Tunisia Big Data Lmi Analysis 2020 Web
Tunisia Big Data Lmi Analysis 2020 Web
Tunisia Big Data Lmi Analysis 2020 Web
MARKET INFORMATION
(LMI) IN TUNISIA
Methodological overview and Analytics insights on
Tunisian Web Labour Market
WORKING PAPER
December 2020
The contents of this draft report are the sole responsibility of the authors and do not necessarily reflect the views of the
ETF or the EU institutions.
TABLE OF ABBREVIATIONS 46
REFERENCES 47
3
FIGURES AND TABLES
FIGURE 3.1. THE KDD PROCESS ......................................................................................................... 7
FIGURE 6.1. EVOLUTION OF ACTIVE POPULATION, 2016–2019 .................................................... 11
FIGURE 6.2. EVOLUTION OF THE UNEMPLOYED POPULATION, 2016–2019 ............................... 11
FIGURE 6.3. UNEMPLOYMENT RATE BY GENDER, 2016–2019 (%) ............................................... 12
FIGURE 6.4. UNEMPLOYMENT RATE BY AGE, 2019 (%) ................................................................. 13
FIGURE 6.5. UNEMPLOYMENT RATE FOR HIGHER EDUCATION GRADUATES, 2016–2019 (%) 13
FIGURE 6.6. EVOLUTION OF THE EMPLOYED POPULATION, 2016–2019 .................................... 14
FIGURE 6.7. DISTRIBUTION OF EMPLOYED POPULATION BY INDUSTRY, 2019 ......................... 15
FIGURE 6.8. SERVICES EMPLOYED POPULATION, 2019 ................................................................ 15
FIGURE 6.9. MANUFACTURING INDUSTRIES EMPLOYED POPULATION, 2019 ........................... 16
FIGURE 6.10. NON-MANUFACTURING INDUSTRIES EMPLOYED POPULATION, 2019 ................ 16
FIGURE 8.1. PUBLICATION OF OJVS BY MONTH, APRIL–SEPTEMBER 2020 ............................... 19
FIGURE 8.2. OJVS BY OCCUPATION (LEVEL 1) ............................................................................... 20
FIGURE 8.3. OJVS BY SKILL LEVEL ................................................................................................... 20
FIGURE 8.4. OJVS BY OCCUPATION (LEVEL 4) ............................................................................... 21
FIGURE 8.5. OJVS BY OCCUPATION (LEVEL 1), AUGUST 2020 ..................................................... 21
FIGURE 8.6. OJVS BY OCCUPATION (LEVEL 4), AUGUST 2020 ..................................................... 22
FIGURE 8.7. OJVS BY EDUCATIONAL LEVEL ................................................................................... 22
FIGURE 8.8. OJVS BY EXPERIENCE .................................................................................................. 23
FIGURE 8.9. OJVS BY TYPE OF CONTRACT..................................................................................... 23
FIGURE 8.10. OJVS BY CONTRACT AND OCCUPATION ................................................................. 24
FIGURE 8.11. OJVS BY CONTRACT AND SKILL LEVEL ................................................................... 25
FIGURE 8.12. OJVS AND EMPLOYED POPULATION BY INDUSTRY .............................................. 25
FIGURE 8.13. OJVS BY INDUSTRY ..................................................................................................... 26
FIGURE 8.14. OJVS FOR THE ADMINISTRATIVE AND SUPPORT SERVICE ACTIVITIES SECTOR
........................................................................................................................................................ 27
FIGURE 8.15. OJVS FOR THE INFORMATION AND COMMUNICATION SECTOR ......................... 27
FIGURE 8.16. EMPLOYED POPULATION BY INDUSTRY, 2019 ....................................................... 27
FIGURE 8.17. OJVS BY KNOWLEDGE REQUIREMENTS ................................................................. 28
FIGURE 8.18. OJVS BY PERSONAL QUALITIES REQUIREMENTS ................................................. 29
FIGURE 8.19. OJVS BY SKILLS REQUIREMENTS ............................................................................. 29
FIGURE 8.20. OJVS BY TOOLS AND TECHNOLOGY REQUIREMENTS.......................................... 29
FIGURE 8.21. OJVS BY SKILLS/KNOWLEDGE REQUESTED FOR SOFTWARE DEVELOPERS ... 30
FIGURE 8.22. OJVS BY SKILLS/KNOWLEDGE REQUESTED FOR SHOP SALES ASSISTANTS .. 31
4
1 INTRODUCTION – PURPOSE OF THE PROJECT
Governments and socioeconomic partners in most ETF partner countries are unanimous on the need
to develop and better use information on labour market and skills dynamics to improve the
performance of education and training, the availability of qualifications and skills for employment, and
the lifelong societal and personal development of individuals. In this context, most partner countries
have been reinforcing their systems, capacities and methods to identify, analyse and anticipate
demand and skills needs in a context of changing economic structures, new types of work, and rapid
digital transformation of occupations and tasks.
While conventional labour market intelligence (LMI), based on regular statistics, specific-purpose
surveys and qualitative methods, has gained ground in ETF partner countries, there is room to further
innovate data sources, improve analytical capacities and modernise the formats and instruments to
visualise and disseminate insights for users (policy makers, socioeconomic partners, education and
training players).
Big Data analytics offers new opportunities to improve LMI and deliver real-time and fine-grained skills
analysis and insights for users. Big Data is all around us. Big Data is characterised by volume, variety,
velocity and – eventually – value. Machine learning and artificial intelligence (AI) algorithms, combined
with immense computing power anytime and anywhere, allow data science to exploit specific Big Data
sources that have great potential to supplement and enrich conventional LMI. This is the case for
online job vacancies (OJVs) managed by a large variety of online job portals and boards.
