Digital Requirements for Systems Analysts in Europe: A Sectoral Analysis of Online Job Advertisements and ESCO Skills

Abstract

Systems analysts can be considered a valuable part of organizations, as their responsibilities and contributions concern the improvement of information systems, which constitute an irreplaceable part of organizations. Thus, by exploring the current labor market of systems analysts, researchers can gather valuable knowledge to understand some invaluable societal needs. In this context, the objectives of this study are to investigate the sets of digital skills from the European Skills, Competences, Qualifications, and Occupations (ESCO) taxonomy required by systems analysts in Europe and examine the key characteristics of various relevant sectors. For this purpose, a tool combining topic extraction, machine learning, and statistical analysis is utilized. The outcomes prove that systems analysts may indeed possess different types of digital skills, where 12 distinct topics are discovered, and that the professional, scientific, and technical activities demand the most unique sets of digital skills across 17 sectors. Ultimately, the findings show that the numerous sectors indeed have divergent requirements and should be approached accordingly. Overall, this study can offer valuable guidelines for identifying both the general duties of systems analysts and the specific needs of each sector. Also, the presented tool and methods may provide ideas for exploring different domains associated with content information and distinct groups.

1. Introduction

Systems analysts are responsible for investigating client data and developing plans to improve information systems (https://esco.ec.europa.eu/en/classification/occupation_main (accessed on 8 February 2025)). Job positions belonging to this category are related to some popular occupations, including data analysts, data scientists, artificial intelligence engineers, as well as green and Information and Communication Technology (ICT) consultants. In total, data science and scientists are associated with applications covering different domains and sectors such as banking, healthcare, sports, and marketing [1], meaning that systems analysts can be occupied in diverse areas of interest as well. As a result, systems analysts should be considered an important job category for supporting valuable infrastructures and services. Due to the plethora of technical content and various ICT principles surrounding relevant jobs and organizations, it is evident that systems analysts should possess various skills, which may concern project management, databases, system development, and relevant tools and programming languages, e.g., SQL and Microsoft Office [2]. Complementary to hard and technical skills, jobs of this type can also require soft skills related to analysis and problem-solving, communication, and teamwork [3].

To study skills concerning a unique occupation or job category, like systems analysts, online job advertisements are the primary data type among other potential resources, e.g., surveys and interviews. For this purpose, a variety of techniques are used for extracting skill analytics, where Natural Language Processing (NLP), statistical methods, machine learning, and graph theory are some of the usual choices [4,5,6,7]. Among the many use cases of skill information, one of the prevalent relevant objectives concerns skill analytics. In experiments of this type, skills within online job advertisements are studied for evaluating patterns, via skill frequencies and co-frequencies, to address various research directions, such as skill demand and gaps [4], and group characteristics, e.g., types of skills–competences [5] and regions [6,7].

Since systems analysts constitute an important part of organizations from different domains and variant scopes, both the research and industrial communities may benefit from insights covering the relevant skills demand and respective requirements of the various sectors. Thus, the current study addresses the existing research gaps concerning the current skill demand of systems analysts in Europe as well as the key characteristics of the different sectors associated with jobs of this type. Overall, we aim to provide answers to three Research Questions (RQs) reflecting the aforementioned objectives. To do so, an RShiny tool (https://github.com/koncharman/ClickTMtool (accessed on 13 February 2025)) that was developed as part of an EU-funded project (https://skillab-project.eu/ (accessed on 17 February 2025)) is utilized. This tool offers functionalities for conducting content analysis and group comparisons via data importing, topic extraction, statistical analysis, and machine learning.

To fulfill the goals of this study, a dataset comprising online job advertisements from EUROSTAT (https://ec.europa.eu/eurostat (accessed on 19 February 2025) of systems analysts is collected, where each observation is associated with a set of digital skills (201 overall) and a unique sector (17 distinct sectors are investigated). These skills are annotated based on the European Skills, Competences, Qualifications, and Occupations (ESCO) taxonomy. Next, the main areas of interest (or topics) surrounding these skills are evaluated through skill frequencies and cluster analysis. Furthermore, the similarities and differences between the sectors are detected through statistical analysis and machine learning, where, in the end, the most unique sectors are discovered as well. By extending the previous analysis, the final procedures of the case study include the identification of the key characteristics of each sector along with the key sectors per topic. To this end, the relative frequency of digital skills and skill topics are considered to track unanticipated frequencies and potential strong relations between sectors and concepts. In multiple cases, significant relative frequencies are uncovered and can be studied as baselines for investigating specific skill groups or sectors according to the preferences of a third party, e.g., individuals or organizations.

The main contribution of this study is the provision of insights on the primary skills required from systems analysts in general and within sectors, which can prove useful to both organizations and individuals for multiple purposes. These purposes can be related to creating learning material and pathways, organizing career paths, evaluating skill gaps, and identifying suitable job candidates. Nonetheless, multiple studies focus on specific occupations, like data scientists [1,8] and data analysts [9], instead of gathering insights concerning the job category of systems analysts. Also, while labor market analytics is considered a frequent concept, more specific research directions like sectoral analysis combined with skill-based investigations can be considered a less common subject. Thus, the current study may also provide valuable ideas for future research that can be expanded to other data properties apart from sectors, e.g., countries and occupations.

The remainder of this study is organized as follows: Section 2 contains the related work associated with projects combining the ESCO taxonomy with online job advertisements. Section 3 first presents the RQs and dataset of this study (Section 3.1) as well as the tool and individual methods used to provide answers to these RQs (Section 3.2), i.e., preprocessing (Section 3.2.1) and analysis (Section 3.2.2). Section 4 demonstrates the results of this study, which are organized according to the main RQs. Section 5 provides extensive discussion about the insights, potential use cases, and contributions of the presented methods and findings. Finally, Section 6 concludes the paper and proposes ideas for future work. A list of all abbreviations in the paper can be found in the Appendix A, in

Table A1. Abbreviations.

Abbreviation	Description	Abbreviation	Description
CAE	Computer-Aided Engineering	NLP	Natural Language Processing
CARET	Classification and Regression Training	NMF	Non-Negative Matrix Factorization
CART	Classification and Regression Tree	NPMI	Normalized Pointwise Mutual Information
DTM	Document Term Matrix	RF	Random Forest
ESCO	European Skills, Competences, Qualifications, and Occupations	RQ	Research Question
FKM	Fuzzy K-Means	TCM	Term Co-occurrence Matrix
GBM	Gradient Boosting Machine	TF	Term Frequency
GloVe	Global Vectors algorithm	TF-IDF	Term Frequency–Inverse Document Frequency
GMM	Gaussian Mixture Model	UMAP	Uniform Manifold Approximation Projection
ICT	Information and Communication Technology	XGB	Extreme Gradient Boosting
LDA	Latent Dirichlet Allocation

.

