall of 75.54% means that Decision Tree

was able to correctly retrieve most of the categorized

categories. The precision of 73.78% means the per-

centage of classiﬁed categories that actually belonged

to that category, which is a positive result.

These satisfactory results in terms of accuracy, re-

call, and precision conﬁrm that the Decision Tree is

a competitive option for the task of subject catego-

rization. Moreover, its interpretability is a signiﬁcant

advantage, allowing a better understanding of the de-

cisions made by the model and facilitating the identi-

ﬁcation of classiﬁcation patterns.

Based on the results presented, we can conclude

that Decision Tree was the model that obtained the

best performance for the subject categorization task.

Its accuracy of 75.54% and F-measure of 74.41% out-

performed the other models evaluated. The inter-

pretability of Decision Tree can be a signiﬁcant ad-

vantage when dealing with subject categorization.

7.2 Relevance Classiﬁer

After classifying the 17,000 articles into subject cate-

gories, they were further grouped into three categories

of relevance: Irrelevant, Less Relevant, and Relevant.

The categorization was based on their citation counts

using statistical metrics such as the mean and standard

deviation and Z-score.

Articles with citation counts lower than the mean

were classiﬁed as ”Irrelevant.” Articles with cita-

tion counts falling between the mean and the up-

per bound of the standard deviation were labeled as

”Less Relevant.” On the other hand, articles with a

z-score greater than 3, indicating a substantial devi-

ation from the mean, were classiﬁed as ”Relevant.”

This approach allowed for a comprehensive and nu-

anced assessment of the articles’ signiﬁcance and im-

pact within their respective subject categories.

Below is a table 2 displaying the distribution of

articles across each category:

In the table 3 in the appendix, we present a table

Enhancing Industrial Productivity Through AI-Driven Systematic Literature Reviews

477

Table 2: Quantity of articles per category.

Category Irrelevant Less Relevant Relevant

Innovation 6317 473 253

Efﬁciency 3606 353 238

Sustainability 1181 97 59

Industry 4.0 1232 98 53

Artiﬁcial Intelligence 184 10 7

showcasing the names of the top 5 most relevant arti-

cles within each category:

8 CONCLUSION

In conclusion, this paper aimed to compare systematic

literature review methods to identify the most efﬁcient

approach for conducting quick and effective literature

reviews. We categorized productivity in the industry

into six principal sectors, namely Innovation, Sustain-

ability, Industry 4.0, Industry 5.0, Efﬁciency, and Ar-

tiﬁcial Intelligence. Subsequently, we conducted sys-

tematic literature reviews for each of these sectors.

To achieve efﬁcient categorization, we employed

four Machine Learning models: Multinomial Naive

Bayes Classiﬁcation (MNBC), Stochastic Gradient

Descent (SGD), Support Vector Machines (SVM),

and Decision Tree Classiﬁer (DTC). Among these

models, the Decision Tree Classiﬁer (DTC) demon-

strated superior performance with a classiﬁcation ac-

curacy exceeding 75%.

To identify the most relevant contributions of sci-

entiﬁc articles in each research domain or category,

we utilized a relevance classiﬁer with two essential

metrics: the standard deviation of the mean and the Z

score.

As a forward-looking suggestion for further re-

search, we highlight the signiﬁcance of incorporating

additional data sources and considering the h-index

as an alternative approach for ranking the relevance

of articles. By expanding the scope of data and incor-

porating diverse evaluation metrics, future studies can

enhance the accuracy and depth of systematic litera-

ture reviews, ultimately contributing to more compre-

hensive insights and informed decision-making in the

domain of industrial productivity.

ACKNOWLEDGMENTS

Thanks to the University of Brasilia (UnB) for sup-

porting this research and to the National Industrial

Learning Service (SENAI) of the National Confed-

eration of Industry (CNI) for partially supporting this

research.

REFERENCES

Adesope, O. O., Lavin, T., Thompson, T., and Ungerleider,

C. (2010). A systematic review and meta-analysis of

the cognitive correlates of bilingualism. Review of ed-

ucational research, 80(2):207–245.

Borracini, G. F. (2021). O impacto dos subs

ıdios na

evoluc¸

ao da produtividade do setor agr

ıcola: e

obtenc¸

ao do t

ıtulo de mestre no mestrado proﬁssional

em economia. IDP/FAB.

Chebil, W., Wedyan, M., Alazab, M., Alturki, R., and

Elshaweesh, O. (2023). Improving semantic informa-

tion retrieval using multinomial naive bayes classiﬁer

and bayesian networks. Information, 14(5):272.

Cohen, A. M., Hersh, W. R., Peterson, K., and Yen, P.-

Y. (2006). Reducing workload in systematic review

preparation using automated citation classiﬁcation.

Journal of the American Medical Informatics Associ-

ation, 13(2):206–219.

de Melo, M. K., Faria, A. V. A., Weigang, L., Nery, A. G.,

de Oliveira, F. A. R., Barreiro, I. T., and Celestino, V.