Creating knowledge out of large volumes of data that are available at high velocity and with great
variety is the major goal of Big Data analysis. It is about value. Analysis of thousands of millions of job
vacancies can describe much about the skills that employers want, in almost real time and in fine-
grained detail. Screening and ranking of OJV portals – the first step of the methodology – can tell us
much about the overall panorama of the online/digital labour market in countries and regions, the
features of the individual job portals, the volume of OJVs posted, and the sectoral and occupational
coverage of OJVs. Most importantly, analysis of OJVs reveals specifics of how employers describe
jobs/tasks, the mix of skills they seek, and the importance they attribute to credentials/qualifications.
OJVs are a rich source of information about the skills and other job requirements that
employers require, information that is difficult to gather via other conventional methods. Data
from OJVs do not replace other types of LMI, but add value and can be combined with
conventional statistical data.
5
demand for digital skills plays an increasing role within all professional profiles, even those not
necessarily related to the information and communication technology (ICT) sector. Knowledge of these
current changes can certainly be gained through the study and analysis of the data that companies
publish on the web when searching for professionals to meet to their needs. An example in this
direction is the growing diffusion of services for online recruitment (e-recruitment) which make it
possible to publish a job request through various web sources, such as online newspapers,
employment agencies, specialised websites and job vacancy aggregators. The development of
algorithms, methodologies and systems for labour market analysis for the synthesis of useful
information for decision making has recently been termed ‘labour market intelligence’, by which we
mean the definition and implementation of AI and Big Data techniques for the processing and
synthesis of labour market data with the aim of supporting decision-making processes.
LMI is an emerging cross-disciplinary field of study that is attracting research interest in both industrial
and academic communities, as summarised below. Since the early 1990s, text classification (TC) has
been an active research topic. It has been defined as ‘the activity of labelling natural language texts
with thematic categories from a predefined set’ (Sebastiani, 2002). Most popular techniques are based
on the machine learning paradigm, according to which an automatic text classifier is created by using
an inductive process that is able to learn, from a set of preclassified documents, the characteristics of
the categories of interest. Recently, TC has been shown to give good results in categorising much
real-life web-based data such as news and social media, and sentiment analysis. Skills extraction
from OJVs can also be associated to the information extraction and named entity recognition
frameworks. The latter has been applied to solve numerous domain-specific problems in the areas of
information extraction and normalisation. In recent years, public administrations have started to
explore new ways of supporting knowledge management and of obtaining detailed and fresh
information about the labour market. Here, administrative information collected by public
administrations has been used to study labour market dynamics, conducting both data quality and
knowledge discovery activities through AI techniques. Unfortunately, administrative data are collected
when people are hired (and only in countries where the state collects such information), so do not
provide information about labour demand.
This problem is also relevant for businesses, and this has motivated the development of several
commercial products providing job seekers and companies with skill-matching tools. Firms have a
strong need to automate human resource department activities; consequently, a growing number of
commercial skill-matching products have been developed in recent years.
We aim to classify OJVs according to a target classification system for building a (language-
independent) knowledge base for analysis purposes, rather than matching resumes on job vacancies.
Our approach aims to build a knowledge-graph to support fact-based decision-making activities for
LMI.
6
enable unstructured data to produce knowledge through the knowledge discovery in databases (KDD)
process. This approach enables management of Big Data for LMI and consists of five main steps
(Figure 3.1): selection, preprocessing, transformation, data mining and machine learning,
interpretation/evaluation. Clearly, this needs to be adapted to the domain of interest, adjusting one
task or step in relation to another.
Selection. Selection of data sources is the first step. Each internet source must be evaluated and
ranked in terms of the reliability of the information. This phase should consider, for example, the
vacancy publication date, the website’s update frequency, the presence of structured data, and any
downloading restrictions. At the end of this phase, a ranking of reliable web sources is produced.
Preprocessing. This step includes data cleaning to remove noise from the data or inappropriate
outliers (if any), deciding how to handle missing data, and identifying a function to detect and remove
duplicated entries (e.g. duplicated vacancies or vacancies with missing values). Data quality and
cleaning are essential tasks in any data-driven decision-making approach, to guarantee the credibility
of the overall process. Identification of duplicated job vacancies is far from straightforward. Job
vacancies are usually posted on multiple websites, and this is a duplication, whereas re-use of the
same text to advertise a similar position is not. Identification of appropriate features for correct
recognition of duplicates in the internet labour market domain is crucial. The preprocessing step
reduces the complexity of the Big Data scenario, mitigating the impact of the veracity dimension
through data quality and cleaning.
Transformation. This step includes data reduction and projection, which aim to identify a unified
model to represent the data, depending on the purpose of the exercise. Furthermore, it may involve
the use of dimensionality reduction or transformation methods to reduce the effective number of
variables or to find invariant representations for the data. Like preprocessing, the transformation step
reduces the complexity of the data set by addressing the variety dimension. It is usually performed by
the usage of extraction, transformation and loading (ETL) techniques, which support the data
preprocessing and transformation phases in the KDD process. Roughly speaking, through ETL, the
data extracted from a source system undergo a series of transformation routines that analyse,
manipulate and then clean the data before loading them into a knowledge base. By the end of this
step, the outcome of which is a clean, well-defined data model, the Big Data variety issue should be
resolved.
7
Data mining and machine learning. The aim of this step is to identify appropriate AI algorithms (e.g.
classification, prediction, regression, clustering, information filtering) by searching for patterns of
interest in a particular representational form, based on the purpose of the analysis. More specifically,
in the context of LMI, it usually requires the use of TC algorithms (e.g. ontology based or machine
learning based) to build a classification function for mapping data items into one of several predefined
classes. This step is crucial as it is mainly devoted to the extraction of knowledge from the data.