2. Literature Review

This section presents the related work of the current study concerning the potential use cases of the ESCO taxonomy when it comes to online job advertisements and skills. In general, the information of the ESCO taxonomy is often combined with many data sources to satisfy variant purposes, apart from online job advertisements. These data types may include other relevant taxonomies [4], resources associated with the labor market demand and supply, e.g., courses [4,10,11], CVs [10,11], information and innovations from the literature [12], and surveys and interviews [13].

Regarding the main objectives of projects exploring online job advertisements, the ESCO taxonomy can be leveraged as a valuable knowledge base towards extracting the main skills from online job advertisements with the use of NLP techniques [4,10,11,14]. By conducting experiments of this type, researchers can study the extracted skills and address various objectives related to skill demand [15] and skill gaps [14]. When it comes to skill gaps and matching, prior studies also extract skills within online job advertisements and other types of data, e.g., CVs [10,11] and courses [4,10,11], to evaluate individuals or potentially improve learning and training material.

Similarly, existing projects are oriented to assist individuals by developing systems and methods with skill or job recommendation mechanisms [16,17], which are some ideas that can indeed offer meaningful guidelines and elevate the labor market. Studies of this nature aim at the improvement of individuals, i.e., usually students or employees, by evaluating employees as well as examining job requirements through skill extraction, to ultimately build job profiles. Complementary to improving employee performance, experiments of this type may assist the planning procedures of organizations by detecting the most suitable candidates and creating effective courses concerning job requirements.

One other relevant area of interest is associated with existing taxonomies, which are investigated along with the textual information and skills of online job advertisements. The common task in this case is to utilize the ESCO taxonomy individually or with other ontologies to develop or refine taxonomies/ontologies by analyzing and extracting information from external data sources, including online job advertisements [18,19,20,21]. The potential changes may concern a variety of purposes, including the assessment of new occupations [19,20] and the improvement of recommendation systems [21], which can lead to establishing new data structures capturing the status of the labor market more accurately.

Regarding data analytics, which is closer to the concepts and outcomes of this study, Giabelli et al. [6] show that by evaluating the skills and occupations derived from online job advertisements, researchers can extract valuable insights for addressing gaps between a plethora of entities (countries in this case). In this context, Kahlawi et al. [7] investigate and compare the skill sets required by different occupations and regions and indeed capture some significant similarities and divergences between these different entities. Using similar approaches, Mankevich and Svahn [15] develop a framework for categorizing job posts with 135 digital competencies to ultimately compare different business units and recruitment types, as well as address the overall skill demand in an automotive company.

Overall, the existing studies show that the skills within online job advertisements can be studied at various levels to uncover similar and distant entities, e.g., regions, occupations, and sectors, and eventually address meaningful research gaps. In addition, researchers have dedicated some of their work to exploring specific types of skills, e.g., digital skills [14], to offer information covering specific areas of interest. In total, the two aforementioned observations support the significance of the current study when it comes to investigating digital skills within different sectors, as previous works investigated digital skills along with various properties characterizing online job advertisements.

3. Materials and Methods

In this section, the main characteristics and goals of the case study towards demystifying the key digital skills and skill topics of systems analysts are presented (Section 3.1). Overall, the scope of this study is to address the digital skill demand of systems analysts in Europe and examine the key characteristics of 17 sectors by addressing their similarities, dissimilarities, and uniqueness. At the same time, the main functionalities of the developed RShiny tool that were utilized for the case study are demonstrated as well (Section 3.2).

3.1. Case Study on Systems Analysts

This study aims to provide answers to three Research Questions (RQs), which indeed reflect the scope and objectives of the paper. The first objective of this study concerns the identification of the main areas of interest associated with the digital skills of systems analysts by examining skill frequencies and skill clusters–topics (RQ₁). The next objective concerns the identification of unique sectors according to the mixture of digital skills required from each one with the use of statistical analysis (RQ₂). In addition, the uniqueness of each sector is investigated through machine learning models, which are trained for classifying the observations of two groups in general, effectively (RQ₂). Going beyond the distinction of sectors, the final objective of this study is to evaluate significant relations between the investigated sectors and skills at both individual and collective levels (RQ₃). With respect to these objectives, the main RQs of this study are the following:

RQ₁: What are the primary digital skills and skill topics of systems analysts?

RQ₂: Which sectors can be distinguished from other sectors?

RQ₃: How are the different types of skills associated with each sector?

To accomplish these objectives and provide sufficient answers to the three RQs, a dataset comprising 81,432 European online job advertisements related to systems analysts was collected. The main information concerning the online job advertisements is provided by EUROSTAT in collaboration with CEDEFOP (https://www.cedefop.europa.eu/ (accessed on 23 February 2025)) via the Web Intelligence Hub and covers the EU27 Member States, EFTA countries, and the United Kingdom.

The primary features of the online job advertisements are the digital skills required to cover each job (201 overall), which are matched against the ESCO taxonomy, and the sector (or economic activity) associated with the job position and publisher. As part of the process, two data structures were established. The first one contains a data frame storing the sector and a new artificial identifier for each observation, while the second one is a binary matrix storing the information about 201 unique digital skills that were found in the job advertisements, where each observation was already annotated by the provider. Each cell of the binary matrix denotes the occurrence of a skill (column) in an observation (row) with values equal to 1, while the absence of a skill is denoted with values equal to 0. It should be mentioned that the retrieved data were all published in 2024, up to Q3 2024, meaning that the demonstrated experiments cover the recent demand for digital skills in Europe when it comes to systems analysts. In

Table 1. Sectors within the collected dataset.

Sector	Sector No	Advertisements
Accommodation and food service activities	1	5000
Administrative and support service activities	2	5000
Arts, entertainment and recreation	3	5000
Construction	4	5000
Education	5	5000
Electricity, gas, steam and air conditioning supply	6	5000
Financial and insurance activities	7	5000
Human health and social work activities	8	5000
Information and communication	9	5000
Manufacturing	10	5000
Other service activities	11	5000
Professional, scientific and technical activities	12	5000
Public administration and defence; compulsory social security	13	5000
Real estate activities	14	4730
Transportation and storage	15	5000
Water supply; sewerage, waste management and remediation activities	16	1702
Wholesale and retail trade; repair of motor vehicles and motorcycles	17	5000

, the seventeen sectors included in the dataset of this study are presented, where approximately 5000 from each one were retrieved in order to keep only the latest online job advertisements and maintain a balance across the sectors. Note that the observations associated with Water supply; sewerage, waste management and remediation activities are significantly less than 5000 due to the rareness of systems analysts in this sector.

3.2. RShiny Tool and Utilized Methods

Regarding the utilized RShiny tool, in brief, the developed application combines a graphical interface with functionalities accommodating machine learning and statistical methods to offer an observatory covering a variety of purposes, including the three RQs. In general, this tool is designed for individuals aiming to conduct content-wise group comparisons, where groups may be established based on one or more categorical variables within an imported dataset, e.g., online job advertisements, news feeds, and online posts. The main structure of the developed tool is illustrated in Figure 1, where the RQs of the presented study are mentioned together with the functionalities used to provide relevant answers.

