A Sustainability Propensity Index to Assess a Migrants Integration Policy Programme: An NLP-Based Approach

This paper presents an experimental methodology for measuring the sustainability propensity of 530 National projects funded by a European Programme focused on migrants and refugees integration using a composite indicator. The study proposes a systematic approach for evaluating the achievement of a policy programme and its impact on both society and the beneficiaries over time. To construct the Sustainability Propensity Index (SPI) for each national project, we employed a sociocultural profiling method called Emotional Text Mining (ETM) applied to the contents their final reports. ETM is grounded in a socio-constructivist approach and utilizes natural language processing procedures (NLP), and a non-supervised bottom-up approach to identify the reports content and assess their sustainability propensity. We analyzed six sections from projects’ final report to identify prerequisites for sustainability. Even if sustainability can only be assessed post-project, early prerequisites can be included in the project planning phase. Hence, using a bottom-up NLP approach, we have extracted these prerequisites for sustainability identified by the scientific literature as project effectiveness, continuity of activities, and contribution to contextual change. We collected text from 530 project final reports, resulting in a corpus of nearly one million terms and 16,000 comparable chunks of text. We applied ETM procedure to identify 19 clusters of similar text chunks, which were then interpreted and assigned to the above mentioned sustainability dimensions on the basis of four experts’ assessments. Subsequently, we calculated indices for effectiveness, continuity, and change, normalizing them between zero and one, while also assessing project criticalities. A SPI was derived by aggregating these indices and subtracting criticalities. The final ranking of the projects was compared with that resulting from traditional qualitative analysis, revealing strong convergence between the two methods. These findings indicate that semi-automatic text mining techniques can streamline the evaluation process, enhancing efficiency without sacrificing accuracy.