Interpretation/evaluation. This final step employs visual paradigms to visually represent the
knowledge obtained, depending on the user’s objectives. In the LMI context, it means considering the
user’s ability to understand the data and their main goal in the LMI field.
The collection of OJVs aims to extract the following variables, classified according to international
standards so that the data are comparable, regardless of the country analysed:
Once internet labour market data have been produced, they must be delivered to end users according
to the needs of stakeholders. For labour market analysts, interactive dashboards have been built that
allow analysis of internet labour market dynamics and trends following a predefined model, ensuring
data integrity and protection. It should be emphasised that a key role in data validation is that of the
national expert. These experts’ knowledge of the country analysed allows them on the one hand to
verify and validate what emerges from the analysis of OJVs and on the other hand to help in the
interpretation of any phenomena that emerge. Only those who know the local area can make the most
effective use of web data, always with the intention of integrating rather than replacing official
statistical data, which represent a stock figure in effect and therefore a ‘snapshot’ at a certain time, in
contrast to web data.
1 ESCO is a multilingual classification system for European skills, competences, qualifications and occupations,
developed by the European Commission. The ESCO classification corresponds to the International Standard
Classification of Occupations (ISCO-08) up to the fourth-digit level. It then extends ISCO through an additional
level of occupations and skills, organised as a graph rather than a tree (i.e. a skill may belong to multiple
occupations).
2 Statistical classification of economic activities in the European Community, see:
https://ec.europa.eu/eurostat/web/nace-rev2
8
4 DEFINITION OF ONLINE JOB VACANCY
The analysis of OJVs and the skills specified in them makes it possible to identify – at a given moment
in time – the main skills required by the company for the professional profile demanded. For example,
the skill of ‘programming’ is unlikely to be explicitly mentioned by the employer in an advertisement for
a software analyst as it is deemed to be implicit; meanwhile, the same skill could be made explicit for a
statistician if this is considered as enabling the profession. In this sense, therefore, a vacancy listing
should not be understood as an enumeration of the skills of the professional profile; standard
taxonomies offer a complete and comprehensive description. Instead, the job advertisement must be
understood as a specification of the competences and skills of the profession that are considered
essential for companies when the data is collected or observed. In other words, vacancy listings allow
attention to be focused on the skills required in real time by the labour market, and thus represent a
valuable tool for investigating changes over time in professions and skills in the different dimensions of
analysis.
OJVs, which are job advertisements, contain two main text fields: a title and a full description. The title
briefly summarises the job position, while the full description field usually includes details of the
position and the relevant skills the employee should hold.
Regarding small companies, generally they don’t use web channel to recruit staff, using instead more
traditional channels. Then, there will almost certainly be few OJVs collected regarding those
companies.
The deduplicated/total ratio is a very relevant indicator for the job market on the web for the analysed
countries. Comparison of data for the two countries analysed, namely Tunisia and Ukraine, shows a
9
substantial difference for this indicator, with the latter having a much lower duplication rate. This is
hypothetically indicative of a business difference applied by the different portals: the market on the
Tunisian web, which is presumably in a phase of growth and consolidation, has rather generalist
portals that are transversal to the various sectors and probably have a high percentage of
announcements in common. Meanwhile, the Ukrainian market is more multifaceted, with portals that
over the years have developed specific markets for different business roles and sectors, specialising
and, consequently, reducing duplication. This index, based on a solid selection of sources obtained
through the landscaping phase, is therefore very interesting and is suggestive of a dynamic of maturity
of the job market on the web.
The current period we are experiencing represents a unique and unprecedented moment in time. The
study and analysis of how the pandemic is reflected on the one hand in the
employment/unemployment data extracted from official statistics and on the other in the web data
certainly represents an interesting aspect. Although we do not have up-to-date data that we can use
for our analysis for the former, for the latter it is possible to make initial observations by analysing the
monthly historical series (April–September 2020). April is the month with the lowest number of
advertisements published (2 521), and this corresponds with the most stringent COVID-19 measures
put in place by the Tunisian authorities. There is a significant increase in announcements precisely
corresponding to the gradual withdrawal of the measures, i.e. from 29 April onwards. The significant
increase in the number of announcements starting from May and even more so from June is therefore
entirely logical, following the recovery of various economic activities. It should be pointed out that the
reduced length of the historical series does not allow for clear and irrefutable conclusions to be drawn.
Thus, this growth will have to be verified to understand whether it is an extemporaneous factor linked
to the exit from the lockdown or a seasonal factor that will recur in ‘normal’ conditions. It is therefore
impossible to estimate the dynamics. Nevertheless, it is important to assess and monitor what is
happening in order to understand how the market is adapting to this historic event. In the future, we
will be able to understand better how the dynamics have stabilised, with what growth rates (already
visible today, after the first months of collection and observation) and with what new emerging skills
and professions. In this sense, the freshness and timeliness of the web data allow us to understand
aspects in almost real time, something that is not possible with the official data.
10
6 DATA FROM OFFICIAL STATISTICS
This section provides a summary description of the labour market in Tunisia through the analysis of
the main indicators developed with respect to the available dimensions, starting with official statistics
(http://www.ins.tn/en/statistiques).
In 2019 the active population was over 4 170 000, 71% of whom (over 2 960 000) were male. Figure
6.1 shows the evolution of the active population from 2016 to 2019; there is a continuously increasing
trend, and 2.9% growth over the period3.
3 The active population is defined as the total number of unemployed people plus the employed civilian workforce.
11
The unemployment rate is high, especially for young people and women.