3.2.1. Creating a Session

Initially, the user launches the application, and the first step is to load a file from a previous session or create a new one. By creating a new session, the user has to follow a series of processes to create a relevant file that stores information about the dataset, groups, and data structures used to conduct content-wise comparisons within the tool. These structures are the Document Term Matrix (DTM), Term Co-occurrence Matrix (TCM), word vectors, and topic extraction models, which are developed for each group and the entire dataset.

Initially, the user imports a file storing the main information of the dataset, where a column with the textual information of the data (optional when a previously established DTM is available) and at least one column with a categorical variable should be imported. The next step is to form groups by selecting one or more categorical variables, where the groups are created by all possible combinations of the selected variables. In the presented case study, the dataset comprises a data frame where a unique online job advertisement is stored per row while an identifier and a sector constitute its features (columns). As mentioned previously, the sector variable is selected to create 17 groups.

After forming the groups, the user can either select a textual feature (column) in the data, containing the textual information of the observations used for text analysis, to form a DTM or import a file of a previously established DTM. In both scenarios, the DTM constitutes the basis for establishing the rest of the data structures within the tool and conducting group analysis. When a DTM is not previously established (Figure 2), the user can select the weighting scheme to apply. In this case, the bag of words (Raw Term Count or Term Frequency (TF)), Term Frequency–Inverse Document Frequency (TF-IDF), and binary weighting (Binary Weighting) schemes are available. Also, the user can select the minimum and maximum frequency ratio of a word to be included in the DTM. Further, the text preprocessing options are the following [22]: (i) lowercase transformation; (ii) punctuation, mention (@), and/or hashtag (#) removal; (iii) number, HTML, question mark, exclamation mark, and/or digit replacement; (iv) stemming; (v) N-gram creation. Alternatively, the user can select the Basic preprocess option, which omits the previous text preprocessing options and applies lowercase transformation and keeps only alphanumeric characters in the texts. After selecting all these options, the user can establish a DTM and TCM for all groups and the entire dataset by pressing the APPLY TEXT PREPROCESSING button. When a previously established DTM is imported, the order of the imported rows should match the rows of the initial file containing the primary information of the dataset.

The next step is to develop word vectors, or term vectors depending on the properties of DTM, based on term co-occurrences, which can be combined with the Global Vectors algorithm—GloVe [23]—or weighted through an introduced standardization approach. Given a term (row), the data structure concerning standardized co-occurrences stores information about the conditional probabilities of other terms (column), given that the term appears. Our initial idea was that terms with a tendency to occur in similar contexts would also exhibit similar conditional probability distributions across all terms. As a result, terms that frequently co-occur in the same documents will have a low distance in this new data structure. The approach for calculating standardized term co-occurrences is described by the following equation, where $m$ is the number of observations/terms in the TCM.

$${TCM\_std}_{i,j}={\displaystyle \frac{{TCM}_{i,j}}{{TCM}_{i,i}}}, \mathrm{f}\mathrm{o}\mathrm{r} \mathrm{i}, \mathrm{j} = 1, 2\dots \mathrm{m}$$

(1)

Optionally, autoencoders and the Uniform Manifold Approximation Projection (UMAP) algorithm [24] can be used with either the GloVe or standardized co-occurrences, where the user can select the number of dimensions in any as well. A vital parameter of the UMAP algorithm reflecting the number of neighbors considered to approximate the new low-dimensional projection of the words is accessible as well.

The goal of this approach, in any configuration, is to establish term vectors in a new multidimensional space, where terms that frequently co-occur in the dataset occur nearby in this new space as well. Through vectors with characteristics of this type, clustering algorithms can effectively capture patterns in the data, which is indeed the next step of the preprocessing phase of the tool. The interface for establishing term vectors is presented in Figure 3.

Regarding the case study of systems analysts, an external file that was created for this study is imported (IMPORT NEW FILE button in Figure 2). This file is the binary matrix described in Section 3.1, which is equal to DTM with binary weighting, and stores information about the existence of a digital skill (column) within an online job advertisement (row). In addition, term vectors were developed using standardized term co-occurrences combined with the UMAP algorithm.

The final step towards creating the file storing the main information of the session is to develop topic models for the entire dataset and per group. For this purpose, the user can select an algorithm, the range of topics, as well as the evaluation approach to determine the optimal number of topics. It should be noted that different settings can be applied for each group and the entire dataset, and currently, there are three approaches for selecting the optimal number of topics. The first one is based on topic coherence [25], the second one is based on topic divergence [26], and the final one combines both topic coherence and divergence via a rank-based score.

Overall, topic coherence scores are used to evaluate the co-frequencies between the top words of each topic, where higher scores are obtained when the top words usually occur together in the dataset, while lower scores indicate the opposite. Within the tool, Normalized Pointwise Mutual Information (NPMI) is employed [25]. This evaluation metric relies on the pairwise word co-occurrences and the overall occurrence of individual words in a dataset to determine whether the most probable words of the extracted topics co-occur frequently. Usually, this metric is selected for topic coherence evaluation due to its scoring mechanism, which is very close to human judgment.

$$NPMI\left(x,y\right)={\displaystyle \frac{\ln \left(\frac{p(x,y)}{p\left(x\right)p\left(y\right)}\right)}{-\ln \left(p(x,y)\right)}}$$

(2)

In our case, $p(x,y)$ is the probability of words x and y co-occurring in a document while $p\left(x\right)$ is the probability of x occurring in a document of the dataset, where $p\left(y\right)$ is calculated similarly. The final score for a topic model is given by the average NPMI evaluations between the top words of each topic. From a different perspective, a topic divergence measure can be used to evaluate the distances between the topics, as indicated based on their top words or other properties. The tool evaluates the overall topic divergence of a topic model by counting the ratio of the unique words within the top words of all topics [26], as shown in the following equation:

$$Divergence\_score={\displaystyle \frac{\#unique(top\_w)}{k\times \#top\_words}}$$

(3)

where $\#unique(top\_w)$ is the number of unique words between all the top words in the topics, $k$ is the number of the extracted topics, and $\#top\_words$ is the number of top words examined per topic.

When it comes to topic extraction algorithms, the user can either train a topic model, using the Non-Negative Matrix Factorization (NMF) [27,28] or Latent Dirichlet Allocation (LDA) [29] algorithms, or train a word–term cluster model, using Fuzzy K-Means (FKM) [30] or Gaussian Mixture Models (GMMs) [31,32,33]. For both approaches, two data structures concerning distributions of words over topics and distributions of topics over documents are the basis for topic analysis and comparison in the tool. Generally, these distributions constitute the standard output for topic modeling algorithms, including LDA and NMF. In the case of cluster models, these distributions are calculated based on term frequencies (DTM) and the term weights assessed by these models for each cluster [34]. The interface for topic extraction is presented in Figure 4, where all_data refers to the analysis conducted for the entire dataset. As mentioned previously, different options can be applied for each group via the input widget labeled as Select groups. In the presented case study, the experiments were conducted in the range of 2 to 30 topics using the GMM approach and the topic coherence evaluation option for the entire dataset.

