Microblogging websites such as Twitter have caused sentiment analysis research to increase in popularity over the last several decades. However, most studies focus on the English language, which leaves other languages underrepresented. Therefore, in this paper, we compare several modeling techniques for sentiment analysis using a new dataset containing Flemish tweets. The key contribution of our paper lies in its innovative experimental design: we compared different preprocessing techniques and vector representations to find the best-performing combination for a Flemish dataset. We compared models belonging to four different categories: lexicon-based methods, traditional machine-learning models, neural networks, and attention-based models. We found that more preprocessing leads to better results, but the best-performing vector representation approach depends on the model applied. Moreover, an immense gap was observed between the performances of the lexicon-based approaches and those of the other models. The traditional machine learning approaches and the neural networks produced similar results, but the attention-based model was the best-performing technique. Nevertheless, a tradeoff should be made between computational expenses and performance gains.

Mapping innovation in companies for the purpose of official statistics is usually done through business surveys. However, this traditional approach faces several drawbacks like a lack of responses, response bias, low frequency, and high costs. Alternatively, text-based models trained on web-scraped text from company websites have been developed to complement or substitute traditional business surveys. This paper utilises web scraping and text-based models to map the business innovation in Flanders with a focus on identifying different types of innovation through topic modelling. More specifically, the scraped web texts are used to identify innovative economic sectors or topics, and to classify firms into these topics using Top2Vec and Lbl2Vec. We conclude that both models can be successfully combined to discover topics (or sectors) and classify companies into these topics which results in an additional parameter for mapping innovation in different regions.

Data science has become increasingly essential for the production of official statistics, as it enables the automated collection, processing, and analysis of large amounts of data. With such data science practices in place, it enables more timely, more insightful and more flexible reporting. However, the quality and integrity of data-science-driven statistics rely on the accuracy and reliability of the data sources and the machine learning techniques that support them. In particular, changes in data sources are inevitable to occur and pose significant risks that are crucial to address in the context of machine learning for official statistics. This paper gives an overview of the main risks, liabilities, and uncertainties associated with changing data sources in the context of machine learning for official statistics. We provide a checklist of the most prevalent origins and causes of changing data sources; not only on a technical level but also regarding ownership, ethics, regulation, and public perception. Next, we highlight the repercussions of changing data sources on statistical reporting. These include technical effects such as concept drift, bias, availability, validity, accuracy and completeness, but also the neutrality and potential discontinuation of the statistical offering. We offer a few important precautionary measures, such as enhancing robustness in both data sourcing and statistical techniques, and thorough monitoring. In doing so, machine learning-based official statistics can maintain integrity, reliability, consistency, and relevance in policy-making, decision-making, and public discourse.