The overall unemployment rate in 2019 was 15.2%, a slight decrease on the previous year, when it
was 15.5%. With regard to gender, the female unemployment rate, at 22.2% in 2019, was significantly
higher than the male rate, at 12.3%, with a difference of about 10 percentage points 4.
The gap between the male and female employment rates is the result of discriminatory attitudes on
the part of employers, based on sociocultural values. In the provinces, the high rate of unemployment
among women may also be caused by their lower mobility, which prevents them leaving their home to
apply for a job, and by the distance to the nearest job centre. A further factor is that women often
study for qualifications that are less in demand on the labour market. Overall, women work in lower-
skilled jobs than men with the same level of qualification.
Female entrepreneurs face many obstacles in Tunisia, the most significant being: i) cultural barriers to
starting a business; ii) a lack of collateral for bank loans, since women own little property in their own
right; and iii) the general scarcity of women in the working population and in executive posts (OECD,
2018).
4
The unemployment rate is calculated as the number of unemployed people as a percentage of the active
population. Unemployed people are those of working age who are without work, who are available for work and
who have taken active steps to find work.
12
With regard to age groups, in 2019 the unemployment rate was 34.4% for young people aged 15–24,
falling to 12.7% for those aged 25–59 and finally to 1.9% for those over the age of 60.
Figure 6.5 shows that the unemployment rate for graduates was 28.2% in 2019 (higher than the
overall unemployment rate, 15.2%); the female unemployment rate is the most worrying, reaching a
value of 38.3%, compared with 16.4% for the male unemployment rate.
Figure 6.5. Unemployment rate for higher education graduates, 2016–2019 (%)
13
6.2 Employment
There has been a positive and increasing trend in the number of employees, the number having risen
from over 3 423 000 in 2016 to over 3 539 000 in 2019, an increase of 3.4%.
The gender composition of the employed population shows a clear prevalence of men: the share was
73% in 2019, and this has remained constant over the period observed.
Gender gaps are smaller than in other MENA countries. However, the employment rate is much lower
for women than for men, and women often have less skilled jobs. Policies that promote women’s
participation in the labour market and their employment, and that provide better guidance on training
courses that help to ensure employment, should be implemented.
14
The female employment rate is one of the highest in the Middle East and North Africa (MENA) region,
but it still falls well short of that of the OECD (23% versus 63%). The proportion of women holding
management positions in the public and private sectors in Tunisia is the highest among MENA
countries, at 14.8% (OECD, 2018).
The data from the official statistics offer a further level of detail, allowing us to analyse which sub-
sectors show the highest and lowest employment rates in Tunisia.
Figure 6.8 shows the percentage of employees in the various sub-sectors of the services sector.
15
• Manufacturing industries (18% of the employed population)
Figure 6.9 shows the shares of employees in the various sub-sectors of the manufacturing sector.
For this sector, almost all employees are in the construction and public works sub-sector, which
accounts for 93% of the total employed population.
16
6.3 Conclusions
The analysis of data from official statistics in Tunisia allows us to extract the following evidence.
1. The active population in 2019 was over 4 170 000, 71% of whom were male. The trend was
positive from 2016 to 2019.
2. The unemployed population in 2019 was over 631 000, down by 1.4% (over 8 700 fewer
people) on the number in 2018.
3. The unemployment rate is high, especially for young people and women. Female
unemployment was 22.2% in 2019, compared with a rate of 12.3% for men. The gap between
the male and female employment rate is the result of discriminatory attitudes on the part of
employers, based on sociocultural values. A further factor is that women often study for
qualifications that are less in demand on the labour market. Overall, women work in lower-
skilled jobs than men who have the same level of qualification.
5. The female employment rate is one of the highest in the MENA region, but it still falls well
short of that for the OECD (23% versus 63%). The proportion of women holding management
positions in the public and private sectors in Tunisia is the highest among MENA countries, at
14.8% (OECD, 2018).
6. In 2019 more than half of the population (52%) was employed in the services sector, followed
by the manufacturing sector (18%), non-manufacturing industry (16%), and finally the
agricultural sector (14%).
17
7 USE OF THE INTERNET
Micro and small firms are unlikely to recruit through formal employment agencies or using the internet.
Their vacancies are hard to classify by type and level of skill. However, they may have computers and
mobile phones that allow them to access the internet. The existing data show that the use of the
internet will not be a major barrier to monitoring job vacancies and the labour market online. According
to a recent survey conducted in 2019 by the National Telecommunication Agency (the Tunisian
telecommunication regulator), two thirds of mobile phone owners use their phones to access the
internet (75% of women and 90% of young people). Overall, almost 70% of individuals use the internet
(80% of women and 90% of young people). The Internet Users Statistics for Africa in 2018 gives a
smaller number, 66.8% instead of 70%. The difference could reflect growth from 2018 to 2019.
Compared to the data for Africa and to the world average (58.8%), Tunisia scores relatively well.
Individuals use the internet for a number of purposes, both personal and professional; the survey
results show that 45% of individuals use the internet at work. This percentage reflects quite
accurately the depth of internet penetration in the production system, and it may be explained by the
size of the informal sectors: about half of labour is in the informal sector where the use of the internet
is limited.
Table 7.1. Proportion of enterprises using the internet in Tunisia by sector, 2015 (%)
Sector %
ICT 99.2
Specialised, scientific and technical activities 98.5
Education 98.5
Other services 97.4
Real estate 94.2
Other administrative services 94.0
Health and social services 90.9
Mining 87.4
Commerce and car maintenance 86.9
Transports and logistics 85.0
Manufacturing 82.8
Construction 81.9
Arts and recreation 80.2
Utilities (electricity, gas, water) 77.7
Hotels and restaurants 61.1
Total 84.7
18
8 ONLINE JOB VACANCIES
The number of OJVs published on the web for Tunisia in the observed period, from 1 April to 30
September 2020, was 37 535.