3.2.2. Data Analysis and Visualization Utilities

After finalizing the DTM, word vectors, and topic extraction models, the user can now finalize the necessary data structures and be transferred to the main dashboard of the tool, which provides a download option to store the session file locally. After downloading the session data, the user may continue with the main utilities of the tool or close the application and load this session file in the future.

Along with the download option, the main dashboard visualizes the basic information of the dataset, i.e., the imported data, the number of data observations and terms, and the overall frequency of each group. In total, apart from the main dashboard, the fundamental utilities of the tool are organized in three tabs named TEXT ANALYTICS, TOPIC ANALYSIS, and CLASSIFICATION MODELS.

Starting from the first one, the main goal is to provide insights covering the terms included in the DTM. The user can first review the main terms per group and for the entire dataset to understand the most prevalent and prevailing concepts within the imported data via bar plots and data tables (RQ₁). Next, the user can compare the groups in the dataset by computing Jensen–Shannon divergence [35] between every pair of groups, where each group is represented by a probability distribution of terms based on the DTM and the observations belonging to the group (RQ₂). Also, a two-dimensional representation of the groups is provided, created using multidimensional scaling [36], to assist the user in detecting similar groups and potential group clusters in a more direct manner (RQ₂). The final utility of the tab includes the identification of key terms within a group based on the frequency of each skill in the group and outside the group. To achieve this, the user selects a target group to be inspected, and then the system provides evaluations for each term according to mutual information [37], the Spearman correlation coefficient [38], and the frequency ratio of the term inside the query versus outside, i.e., the rest of the dataset (RQ₃). Regarding the last approach, a score equal to 2 would mean that the term is two times more likely to occur in the selected group compared to the rest as a whole. Overall, the existence of a term (binary weighting) is considered for evaluating its importance instead of other weighting schemes in all approaches. Similarly, the evaluations are performed for a binary target variable, where the observations of the selected target group are labeled with values equal to 1, while the rest of the observations are labeled with values equal to 0.

The second tab is oriented to the topic properties extracted from each group and the entire dataset. First, the user can review the top terms per topic to understand the underlying subjects and concepts within the data (RQ₁). Like the previous tab, a utility for evaluating key topics per group is provided. Hence, the user is able to identify rare subjects and directions within groups, i.e., subjects occurring more frequently inside a group than the rest of the dataset (RQ₃). For this purpose, the tool offers functionalities for evaluating the percentage of a topic in a group versus outside the group. Given a topic and a group, this percentage is calculated as the average weight of the topic in the observations of the group versus the average weight of the topic in the observations outside the group. Information of this type can reveal topics that are more relevant or irrelevant to a group compared to the rest of the dataset.

The final tab offers functionalities for training and evaluating classification machine learning models, where the DTM or the topic weights of each observation can be used as independent variables while the group of each observation constitutes the dependent variable. The practical usage of this tab is to reveal whether the observations in a group can be effectively distinguished from the observations belonging to other groups (RQ₂). Via the tool, the user can conduct experiments to either predict the group of the observations or investigate the uniqueness of one target group with negative sampling, i.e., values equal to 1 for the target group (positive instance) and values equal to 0 for observations that do not belong to the target group (negative instances). Currently, the available performance metrics are Accuracy, Precision, Recall, and F1-Score while the employed algorithms are implemented using the Classification and Regression Training library—CARET [39]. These algorithms are the Classification and Regression Trees (CARTs), Random Forest (RF), Gradient Boosting Machines (GBMs), and Extreme Gradient Boosting (XGB). It should be mentioned that any potential occurrence of class imbalance can be handled by the user via oversampling or downsampling and that the selected data are split into training (70%) and testing (30%) subsets to evaluate classification performance. All in all, a group that can be effectively separated from the rest, i.e., high classification performance, indicates that the group is indeed associated with some unique combinations of terms and concepts. Findings of this type may provide valuable insights to the end user depending on the context of the analysis. In the presented study, high accuracy levels would indicate that the job positions of a sector demand combinations of digital skills that are not common in the rest of the sectors.

4. Experimental Results

This section presents the main results of this study, reflecting the primary digital skills of system analysts and the key characteristics of each sector. This section is organized according to the presented RQs, where the main skills and skill topics are demonstrated and discussed in Section 4.1 (RQ₁), the evaluations regarding the potential distinction of each sector are presented in Section 4.2 (RQ₂), and Section 4.3 presents the key digital skills and skill topics of each sector (RQ₃).

At this point, it is worth providing an inclusive summary of the steps and settings that were followed to create all the necessary data structures, i.e., DTM, TCM, word vectors, and topic models, for creating the session file used in the following experiments. These steps are also discussed in Section 3.2.1 across the different methodological parts that were followed to establish a session file. Prior to the use of the tool, two datasets were constructed, one storing the sector information and one the skill information (see Section 3.1). First, the former file was imported to create 17 distinct groups, 1 per sector. Next, instead of proceeding to text preprocessing pipelines for creating a DTM, the second file was imported, which also constitutes a DTM with binary weighting. Furthermore, to establish word vectors, the skill co-occurrences were imported to the UMAP algorithm to project the skills in a new vector space with five dimensions. These skill vectors constituted the inputs to the GMM algorithm, where 29 different models were trained in the range from 2 to 30 topics/clusters to find the optimal one. To evaluate the models, the topic coherence approach in the tool was followed, i.e., NPMI. The later analysis constitutes the final step for creating a session file, as all data structures required for content analysis and group comparison (Section 3.2.2) were developed at that point, i.e., DTM, TCM, word vectors, and topic models. Regarding the following analysis, the approaches of the tool that were followed to provide answers to the RQs of this study are mentioned across the different subsections.

4.1. Key Skills and Skill Topics (RQ₁)

Regarding RQ₁, the most frequent digital skills were first reviewed to identify the primary directions outlining the skill demand of systems analysts (

Table 2. Top digital skills of systems analysts in Europe.

Skill	Frequency	Skill	Frequency
Have computer literacy	47,158	Use spreadsheets software	15,563
Computer programming	33,543	Analyse software specifications	15,337
Business ICT systems	29,054	Java computer programming	12,795
Database	28,326	Administer ICT system	11,730
Use Microsoft Office	26,800	Online analytical processing	11,046
Process data	21,520	Cloud technologies	9250
Perform data analysis	20,578	Use scripting programming	9175
Use object-oriented programming	19,340	SQL	8929
Office software	17,582	Hardware components	8719
Computer science	17,249	Manage database	8133
ICT system programming	15,829

). It should be mentioned that the rest of the skills occurred in less than 6000 online job advertisements, which is less than 10% of the dataset. In summary, the table indicates that the primary skill directions mostly concern computer programming (computer programming, use scripting programming) and related languages (Java, SQL, Python, PHP), data analysis (process data, perform data analysis, databases, manage database), office and presentation software (use Microsoft Office, office software), and administration, business, and maintenance (have computer literacy, business ICT systems, administer ICT system). Some other areas of interest within the most frequent digital skills include computer science, cloud technologies, hardware, and software specifications.