R. R. (2022). Few-shot approach for systematic lit-

erature review classiﬁcations. Proceedings of the 18th

International Conference on Web Information Systems

and Technologies WEBIST.

Galv

ao, M. C. B., Pluye, P., and Ricarte, I. L. M.

(2017). M

etodos de pesquisa mistos e revis

oes de

literatura mistas: conceitos, construc¸

ao e crit

erios de

avaliac¸

ao. InCID: Revista de Ci

encia da Informac¸

e Documentac¸

ao, 8(2):4–24.

Howick, J. (2011). The oxford 2011 levels of evidence.

http://www. cebm. net/index. aspx? o= 5653.

Leong, W. C., Bahadori, A., Zhang, J., and Ahmad, Z.

(2021). Prediction of water quality index (wqi) us-

ing support vector machine (svm) and least square-

support vector machine (ls-svm). International Jour-

nal of River Basin Management, 19(2):149–156.

Mundial, B. (2018). Emprego e crescimento: a agenda da

produtividade. Bras

ılia: Grupo Banco Mundial.

Nakkiran, P., Kaplun, G., Kalimeris, D., Yang, T., Edelman,

B. L., Zhang, F., and Barak, B. (2019). Sgd on neu-

ral networks learns functions of increasing complex-

ity. arXiv preprint arXiv:1905.11604.

Santos, M. R. d., Pandullo, R. P., and Souza, E. C. d. (2019).

Evoluc¸

ao da produtividade total dos fatores e da pro-

dutividade da m

ao de obra da ind

ustria brasileira: uma

alise setorial e comparativa. Universidade Cat

olica

de Bras

ılia.

Torezani, T. A. (2018). Evoluc¸

ao da produtividade

brasileira: mudanc¸a estrutural e din

amica tecnol

ogica

em uma abordagem multissetorial. Universidade Fed-

eral do Rio Grande do Sul.

Wang, F., Wang, Q., Nie, F., Li, Z., Yu, W., and Ren, F.

(2020). A linear multivariate binary decision tree clas-

siﬁer based on k-means splitting. Pattern Recognition,

107:107521.

Weigang, L., Enamoto, L. M., Li, D. L., and Rocha Filho,

G. P. (2022). New directions for artiﬁcial intelli-

gence: human, machine, biological, and quantum in-

telligence. Frontiers of Information Technology &

Electronic Engineering, 23(6):984–990.

WEBIST 2023 - 19th International Conference on Web Information Systems and Technologies

478

APPENDIX

Table 3: Relevance articles extracted from WoS classiﬁed by category.

Category Title

Artiﬁcial

Intelligence

Brave new world: service robots in the frontline

Industry 5.0-A Human-Centric Solution

Agriculture 4.0: Broadening Responsible Innovation in an Era of Smart Farming

Industrial internet of things: Recent advances, enabling technologies and open challenges

From Industry 4.0 to Agriculture 4.0: Current Status, Enabling Technologies, and Research

Challenges

Efﬁciency

Impact of energy conservation policies on the green productivity in China’s manufacturing

sector: Evidence from a three-stage DEA model

The Role and Impact of Industry 4.0 and the Internet of Things on the Business Strategy of

the Value Chain-The Case of Hungary

Energy and CO2 emissions performance in China’s regional economies: Do market-oriented

reforms matter?

Modeling the role of environmental regulations in regional green economy efﬁciency of

China: Empirical evidence from super efﬁciency DEA-Tobit model

Enhancing microalgal biomass productivity by engineering a microalgal-bacterial community

Industry 4.0

State-of-the-art in surface integrity in machining of nickel-based super alloys

The cost of additive manufacturing: machine productivity, economies of scale and

technology-push

The link between Industry 4.0 and lean manufacturing: mapping current research and estab-

lishing a research agenda

Multi-response optimization of minimum quantity lubrication parameters using Taguchi-

based grey relational analysis in turning of difﬁcult-to-cut alloy Haynes 25

A Survey on Industrial Internet of Things: A Cyber-Physical Systems Perspective

Innovation

Innovation in the pharmaceutical industry: New estimates of R&D costs

Inﬂuence of tribology on global energy consumption, costs and emissions

Understanding the implications of digitisation and automation in the context of Industry 4.0:

A triangulation approach and elements of a research agenda for the construction industry

Environmental regulation and competitiveness: Empirical evidence on the Porter Hypothesis

from European manufacturing sectors

Different Types of Environmental Regulations and Heterogeneous Inﬂuence on Green Pro-

ductivity: Evidence from China

Sustainability

Natural ﬁber reinforced polymer composites in industrial applications: feasibility of date

palm ﬁbers for sustainable automotive industry

Nanotechnology in Sustainable Agriculture: Recent Developments, Challenges, and Perspec-

tives

’Green’ productivity growth in China’s industrial economy

Sustainable manufacturing in Industry 4.0: an emerging research agenda

Can carbon emission trading scheme achieve energy conservation and emission reduction?

Evidence from the industrial sector in China

Enhancing Industrial Productivity Through AI-Driven Systematic Literature Reviews

479