The source that published the highest number of job advertisements was Tanitjobs, with over 19 000
and a share of 52%, followed by Jora with around 5 500 (15%), Keejob with over 4 800 (13%) and
emploi.nat with over 2 200 (6%).
Analysis of the historical series allows us to identify, where present, the seasonal factors that
characterise the territory analysed and detect any critical issues. However, it should be remembered
that the shortness of the historical series does not allow us to draw clear and irrefutable conclusions,
but only to identify the inputs to be kept in mind as the observed period progresses. April was the
month in which the lowest number of advertisements were published (2 521).
This data can be attributed to the effects of the COVID-19 pandemic. The Tunisian authorities
announced on 29 April that the measures aimed at curbing the spread of the virus would be gradually
revoked starting from 4 May, thus putting an end to the lockdown. With the measures loosened, parts
of the food and construction sectors resumed operations, as did government employees and public
transport services, at least partially. The restrictions were then further relaxed on 11 May with the
reopening of clothing stores and shopping centres. It would therefore be logical to expect a significant
increase in announcements starting from May, and even more so from June, following the return of
various economic activities. However, this growth will have to be verified when there is a longer
historical series available in order to understand whether it is an extemporaneous factor linked to the
exit from the lockdown or a seasonal factor that will recur in ‘normal’ conditions.
However, the overall trend of advertising on the web is growing, with web channels increasingly used
by companies to advertise profiles that are difficult to fill and/or are highly specialised, and not always
reachable by standard recruitment channels.
As shown in Figure 8.1, the number of advertisements grew by 30% from April to May, increased by
138% from May to June, suffered a slight decline from June to July (−13%), and then returned to
growth in August, reaching the highest value during this period, with almost 9 000 advertisements.
19
8.1 Occupation
The type of profession demanded on the labour market in OJVs Tunisia is one of the most important
dimensions. The highest number of advertisements were for highly specialised levels, that is,
professionals, with 34.2% of the total, followed by technicians and associate professionals with
approximately 26.63%, service and sales workers with 12.14% and clerical support workers with
11.13%. In last position in terms of the number of advertisements posted on the web are elementary
occupations, professions with a low level of specialisation, with a share of 3.21%. Thus, the analysis
shows that companies looking for staff use the web to search for high- and medium-specialisation
profiles, while they almost certainly resort to more ‘standard’ channels to search for low-specialisation
profiles.
An analysis of professions by skill level shows that the profiles in most demand on the web are those
that are highly skilled. High-skill profiles account for 68.1% of the total advertisements, followed by
medium-skill profiles with 26.5%, and finally low-skill profiles with only 5.4%.
20
The profession most requested in Tunisia during the period observed is software developer, with over
2 500 advertisements. This profile has a high level of specialisation and belongs to the category of
professionals. It is followed by contact centre salesperson with over 1 600 advertisements and
administrative and executive secretary with over 1 400 advertisements.
For August, which saw a peak in the number of advertisements, the analysis of professions reveals a
slight reduction in the share of professionals and managers, and an increase for service and sales
workers (13.27%, compared with 12.14% overall) and those with a low level of specialisation.
Considering the professions in most demanded, in August the first two positions were held by software
developer and contact centre salesperson; in third position was information and communication
technology user support technician. The profession of shop sales assistant ranked fifth compared to
ninth in the overall ranking for the period as a whole; this rise could be linked directly to the
21
seasonality of August, when there is an increase in tourism and therefore in the search for professions
linked to this economic activity.
Meanwhile, the majority of announcements requiring a master or equivalent qualification are aimed at
professionals, with a share of over 59%; the same is observed for announcements that require a
bachelor or equivalent, with professional profiles accounting for 56%, while the demand for
announcements aimed at managerial profiles increases for this specific education requirements
(10.91%).
In contrast, the proportion of advertisements that mention primary education is not very significant,
with a value of only 1.48%; this demonstrates that employers’ research and recruitment for these
22
profiles takes place using other channels, as the official statistics indicate that such profiles are very
significant for the country.
8.3 Experience
Analysis of the previous professional experience dimension reveals that a significant share of
advertisements (22.25%) do not specify the years of experience required.
In general, OJVs ask for previous experience of 2–4 years (22.36%), followed by profiles with little
quantifiable experience (less than 1 year) (16.13%) and 1–2 years’ experience (11.13%); the share of
advertisements aimed at those who have no experience is significant, at 7.62%.
Overall, profiles requiring significant previous experience (over 2 years) account for a total of 36.65%
of the advertisements analysed and the share of those requiring more than 10 years of experience is
6.2%. For the latter, it is observed that these advertisements, in addition to calling for such a
significant amount of experience, ask for a master or equivalent qualification in 35.88% of cases and
relate to the role of manager in 13.92% of cases.
23
Analysis of the relationship between profession and type of contract indicates that a larger share of
permanent than temporary contracts are for highly specialised profiles; 42.4% of permanent contracts
are for professionals, compared to 31.5% of temporary contracts, and 9.5% of permanent contracts
are for managers, compared to 6.5% of temporary contracts. In contrast, low-specialisation profiles,
and in particular elementary occupations, account for 3.0% of temporary contracts compared to 1.3%
of permanent ones.
Summarising with regard to the level of specialisation, 75.4% of the advertisements offering
permanent employment contracts are for the high level of specialisation, while 67.8% of temporary
contracts relate to this level. A larger proportion of temporary contracts than permanent contracts are
at the medium and low levels of specialisation.