When it comes to the extracted topics (or skill clusters), the GMM algorithm provided the highest coherence under 12 topics among the 29 models (2 to 30 topics), which corresponds to the model that will be further explored for the rest of this study. Additional details regarding the evaluations of all four algorithms in the range from 2 to 30 topics can be found in the Appendix A, in

Table A2. Performance evaluations of topic extraction models (topic coherence).

Topics	GMM	FKM	LDA	NMF
2	0.199	0.356	0.087	0.540
3	0.513	0.513	0.110	0.520
4	0.316	0.440	0.124	0.504
5	0.229	0.507	0.108	0.491
6	0.310	0.556	0.102	0.511
7	0.485	0.462	0.097	0.467
8	0.542	0.546	0.087	0.425
9	0.553	0.498	0.092	0.392
10	0.562	0.508	0.109	0.411
11	0.548	0.464	0.118	0.403
12	0.565	0.484	0.114	0.367
13	0.530	0.474	0.101	0.372
14	0.519	0.455	0.104	0.394
15	0.541	0.481	0.098	0.375
16	0.503	0.438	0.111	0.357
17	0.504	0.425	0.120	0.409
18	0.475	0.424	0.105	0.392
19	0.455	0.428	0.109	0.401
20	0.446	0.416	0.132	0.392
21	0.434	0.430	0.113	0.386
22	0.445	0.413	0.119	0.405
23	0.446	0.428	0.125	0.399
24	0.429	0.411	0.109	0.391
25	0.419	0.451	0.122	0.394
26	0.390	0.428	0.123	0.378
27	0.426	0.436	0.112	0.391
28	0.422	0.438	0.125	0.368
29	0.419	0.445	0.124	0.372
30	0.405	0.404	0.118	0.363

, where the selected model achieved the highest coherence across the experiments. The main skills per topic are presented in

Table 3. Topic details of digital skills within systems analysts’ online job advertisements.

No	Topic Title	Prevalence	Top Skills
1	Developing ICT systems	19,508.1	computer programming; business ICT systems; ICT system programming
2	Microsoft Office and office tasks (spreadsheets, presentation, etc.)	22,273.0	have computer literacy; use Microsoft Office; office software; use spreadsheets software
3	Web programming	1175.3	web programming; use query languages; PHP
4	Scripting programming for extending applications (e.g., e-learning platforms)	2679.4	use scripting programming; e-learning; Python
5	Data processing and analysis	11,491.9	database; process data; perform data analysis
6	Database management	2104.6	SQL; manage database
7	Software specifications and designs, and online presentation of multidimensional data	9961.4	use object-oriented programming; analyse software specifications; Java computer programming; online analytical processing
8	System administration	3211.0	administer ICT system; manage ICT data architecture; maintain ICT system; deploy ICT systems; ICT system integration
9	Web analytics and e-commerce	351.8	web analytics; use content management system software; e-commerce systems; work with e-services available to citizens; use analytics for commercial purposes
10	System infrastructure and components	8343.0	computer science; cloud technologies; hardware components; ICT infrastructure
11	Networks	304.0	ICT networking hardware; ICT network simulation; define ICT network design policies; protect ICT devices; Cisco
12	Antivirus and threat detection	28.4	implement antivirus software; manage IT security compliances; remove computer virus or malware from a computer; NESSUS

, where a comprehensive title is also given according to the core meaning of these skills. In addition, the column prevalence indicates the probability sum of all observations associated with each topic. It should be noted that different numbers of top skills are presented in each case, as significant weight differences between the skills have been observed. For example, in Topic 3, PHP is assigned a weight close to 26%, while the skills design user interface and CSS were assigned a weight lower than 6%.

In total, the main content and skills of each topic agree with the previous analysis, while some additional areas of interest were revealed. These directions concern web programming (Topic 3), web analytics (Topic 9), networks (Topic 11), and information security (Topic 12). In addition, the employed clustering approach succeeded at discriminating topics carrying semantic similarities, e.g., Topic 1 and Topic 10, Topic 3 and Topic 9, Topic 5 and Topic 6. These findings make the respective analysis an important one when it comes to understanding the core requirements of systems analysts. Accordingly, systems analysts can be characterized as a demanding category of occupations, as the potential candidates should be specialized in diverse domains requiring different technologies and background knowledge. Moreover, based on the prevalence of each topic, there is a clear separation between major and minor topics, meaning that individuals should prioritize the major ones to compete in the labor market. Based on these evaluations, the requirements related to Topic 1 and Topic 2 can be considered among the primary concerns of systems analysts, followed by Topic 5, Topic 7, and Topic 10.

In summary, the results in this section can offer valuable insights into individuals intending to fill relevant job positions, as the analysis addressed both the most prevalent required skills and skill topics. Thus, the current study can be reviewed for understanding the different directions of systems analysts and selecting specific skills to learn/acquire according to the interests, prior expertise, and preferred job positions of each individual. Undoubtedly, relevant organizations may also benefit from the outcomes of this section by acknowledging the key activities of their competitors to potentially develop new components and compete in the labor market more effectively.

4.2. Similarities and Dissimilarities Between the Sectors (RQ₂)

The current section provides the main outcomes concerning the potential distinction of unique sectors through statistical analysis and machine learning models. In the following figure (Figure 5), a two-dimensional representation of the 17 sectors surrounding systems analysts is demonstrated.

Figure 5 indicates that the Professional, scientific and technical activities can be considered as the most diverse sector, followed by the following: (i) Public administration and defence; compulsory social security, (ii) Electricity, gas, steam and air conditioning supply, (iii) Human health and social work activities, and (iv) Financial and insurance activities. With more detail, Figure 6 presents the divergence evaluations comparing the Professional, scientific and technical activities sector and the rest. Overall, the utilized measure ranges from 0 to 1, where values closer to 1 indicate total independency, while values closer to 0 indicate identical probability distributions, i.e., similar sectors/groups. The figure shows that the minimum divergence is higher than 0.03, while the rest of the sectors were associated with at least one lower evaluation. It should be mentioned that when considering all skills, not only the digital ones, the Electricity, gas, steam and air conditioning supply sector obtained the lowest minimum divergence (close to 0.10) from the other sectors. The other sectors were associated with a minimum divergence close to 0.05 or lower.

