A set of data generated from databases and repositories of science, technology, research and innovation related to Colombian scientific productivity, case study, was constructed. Several factors were included in the evaluation of the impact of research products on the materialization of sustainable development objectives, such as:

The generation of new knowledge published in scientific journals indexed in Scopus and Web of Science.

A seven-year dataset (2014-2021) was constructed with the following characteristics:

4568 books as research output.

One of the most important characteristics of the dataset used for this research is that all the information is in text format, and we sought to guarantee the quality of the data and the construction of a robust analysis model that would allow us to obtain significant results. The methodology implemented used data in text format: content of books, chapters, journals, working papers, public policies, guiding documents of the Sustainable Development Goals (SDGs), etc. The process of improving the quality of the data used in this research is based on a text format.

The process of improving the quality of the data consisted of carrying out activities to clean and select the most important information. For this purpose, the proposal of Rentería et al. (2022) will be used, where the following algorithmic sequence was established:

Tokenization. This is a process by which the text is segmented into words, i.e., a textual corpus is transformed into a set or list of words. For example, in the text “National System of Science and Technology”, its tokenization is: “System”, “National”, “Science” and “Technology”.

Once the textual corpus was obtained, the process of evaluating the configurations of the most important terms in each of the scientific products began, determining their co-occurrence, sequentially and location.

Subsequently, a technique was selected from the existing ones for topic modeling using Bayesian statistics; among the available techniques, the following stand out: Latent Semantic Indexing (LSA) ( Zhao et al., 2015 ) the Probabilistic Latent Semantic Analysis (PLSA) ( Hofmann 2001, 1999), and Latent Dirichlet Allocation ( lda) (Blei et al., 2015). According to Duran et al. (2015) , lda is one of the techniques that has generated more significant contributions to the study from different areas of knowledge: social sciences ( Ramage et al., 2009 ; Li and Lei, 2021; Jacobi et al., 2016 ), economics ( Hong et al., 2016 ), health ( Paul and Dredze 2013; 2014) among others.

For this research, the selected approach was the lda because the topics constructed with this model are highly coherent ( Stevens et al., 2012 ) and, therefore, it is a useful tool to assess the relevance of the results obtained from Colombian scientific productivity, related to the challenges established in the SDGs.

For the construction of the lda algorithm, the proposal of Duran et al. (2015) was used, where given a textual corpus D composed of M documents, with a document d with N _d ( ) the construction of topics was done as follows:

Select a multinomial distribution ф _t for the topic ( ) from a Dirichlet distribution with parameter β.

One of the most important aspects for this algorithm are the variables Ф and θ, and the hyperparameters α and β that, as recommended by ( Duran et al., 2015 ) must be defined from the perspective of latent variables. Therefore, under this principle for the estimation of the parameters in the textual corpus D.

and

As the parameters θ and Ф are part of the integration, it is necessary to build simulation models to establish the inference of these values, according to ( Blei et al., 2003 ; Rogers et al., 2005 ; Griffiths and Steyvers, 2004; Shivashankar et al., 2011 ; Coelho et al., 2010 ) the best computational alternative for this inference is Markov Chain Monte Carlo. Once the topics have been generated, it is necessary to establish a quality metric for the results obtained, which means, to define the optimal number of topics required to adequately classify all the words of the textual corpus D. In this sense, the most appropriate indicator is Perplexity, which is a measure of information theory, to evaluate how well a statistical model describes a data set ( Duran et al., 2015 ), therefore, the lower the quantification of Perplexity, the quality of the results obtained with the model is more reliable. The Perplexity equation ( ϛ) that will be used in this research is the one proposed by ( Duran et al., 2015 ; Blei et al., 2003 ):

In addition to Perplexity, the coherence indicator ( 𝜗) is adhered to evaluate the consistency of the words that make up each topic Ф _i, Ф _j (see equation 3), to later make a paired comparison based on the Kullback divergence metrics – Leiber (dkl) to establish the main similarities and differences from the distributions of the words in the topics and between the SDGs ( Mei et al., 2007 ).

Regarding this SDG, it was important to consider aspects included in the multidimensional poverty models proposed by Alkire et al. (2015), who calculated poverty from a synthetic indicator, using a weighting system based on the importance of each dimension in the configuration of deprivations antagonistic to well-being.

The findings of this SDG highlight the presence of important topics to address a set of solutions to the problems underlying the term, including terms such as “development”, “economic” and “health”, which are decisive aspects to solve the needs of poverty.

Therefore, in addition to incorporating the aforementioned factors, it is necessary to involve aspects related to habitat, and other factors that promote social wellbeing, such as access to quality public services, health services, among others.

This lack is reflected even in the smaller topics (smaller number of terms), demonstrating that the country’s scientific development is not generating the evidence to solve poverty problems as established by Ramírez et al. (2017) , Pinilla-Roncancio (2018) and Manzano-Núñez et al. (2022) . Finally, Cuesta and Pico (2020), mention that these poverty and inequality gaps have become more acute due to the pandemic.

In the case analyzed, in the sustainable development goal “Zero Hunger”, scientific development has generated important results in aspects such as: “Agriculture”, “Climate Change”, “Economy”, “Social Capital” and “Bioeconomy”, which according to Khanal et al. (2021) , are the keys to promote strategies to reduce hunger. “Social Capital” and “Bioeconomy” that according to Khanal et al. (2021) , are the keys to promoting strategies to reduce hunger because the alterations generated because of climate change affect the availability and quality of the product, reducing the presence of some nutrients, proteins and vitamins required for health welfare.

Blesh et al. (2019) , mentions that to ensure the construction of a sustainable model of zero hunger, in addition to a good food security system in the country, the support of a good health system, highly inclusive and based on the approach of prioritization of care from the social determinants of well-being, should be considered.

On the other hand, Sunderland et al. (2019) , highlights that factors such as adaptation and resilience to climate change should be on the government’s agenda to achieve the expected results and impacts driven by the guarantee of zero hunger in the population. Although these aspects are part of the most important issues (based on the proportion of terms and main components since the creation of the objective) and coherent, the intra-theme analysis shows the lack of a homogeneous model that guarantees its adaptability to the territorial needs of the country, since most of the strategies implemented are focused on the country’s capital cities, and not on isolated rural contexts and territories.

Health and well-being is another of the objectives that has a series of initiatives that have been developed in the country involving plans and programs supported by the Model of Comprehensive Health Care ( Hernández et al., 2019 ), and whose key aspects are found in the topics with the highest proportion of objective terms (“Health”, “Quality-Health”, “Care-Differentiated”, “Coverage-Integral”).

The study strongly suggested that the model established prioritizes the insurance of the population in the General Social Security Health System (SGSS), from an inclusive and differentiated approach, considering the socioeconomic gaps of the different population groups in the territory. In this way, the country has achieved health coverage of over 90 % of the population and has established a population screening model considering the vulnerability approaches proposed by Diderichsen et al. (2019) , which facilitates access to specialized health services.

Such contribution is reflected in the identification of the most relevant topics: “health”, “care”, “sanitary”, “humanization”, “risk”, “inclusion”, “social class” and “stratification”. However, researchers such as Roncancio et al. (2020) , mention that the development of this SDG in the country is not only achieved by guaranteeing the health system for the entire population, but it is also necessary to improve the incorporation of the social determinants of health, since there are sociodemographic and geographic factors that have had an important impact on the configuration of vulnerability that exacerbates the complications of some pathologies in population groups ( Roncancio et al., 2020 and Diderichsen et al., 2019 ).