24
Figure 8.11. OJVs by contract and skill level
8.5 Industry
The services sector accounts for the most vacancies on the web, with an 84% share of
advertisements in the observed period, followed by the manufacturing sector (16%) and finally the
agricultural sector with an insignificant share (0.2%).
Comparing the web data with the official statistics data – in terms of employees by sector –
immediately reveals that the web channel underestimates employment in the agricultural sector: in
fact, this sector is very significant, accounting for 14% of the workforce. The same is observed for the
manufacturing sector: in terms of employed population, it accounts for 34%, although the web channel
is used only for 16% of published advertisements. The opposite is the case for the services sector:
OJVs are crucial in recruiting personnel in this sector, representing 84% of all advertisements, while it
accounts for 52% of the employed population. It should be clarified that the intention of this analysis is
to capture macro evidence from the comparison of the available sources, keeping in mind both the
diversity of their characteristics and also the different time periods compared (web data is updated to
2020, while the official data refers to 2019). In this sense, OJVs become a complementary tool for the
analysis of data from official sources, with no expectation of a perfect match between the two types of
data. Instead, the depth and freshness of data from OJVs can be exploited alongside the consistency
of the official data.
25
Analysis of the advertisements in the various sub-sectors allows us an even better understanding of
what was previously observed in terms of under- or over-estimation of data from the web compared to
official employment statistics.
On the web, the largest share of advertisements are for jobs in the administrative and support service
activities sector, with 27% of the total advertisements, followed by the information and communication
sector with 15%, the professional, scientific and technical activities sector and manufacturing, with
12% each, and financial and insurance activities with 7%. It can be observed that precisely because of
the distinctive character of the web channel, there is little recruitment of staff for the public sector (only
6% of advertisements), which typically seeks personnel through public selection calls; the situation is
similar for the education sector (2%). There is also little demand for accommodation and food service
activities (3%), despite this sector having a very high turnover linked to seasonal factors; this suggests
that OJVs are not the preferred channel for recruiting personnel in the sector, but that more traditional
channels such as word of mouth are favoured.
26
The sector with the largest number and share of advertisements (27%), administrative and support
service activities, includes searches for the professions shown in Figure 8.14, which are consistent
with the nature of the sector. In particular, accounting associate professionals is in the first position,
with over 700 advertisements.
Figure 8.14. OJVs for the administrative and support service activities sector
In second position in terms of number of advertisements, with a 15% share of the total, is the
information and communication sector. This confirms that the job market is in rapid transformation and
is increasingly oriented towards the search for digital profiles, which are becoming increasingly
strategic and in demand in all regions and businesses. With regard to the most in-demand profiles in
this sector, in first position is software developer with about 700 announcements in the information and
communication sector alone; in general, this is the most in-demand profession on the web, regardless
of the sector analysed, with over 2 500 advertisements in the period observed, confirming that digital
professions are also important in non-ICT sectors.
Analysis of the distribution of employees in the various sub-sectors shows that the largest sector for
employment in Tunisia is education, health and administration services, with 19%, in contrast to the
situation for OJVs, where the education sector represents only 2% of the announcements, as
explained above.
27
8.6 Skills
The skills required in advertisements can be considered as distinguishing: it’s not common that an
advertisement includes basic skills for a specific profession, those are implicit in the definition of the
specific occupation required. This dimension is fundamental, as it allows the detection of market
trends and any so-called ‘emerging’ professions, and the characteristics they have in terms of
knowledge/skills, in order to allow adequate vocational programming starting from training courses,
and thus to avoid the phenomenon of mismatching.
The importance of the information and communication sector in terms of job advertisements published
on the web is confirmed by the ever-increasing demand for skills in the digital field. Digital skills are
now a strategic factor for the competitiveness of the socioeconomic system, but awareness of their
importance is still limited, as are the skills, among companies, public administrations and citizens.
While technology lends itself to being a great support to increasingly complex decision-making
processes, it is also necessary to work to strengthen those skills that can never be replaced, such as
critical thinking and emotional intelligence. The challenge, therefore, is to make the educational and
training offer consistent with changes in the knowledge society, accompanying the process of
technological innovation without being engulfed by technologies, but by governing them.
With regard to the knowledge and skills requirements in advertisements aimed at the Tunisian territory
in the time frame concerned, 68.75% of advertisements are looking for applicants who have
knowledge in the field of economics and management; knowledge relating to the science and
technology sector is included in 42.14% of advertisements, humanities in 38.94% and, finally,
industrial knowledge in 20.9%.
28
In terms of personal attitudes, in first place in terms of frequency of advertisements is
adaptability/flexibility (required in 77.68% of the advertisements), followed by responsibility, requested
in 52.83% of the announcements, and monitoring and leadership in 42.16% of cases. Lastly, and
mentioned in only 5.89% of advertisements, is enthusiasm.
The most in-demand skill relates to languages, which is explicitly mentioned in 47.15% of
advertisements, closely followed by teamwork (41.94%). Basic computer skills are mentioned in
40.98% of the advertisements.
Finally, with regard to tools and technology, in first position is Office suite software, which is required in
68.59% of advertisements; this confirms that digital skills are now also transversal to professions not
strictly connected to the information and communication sector.
We also note the significant share of profiles mentioning programming and scripting languages, a
requirement in 22.98% of advertisements.
29
The distribution of the skills/knowledge required is clearly linked to the required profession.