To provide sufficient evidence about the potential uniqueness of systems analysts within each sector, the available machine learning capabilities of the developed RShiny tool were utilized as well. First, an experiment without a target group was conducted in an attempt to predict the sector of online job advertisements, where a maximum accuracy close to 33% was observed. When considering that 17 sectors are investigated, the latter outcome indeed shows that the different sectors possibly require unique sets of skills. Hence, it is evident that the divergence levels of the previous analysis can be considered significant.

Moreover, an experiment was conducted for each potential pair of sectors, where the observations of one sector were labeled with values equal to 1 while the observations of the other sector were labeled with values equal to 0. Given the fact that the datasets were balanced across the two potential classes in most experiments (5000 observations each), except for the Water supply; sewerage, waste management and remediation activities sector, accuracy levels higher than 50% can be considered as significant for providing insights regarding RQ₂. In

Table 4. Performance evaluation with negative sampling.

Sector	Mean	Median	Min	Max
Accommodation and food service activities	0.763	0.759	0.697	0.861
Administrative and support service activities	0.723	0.721	0.64	0.828
Arts, entertainment and recreation	0.734	0.717	0.692	0.821
Construction	0.733	0.737	0.684	0.816
Education	0.746	0.748	0.685	0.84
Electricity, gas, steam and air conditioning supply	0.786	0.781	0.745	0.846
Financial and insurance activities	0.753	0.745	0.695	0.831
Human health and social work activities	0.76	0.763	0.692	0.827
Information and communication	0.745	0.744	0.683	0.84
Manufacturing	0.728	0.73	0.667	0.822
Other service activities	0.722	0.72	0.646	0.819
Professional, scientific and technical activities	0.829	0.831	0.801	0.853
Public administration and defence; compulsory social security	0.754	0.764	0.653	0.85
Real estate activities	0.751	0.758	0.684	0.788
Transportation and storage	0.773	0.766	0.734	0.852
Water supply; sewerage, waste management and remediation activities	0.774	0.779	0.695	0.845
Wholesale and retail trade; repair of motor vehicles and motorcycles	0.731	0.721	0.695	0.819

, the main information concerning the accuracy of the models trained with the XGB algorithm is presented, where this algorithm was selected due to its overall high performance compared to the rest in the experiments of this study.

All in all, the experiments show that, in many cases, the accuracy levels were higher than 75%, while the Professional, scientific and technical activities sector is the only one with evaluations higher than 80% in all examined metrics. The latter finding suggests that each sector demands unique combinations of digital skills, which should be thoroughly addressed. It also proves that the detected divergence indicators that were discussed previously are indeed significant in distinguishing online job advertisements across the sectors.

At this point, it should be mentioned that the accuracy levels were significantly increased, with a rate close to 10%, when the rest of the non-digital skills were used in these experiments as well. However, there are some odd ESCO skills that would not provide significant insights when it comes to distinguishing unique sectors, e.g., implement nursing care, read people, and accounting techniques. According to the median and mean accuracy per sector, Electricity, gas, steam and air conditioning supply can also be distinguished as a unique sector, while the Water supply; sewerage, waste management and remediation activities and Transportation and storage sectors are associated with high evaluations as well.

Regarding RQ₂, both approaches suggested that the systems analysts associated with Professional, scientific and technical activities can be considered as the sector demanding the most unique combinations of digital skills, followed by the Electricity, gas, steam and air conditioning supply sector. In total, the experiments indicated that all sectors can be usually distinguished from the rest, meaning that individuals with unique combinations of skills should explore the differences between sectors to identify the most suitable ones according to their experience and preferences.

4.3. Key Digital Skills and Topics of Systems Analysts (RQ₃)

The current section extends the findings of the previous one by presenting an in-depth analysis of the key skill directions demanded from the different organizations when it comes to systems analysts. To evaluate individual digital skills, the mutual information was first calculated, while the frequency ratio of a skill inside versus outside each sector was also considered to understand whether a skill is associated with higher or lower relative demand in comparison to the other sectors.

Based on these evaluations,

Table 5. Skill categories and technologies associated with each sector.

Sector	Skill Categories	Programming Languages and Technologies
1	Databases; Cloud technologies	SQL (2.6); cloud technologies (2.1); Python (2.5); use spreadsheets software (1.7); Java (1.7); PHP (2.2); Hadoop (3.5)
2	Cyber security	Implement antivirus software (5.9); Cisco (2.2); TypeScript (2.4)
3	ICT systems and projects; ICT communications	Use communication and collaboration software (2.5); Automation technology (1.7); CSS (1.8); Microsoft Access (1.5)
4	Text processing; security legislation	Use word processing software (1.7)
5	Research and innovation; Data analysis; Cloud technologies	QlikView.Expressor (2.7); Cloud technologies (1.7); Visual Studio NET (1.7); Python (1.3)
6	ICT market; Presentations	use presentation software (1.7)
7	ICT systems and projects; Cyber security	NoSQL (2.7); iOS (2.0); Hadoop (1.9); WordPress (1.8); JavaScript (1.6)
8	ICT communications; ICT systems and projects; Software design and specifications; Text processing	Use communication and collaboration software (5.4); Visual Studio NET (3.3); Microsoft Access (2.9); use word processing software (1.5)
9	Cyber security; ICT communications; Networks; Mobile; ICT Systems and projects; Data analysis; Text Processing; IoT	SAP Data Services (5.9); IOS (2.8); Use communication and collaboration software (2.7); Android (2.1); use word processing software (1.6); Visual Studio NET (1.6); Microsoft Access (1.5);
10	Web	Objective C (1.6); NoSQL (1.5); C (1.4)
11	Data analysis; Quality control; Office tasks; ICT market	Office software (1.6); Oracle Application Development Framework (1.9); PHP (1.5); SQL (1.3); use query languages (1.5)
12	Data analysis (online analytics); Software design and specifications	Microsoft Access (2.2); use content management system software (1.9); SAP Data Services (1.9)
13	CAE; Office tasks; ICT market; Cyber security; ICT systems and projects; Quality control; IoT	Use CAE software (2.8); Office software (1.9); C (4.0); implement antivirus software (1.9); TypeScript (1.6); NoSQL (1.5); PHP (1.3)
14	ICT systems and projects; Cyber security	Query languages (1.8); SQL Server (1.8)
15	Data analysis (analytics and machine learning); Cyber security; Web; Presentations	CSS (2.9); WordPress (2.3); implement antivirus software (1.9); Python (1.7); use presentation software (1.7); C (1.7); Java (1.4)
16	ICT systems and projects; Data analysis (online analytics)	Use content management system software (2.2); Objective C (1.7)
17	Data analysis (online analytics); Databases	NoSQL (1.7); SQL Server (1.7); Objective C (1.6); WordPress (1.6)

presents the skill categories along with the primary programming languages and technologies (or types of technologies) associated with each sector, where a score is provided for each one. For example, in the first sector (Accommodation and food service activities), the digital skill SQL is 2.6 times more likely to occur in job positions compared to a random position belonging to one of the other sectors. In addition, skill categories were assigned to sectors when either multiple significant skills with semantic similarities are observed, e.g., manage database, databases, or a frequent individual skill was detected. For this purpose, the ESCO taxonomy was manually inspected to find semantic similarities, where some new concepts are also introduced after the inspection of the individual skills and skill categories, e.g., Computer-Aided Engineering (CAE).