As previously mentioned, the profession in most demand in Tunisia during the observed period is
software developer, with over 2 500 OJVs, and requests for knowledge/skills for this profession are
reported in Figure 8.21. It is immediately evident that in terms of knowledge, science and tech ranks
first (requested in 95.08% of the advertisements). In terms of personal skills, creativity and innovation
appears to be valued, and is in second position here, compared with fourth position in the overall
figure when it is not linked to the profession. Meanwhile, knowledge of languages is of little importance
for software developers; instead, ICT development is in first place, appearing in 95.46% of
advertisements. Finally, in tools and technology, first position goes programming and scripting
languages, which is requested in 79.27% of the advertisements for software developers.
Figure 8.22 gives another example to further demonstrate how the demand for and importance of
different skills/knowledge changes according to the profession selected; the high quality of the
extracted and processed data is also evident from the excellent coherence between the
skills/knowledge required and the profession.
The shop sales assistant profession had over 1 000 OJVs in the observed period. It is immediately
evident that in terms of personal qualities, the most important characteristic is responsibility, which is
30
required in 70.93% of the relevant advertisements; in terms of fundamental skills, sales and marketing
activities is in the top position and is required in 79% of the advertisements; finally, knowledge of
Office suite software tools is important, and is required in 87.5% of cases, further confirmation of the
key role of transversal digital skills to all sectors and profiles.
8.7 Conclusions
Compared with traditional employment channels (such as employment services, recruitment agencies,
mass media, friends and acquaintances), online job portals can provide a broader information base
and more efficient ways to find a job, and give a better and more complete understanding of the
situation on the labour market and the trends in its development. This can strengthen state policy in
the field of employment and training, offer more effective tools for the work of the state employment
service, provide better signals for the education system and additional education for adults, and
facilitate the integration of socially vulnerable people into the full life of society.
1. The number of advertisements on the web is growing, demonstrating that OJVs are
increasingly used by companies to recruit employees for profiles that are difficult to fill and/or
highly specialised, and not always reachable by standard recruitment channels.
.
2. Companies looking for staff on the web use it to recruit employees for high- and medium-
specialisation profiles, while they almost certainly use more ‘standard’ channels to search for
low-specialisation profiles. Advertisements for professionals represent a 34% share of the total
advertisements, followed by technicians and associate professionals with about 27%. In
summary, with respect to the level of specialisation, high-level profiles represent 68.1% of the
total advertisements, while low-level profiles account for only 5.4%.
31
3. In Tunisia the most in-demand profession is software developer, with over 2 500
advertisements, confirming that the job market is rapidly changing and is increasingly oriented
towards digital profiles, which are becoming increasingly important strategically and are in
demand in all regions and businesses. The information and communication sector is in second
position in terms of the number of advertisements.
4. A significant number of job advertisements on the web relate to profiles with a master or
equivalent qualification, the share being 25.75%; in contrast, profiles requiring primary
education are not very significant, with share of less than 2%.
5. Services is the sector that looks for the most personnel on the web, with an 84% share of
advertisements in the observed period, followed by the manufacturing sector with 16% and
finally the agricultural sector with an insignificant share (0.2%). Compared with the official
statistics, OJVs seem to underestimate the importance of the agricultural sector, which is very
significant for Tunisia as it employs 14% of the workforce.
7. Comparing required skills and the demanded profession in the advertisements, it’s possible to
identify a strong correlation: skills are strictly related to the specific occupation, confirming the
quality of the analysed data.
32
ANNEX – WEBSITE CHARACTERISTICS
Rough position in the Google ranking: the rough position of the website in the Google ranking list
resulting from the queries ‘emploi + name of the country’ and ‘job + name of the country’. The value
can be either first (second) page, which means the website appears in the first (second) page of the
Google results, or ‘other’, to represent that the web page is listed from the third page onwards.
Type of job portal: defines whether the website is a primary job portal, a secondary job portal or a
combination of job portal and secondary functions.
Type of operator: refers to the typology of the website, i.e. whether the website is related to a
recruitment agency (e.g. GiGroup) or to a national newspaper (e.g. the Jobs section of the Guardian
website), whether it is a specialised website (e.g. Monster), a public, sectoral or company website or a
classified ads portal.
OJV volume (approximate number of OJVs): the number of vacancies included on the website at the
time of the analysis.
Geographical scope: defines whether the source is regional or national (e.g. the Czech portal), or
whether it has an international dimension (e.g. Monster is almost worldwide).
Sectoral scope: defines whether the website refers to only one sector or to the whole labour market
(defined as ‘one industry’ vs ‘all industries’).
Publication date of OJV: indicates whether or not the publication date of the vacancy is present.
Update frequency: indicates the frequency of update of the sources (‘daily’ or ‘not daily’).
OJV characteristics
Type of contract: defines whether the type of contract in the vacancy description text is structured,
textual or not available.
Working time: defines whether the working time in the vacancy description text is structured, textual
or not available.
Sector: defines whether the sector in the vacancy description text is structured, textual or not
available.
City: defines whether the city in the vacancy description text is structured, textual or not available.
District: defines whether the district in the vacancy description text is structured, textual or not
available.
Region: defines whether the region in the vacancy description text is structured, textual or not
available.
33
Qualification level: defines whether the qualification level in the vacancy description text is
structured, textual or not available.
Wage: defines whether the wage in the vacancy description text is structured, textual or not available.
Language: lists the language used on the website (not included in values).