All in all, the different areas of interest discussed in Section 4.1 are also significant in multiple sectors according to the skill evaluations. In addition, the categories associated with Data analysis (seven sectors), ICT systems and projects (seven sectors), and Cyber security (six sectors) were evaluated with high relative frequency in most sectors. Regarding the practical usage of these outcomes, individuals may review the above analysis to identify sectors covering their expertise when it comes to the different areas of interest and technologies.

With more detail, while most sectors were associated with more than one area of interest, the Electricity, gas, steam and air conditioning supply and Manufacturing sectors were related to distinct concepts, i.e., presentation software and web programming, respectively. Furthermore, the Education sector should be considered as the most appropriate sector for Research and innovation activities while the Information and communication sector was proved as the most generic option; i.e., many areas of interest were observed.

In addition, while the Public administration and defence; compulsory social security sector was also associated with multiple skill categories, CAE was evaluated with high relative frequency only within this sector. In total, apart from these observations, which are associated with distinct sectors, individuals may select some unique concepts of their interest with few overlaps, such as Web analytics, Software design and specifications, IoT, and Quality control, to effectively identify and review the sectors with the highest probability of including the desired concepts in relevant job positions.

As the latter experiments were completed through qualitative analysis and manual inspection, the topics extracted from the whole dataset were also evaluated based on the ratio of each topic inside a sector versus outside. Through this analysis, a more structured overview can be provided, as the topics are represented by distinct digital skills and concepts that can cover the entire dataset. Overall, the most significant observations of sectors obtaining the highest and lowest evaluations per topic are presented in

Table 6. Most informative sectors per topic.

Topic	High Frequency	Low Frequency
1	Professional, scientific and technical activities (1.26)	Electricity, gas, steam and air conditioning supply (0.64)
2	Electricity, gas, steam and air conditioning supply (1.85)	Professional, scientific and technical activities (0.69)
3	Manufacturing (1.54) Accommodation and food service activities (1.49)	Professional, scientific and technical activities (0.54) Electricity, gas, steam and air conditioning supply (0.50)
4	Transportation and storage (1.59) Accommodation and food service activities (1.52)	Public administration and defence; compulsory social security (0.59) Electricity, gas, steam and air conditioning supply (0.45)
5	Education (1.54) Other service activities (1.39)	Professional, scientific and technical activities (0.61)
6	Accommodation and food service activities (2.2)	Professional, scientific and technical activities (0.48) Electricity, gas, steam and air conditioning supply (0.47)
7	Professional, scientific and technical activities (1.9) Human health and social work activities (1.5)	Electricity, gas, steam and air conditioning supply (0.64) Financial and insurance activities (0.55)
8	Financial and insurance activities (1.86) Real estate activities (1.58)	Public administration and defence; compulsory social security (0.63) Electricity, gas, steam and air conditioning supply (0.42)
9	Water supply; sewerage, waste management and remediation activities (2.4) Professional, scientific and technical activities (2.35) Wholesale and retail trade; repair of motor vehicles and motorcycles (1.81)	Human health and social work activities (0.54) Manufacturing (0.54) Education (0.50) Transportation and storage (0.49) Financial and insurance activities (0.46)
10	Accommodation and food service activities (1.37)	Human health and social work activities (0.67)
11	Administrative and support service activities (2.50) Information and communication (2.02)	Electricity, gas, steam and air conditioning supply (0.45) Professional, scientific and technical activities (0.26)
12	Administrative and support service activities (6.32)	Arts, entertainment and recreation (0) Professional, scientific and technical activities (0) Water supply; sewerage, waste management and remediation activities (0)

, where higher evaluations indicate higher demand within a sector than in any other sector, while lower evaluations indicate lower demand.

According to the above table, many topics indicated significant observations with both high and low evaluations. Apart from the extreme observations of Topic 12, which covers a small proportion of the data, the topics concerning Office tasks (Topic 2), Database management (Topic 6), Software specifications and designs and online presentation of multidimensional data (Topic 7), System administration (Topic 8), Web analytics (Topic 9), and Networks (Topic 11) provided the most informative outcomes, with ratios exceeding 1.8. The rest of the maximum evaluations were lower or close to 1.5 and indicate that these topics are not significantly concentrated around unique sectors. Also, it should be noted that the manual inspection from the previous analysis matches the outcomes of this analysis on multiple occasions, including presentation software (Topic 2), web programming (Topic 3), databases (Topic 6), and web analytics (Topic 9). Regarding the lowest evaluations, the Professional, scientific and technical activities and Electricity, gas, steam and air conditioning supply sectors occurred in most topics, which is an observation further explaining their overall divergence from other sectors. On the other hand, the highest evaluations per topic were not concentrated around distinct sectors, as most sectors were evaluated with high relative frequency in at least one topic.

In summary, the experiments of this section pointed out significant associations between the different concepts surrounding systems analysts. Thus, job seekers with unique sets of skills may review the above tables to identify the most suitable sectors, according to their preferences. For example, job seekers who are interested in web analytics (Topic 9) will be guided to explore organizations within the Water supply; sewerage, waste management and remediation activities, Professional, scientific and technical activities and Wholesale and retail trade; repair of motor vehicles and motorcycles sectors according to their activity and requirements. From a different perspective, systems analysts focusing on a specific sector may also review the outcomes of this study to detect the most appropriate skills according to the requirements of relevant job positions. In this context, employees in Administrative and support service activities can be guided into digital skills that primarily concern cybersecurity and networks instead of focusing on other more generic concepts.

5. Discussion

The various characteristics of online job advertisements can be considered an invaluable knowledge base for understanding labor market demand from multiple perspectives covering regional, industrial, and sectoral characteristics. In this spectrum, the current study provided insights regarding the digital ESCO skills required by systems analysts in Europe. The analysis was conducted for a dataset containing 81,432 online job advertisements posted from Q1 to Q3 2024, where each job/observation was associated with a unique sector (17 in total) and a set of digital ESCO skills (201 unique digital skills were detected). The respective outcomes showed that there are different types of skills required by systems analysts, which can be categorized into major (more frequent) and minor (less frequent). In addition, the key differences between the sectors were also addressed by evaluating the frequencies of digital skills and skill topics with each one. Also, this study presented a tool and methods that were utilized to derive the desired outputs, which ultimately led to providing insights covering the three RQs of this study. As the imported dataset followed some standards regardless of its special properties, this tool can be used for other types of data as well.

Regarding the main RQs of this study, the analysis first showed that organizations within the European labor market demand diverse sets of digital skills to cover the requirements of systems analysts (RQ₁). To reveal the main digital skills and groups of skills required from job positions of this type, the occurrences and co-occurrences of the detected digital skills were analyzed. The core concepts of the most frequent digital skills within the dataset indicated that the primary duties of systems analysts concern administrative, business, office, and data management tasks along with a variety of programming languages, e.g., PHP, Python, and Java. These concepts were also confirmed by a topic extraction approach, where the GMM algorithm was used to address 12 skill topics/clusters overall. Through this analysis, some less frequent directions were also revealed, including cybersecurity, networks, web analytics, and e-commerce, hence confirming the variety of skills related to systems analysts. In addition, the latter analysis was indeed valuable in understanding the relationships between the different skills by identifying subsets of skills that are more likely to occur or not occur together. On top of that, skills with semantic similarities were also separated into different topics since they covered different concepts, e.g., Developing ICT systems and System administration.

Furthermore, the most unique sectors were detected through two different approaches (RQ₂), based on the similarities/dissimilarities between all groups. In the first approach, we calculated probability distributions of skills over the sectors, which were later used to measure the divergence between each pair. Additionally, a two-dimensional projection was provided, using the divergence evaluations, to identify sectors associated with more unanticipated requirements compared to the rest. Briefly, both the projection and the pairwise comparisons indicated that the Professional, scientific and technical activities sector is the most diverging one, as it obtained the highest minimum distance from other sectors and significantly higher distances overall. Going beyond this approach, multiple machine learning experimental setups were also conducted to explore whether the online job advertisements of each sector can be distinguished from the rest or not, with the exclusive use of the required digital skills. In short, the outcomes indicated that the accuracy levels were higher than 75% in most cases, while the Professional, scientific and technical activities sector was distinguished as the most unique sector once again. In addition, the machine learning models trained for modeling all 17 sectors, instead of modeling 2 sectors at a time, achieved high accuracy levels (close to 33%). This finding indicated that the different sectors indeed demand unique sets of skills when it comes to systems analysts.

Since the previous analysis suggested that different sectors require different sets of skills, this study also revealed the most unanticipated areas of interest within each sector and the most relevant to each topic sector (RQ₃). Initially, by combining information from the ESCO taxonomy with evaluation approaches based on statistical analysis, i.e., mutual information and frequency ratios, we assigned multiple areas of interest to each sector. At the same time, the most unanticipated technologies and programming languages within each sector were also identified and presented. The outcomes showed that there are some sectors associated with multiple areas of interest, e.g., Information and communication, Human health and social work activities, while there are others with more domain-specific responsibilities, i.e., Electricity, gas, steam and air conditioning supply, and Manufacturing. In a more collective manner, the most relevant to each topic sectors were addressed as well. The analysis revealed that the skills concerning office tasks (Topic 2), database management (Topic 6), software designs and online presentations (Topic 7), system administration (Topic 8), web analytics (Topic 9), and networks (Topic 11) are more frequent in distinct sectors and not balanced across the investigated sectors. Overall, the findings from both approaches indicated that each sector requires distinct sets of skills, hence showing that they should be engaged in different ways by third parties with specific interests in the labor market.

Ultimately, the main implications and contributions of the current study concern the potential future usage of the methods and main findings of the demonstrated experiments. First, individuals aspiring to work as systems analysts may review the presented outcomes to identify the core digital skills required to fulfill a relevant position. When it comes to achieving higher performance, experienced individuals with unique sets of skills and interests can find the most suitable sectors through the demonstrated experiments as well. Similarly, job seekers interested in a specific sector can detect the core differences between the general standards of systems analysts and this sector. Eventually, individuals may possibly increase their value as employees by identifying and focusing on specific sets of skills with higher demand rather than limiting their knowledge to less significant skills. Moreover, organizations requiring systems analysts may also inspect the outcomes of this study to outline the status of the labor market and the needs of immediate competitors, as expressed by each sector. In this way, they can filter their online job advertisements to adapt to the standards of systems analysts, as well as provide familiar and attractive descriptions to job seekers and potential applicants. Also, by evaluating the requirements of immediate competitors, organizations can possibly detect technologies and areas of interest that may adapt to their needs. In this way, third parties can develop or expand both new and existing ideas and products. Finally, both organizations and individuals aiming at establishing effective training courses and learning material can leverage the different outcomes of this study and develop programs focusing on specific significant skills and sets of skills, with respect to the labor market and their unique characteristics, properties, and interests.

When it comes to the employed methods, a tool for conducting content-based comparisons between groups was utilized. Apart from experiments concerning skill demand, the respective methods can be adapted to other applications exploring content-based differences between groups. To implement these methods, researchers may use any kind of dataset containing content features (text usually) as well as create groups based on one or more variables. Afterward, the main offerings of the employed approaches, which can be matched against the RQs of this study, are the following:

• Understand the most prevalent concepts within a dataset via NLP and statistical methods (RQ₁);
• Outline the primary directions with the use of topic analysis (RQ₁);
• Find similar and distant groups (RQ₂);
• Identify unique groups with exceptional content mixtures (RQ₂);
• Measure the degree of separation between groups through machine learning (RQ₂);
• Evaluate the key features and topics characterizing each group using different evaluation approaches (RQ₃).

6. Conclusions

The current study presented an exploratory analysis of online job advertisements related to systems analysts with the use of data mining, machine learning, and statistical techniques. The goal was to address the digital skill demand of systems analysts in Europe by evaluating the primary areas of interest characterizing these skills and by providing information concerning the different sectors surrounding occupations of this type. In total, the analysis showed that the job positions of systems analysts may require diverse skills and duties covering a variety of ICT directions like cyber security, networks, and web analytics. Furthermore, through sectoral analysis, the experiments indicated that these requirements are affected and vary based on the sectors associated with each organization. In addition, the Professional, scientific and technical activities sector was declared as a demanding sector requiring the most unique combinations of digital skills across the 17 investigated sectors.

Overall, the main purpose of this study was to offer valuable analytics that can be leveraged by third parties with various interests and scopes. In this context, we believe that the current study can primarily assist individuals in evaluating the labor market of systems analysts to understand and address potential skill gaps, identify suitable organizations and sectors, and ultimately, improve their value as employees. Accordingly, organizations can review the current study to extract insights regarding their competitors within the labor market and scout candidate technologies for improving their information systems and strategic position.

For future work, one noteworthy idea is to compare the findings of this study with global analytics and address the gaps and differences between the European labor market of systems analysts and the rest. Furthermore, as systems analysts belong to the wider category of ICT professionals, future projects might also explore systems analysts at occupational level, not only sectoral, to capture their key digital skills from a different point of view. Similarly, researchers may be also inspired by the experiments of this study to adapt and expand the proposed methods into different concepts or even other types of job positions. At last, third parties interested in education may also combine the findings of the current study with analytics derived from their databases and curriculum to design courses according to the needs of the labor market, to eventually help students to approach potential job and career opportunities more effectively.