1. Keejob https://www.keejob.com/
Occupation: Textual
Sector: Structured
City: Structured
District: Structured
Region: Structured
Language: French
2. Tanitjobs https://www.tanitjobs.com/
34
Type of job portal: Primary job portal
Occupation: Textual
Sector: Structured
District: Structured
Region: Structured
Language: French
Occupation: Textual
35
Type of contract: Textual
Sector: Structured
Region: Structured
4. Tunisie-emploi https://www.tunisie-emploi.tn/
Occupation: Textual
Sector: Structured
City: Various
District: Various
Region: Various
36
Language: French + Arabic
5. ReKrute https://www.rekrute.com/
OJV volume: 75
Occupation: Textual
Sector: Structured
City: Various
District: Structured
Region: Structured
Language: French
6. Farojob https://www.farojob.net/
37
Sectoral scope: All industries
Occupation: Textual
Sector: Structured
Region: Structured
Language: French
7. Jora https://tn.jora.com/
Occupation: Textual
Sector: Various
City: Various
38
District: Various
Region: Various
8. Offre-emploi https://www.offre-emploi.tn/
Occupation: Textual
Sector: Structured
Region: Structured
Wage: Textual
Language: French
9. Jobi https://www.jobi.tn/#!/
39
Type of job portal: Primary job portal
Occupation: Textual
Sector: Structured
City: Structured
District: Structured
Region: Structured
Wage: Structured
Language: French
Occupation: Textual
40
Type of contract: Textual
Sector: Structured
City: Structured
District: Structured
Region: Structured
Wage: Various
Language: French
Occupation: Structured
Sector: Structured
Region: Structured
41
Language: French
Occupation: Textual
Sector: Structured
City: Structured
District: Structured
Region: Structured
42
Sectoral scope: All industries
Occupation: Textual
City: Structured
District: Structured
Region: Structured
Language: French
OJV volume: 33
Occupation: Textual
43
District: Not available
Region: Structured
Language: Arabic
OJV volume: 60
Occupation: Textual
Sector: Structured
Region: Structured
Wage: Structured
44
Type of job portal: Primary job portal
Occupation: Textual
Sector: Structured
Region: Structured
45
TABLE OF ABBREVIATIONS
AI Artificial intelligence
TC Text classification
46
REFERENCES
Boselli, R. et al., ‘WoLMIS: a labor market intelligence system for classifying web job vacancies’,
Journal of Intelligent Information Systems, Vol. 51, No. 3, 2018, pp. 477–502.
Boselli, R., Cesarini, M., Mercorio, F. and Mezzanzanica, M., ‘A model-based evaluation of data
quality activities in KDD’, Information Processing & Management, Vol. 51, No. 2, 2015, pp. 144–166.
Boselli, R., Cesarini, M., Mercorio, F. and Mezzanzanica, M., ‘Classifying online Job Advertisements
through Machine Learning’, Future Generation Computer Systems, Vol. 86, 2018, pp. 319–328.
Cedefop, Real-time Labour Market information on skill requirements: Feasibility study and working
prototype, Cedefop Reference number AO/RPA/VKVET-NSOFRO/Real-time LMI/010/14, Contract
notice 2014/S 141-252026 of 15/07/2014, 2014. Last accessed March 2019 at: https://goo.gl/qNjmrn
Cedefop, Real-time Labour Market information on skill requirements: Setting up the EU system for
online vacancy analysis AO/DSL/VKVET-GRUSSO/Real-time LMI 2/009/16. Contract notice - 2016/S
134-240996 of 14/07/2016, 2016. Last accessed March 2019 at: https://goo.gl/5FZS3E
Colombo, E., Mercorio, F. and Mezzanzanica, M., ‘Applying machine learning tools on web vacancies
for labour market and skill analysis’, in Terminator or the Jetsons? The Economics and Policy
Implications of Artificial Intelligence, 2018.
ESCoE (Economic Statistic Centre of Excellence), ‘Using administrative and big data to improve
labour market statistics’, 2019. Last accessed March 2019 at:
www.escoe.ac.uk/projects/usingadministrative-big-data-improve-labour-market-statistics/
Fayyad, U., Piatetsky-Shapiro, G. and Smyth, P., ‘The KDD process for extracting useful knowledge
from volumes of data’, Communications of the ACM, Vol. 39, No. 11, 1996, pp. 27–34.
Institut National de la Statistique (INS) 2017: ‘Indicateurs fondamentaux sur l’utilisation des
technologies de l’information et de la communication par les entreprises en
2015’. http://www.ins.tn/fr/publication/indicateurs-fondamentaux-sur-l’utilisation-des-technologies-de-
l’information-et-de-la-1
Mercorio, F. and Mezzanzanica, M., ETF, ‘Feasibility Study for Tunisia and Morocco to identify,
validate and rank web job vacancy sources – practical guidance’, ETF, unpublished, 2019.
Mezzanzanica, M. and Mercorio, F., ‘Big Data Enables Labor Market Intelligence’, in Encyclopedia of
Big Data Technologies, Springer International Publishing, 2018, pp. 1–11.
Mezzanzanica, M. and Mercorio, F., ETF (European Training Foundation), ‘Big Data for labour market
intelligence: An introductory guide’, ETF, 2019.
47
Mezzanzanica, M., Mercorio, F. and Colombo, E., ‘Digitalisation and Automation: Insights from the
Online Labour Market’, in Larsen, C., Rand, S., Schmid, A. and Dean, A. (eds.), Developing Skills in a
Changing World of Work, Rainer Hampp Verlag, 2018, p. 259.
Penneck, S. et al., ‘Using administrative data for statistical purposes’, Economic & Labour Market
Review, Vol. 1, No. 10, 2007, p. 19.
Sebastiani, F., ‘Machine learning in automated text categorization’, ACM Computing Surveys, Vol. 34,
No. 1, 2002, pp. 1–47.
Vankevich, A., ETF (European Training Foundation), ‘Landscaping of the web labour market in
Belarus, and ranking of online job vacancy sources’, September 2020.
48