Romanian Dataset Research Articles

LLaMA 3 vs. State-of-the-Art Large Language Models: Performance in Detecting Nuanced Fake News

This study investigates the effectiveness of a proposed version of Meta’s LLaMA 3 model in detecting fake claims across bilingual (English and Romanian) datasets, focusing on a multi-class approach beyond traditional binary classifications in order to better mimic real-world scenarios. The research employs a proposed version of the LLaMA 3 model, optimized for identifying nuanced categories such as “Mostly True” and “Mostly False”, and compares its performance against leading large language models (LLMs) including Open AI’s ChatGPT versions, Google’s Gemini, and similar LLaMA models. The analysis reveals that the proposed LLaMA 3 model consistently outperforms its base version and older LLaMA models, particularly in the Romanian dataset, achieving the highest accuracy of 39% and demonstrating superior capabilities in identifying nuanced claims, over all the compared large language models. However, the model’s performance across both languages highlights some challenges, with generally low accuracy and difficulties in handling ambiguous categories by all the LLMs. The study also underscores the impact of language and cultural context on model reliability, noting that even state-of-the-art models like ChatGPT 4.o and Gemini exhibit inconsistencies when applied to Romanian text and more than a binary true/false approach.

The digital (il)literacy of local and regional politicians and civil servants –An analysis based on recent data from Romania

In 2021/2022 a consortium of European universities delivered a scientific basis for a policy fighting fake news and hate speech with support of the Congress of Local and Regional Authorities of the Council of Europe, describing also the technical and legal basics of both phenomena. In this context a survey was performed with the delegates of the Congress and the book, which contained also the survey results, was presented to the Congress in its Spring Session 2022. In 2023 a broader survey was done in a selection of countries, which also returned 675 filled-in questionnaires from Romania. The part of the questionnaire dealing with possible remedies contained questions which remedies are considered technically and legally feasible by the participants. Confirming the results of the survey in 2021/2022 the results for Romania showed a severe lack of basic knowledge about how the internet works and its governance. In this paper we want to show the results for Romania in detail, discuss them and suggest possible training measures which seem necessary to make the local and regional politics and administration fit for the Digitalization.The analysis of data from other European countries showed that this phenomenon is not restricted to Romania but is widely recognizable all over Europe. It also confirms the results from the survey among the Congress delegates. Therefore, this paper may be based on the Romanian dataset, however it signifies a pan-European issue.The authors are grateful to Vice Dean Nicolae Urs (Babeș-Bolyai University Cluj-Napoca) and Catalin Vrabie (National University of Political Studies and Public Administration) for their support in spreading the survey among Romanian municipalities.

Evaluation of Language Models on Romanian XQuAD and RoITD datasets

Natural language processing (NLP) has become a vital requirement in a wide range of applications, including machine translation, information retrieval, and text classification. The development and evaluation of NLP models for various languages have received significant attention in recent years, but there has been relatively little work done on comparing the performance of different language models on Romanian data. In particular, the introduction and evaluation of various Romanian language models with multilingual models have barely been comparatively studied. In this paper, we address this gap by evaluating eight NLP models on two Romanian datasets, XQuAD and RoITD. Our experiments and results show that bert-base-multilingual-cased and bertbase- multilingual-uncased, perform best on both XQuAD and RoITD tasks, while RoBERT-small model and DistilBERT models perform the worst. We also discuss the implications of our findings and outline directions for future work in this area.

Deep Learning Approach for Automatic Romanian Lemmatization

This paper proposes a deep learning sequence-to-sequence approach to improve the task of automatic Romanian lemmatization. The study compares 24 systems using different combinations of recurrent, convolutional and attention layers, while the text input consists of word-lemma pairs, both at word and trigram level. As Romanian is a low resourced language in the field of text processing, the aim of this study is to use as little input information as possible. Thus, to increase the lemmatization accuracy, additional lexical features (such as context and POS tags) have been provided only gradually. For the trigrams case, two scenarios were proposed: to predict the lemma for every word from the sequence or to predict the lemma only for the word in the middle.The lemmatizers have been analyzed on two Romanian datasets: the Romanian Explicative Dictionary (DEX) and the belletristic subset of the CoRoLa corpus. For the DEX dataset, the best results were obtained with the LSTM-based systems at both word (99.32%) and character level (99.43%). For the CoRoLa subset, the CNN-based architecture outperforms at trigram (95.86%) and word level (99.09%) while the LSTM-stacked system obtained the highest accuracy at character level (98.78%).

Aerial Point Cloud Classification Using an Alternative Approach for the Dynamic Computation of K-Nearest Neighbors

Abstract The paper reports some methods to select the optimal number of neighbors and to use eigenfeatures for aerial point cloud classification. In the literature, the neighborhood selection is performed using different methods. In this paper, we propose an approach that uses the region growing algorithm. The input data is an aerial point cloud, part of the Romanian Dataset from LAKI II Project. To test our approach, we used a small dataset from the city of Marghita, Bihor County. We report the technical background for classification process and all technical details of the workflow used with insight analyses and comparisons. The work was realized within the VOLTA project (VOLTA, 2017), a RISE Marie-Curie action designed to do research and innovation activities among partners and to exchange knowledge, methods and workflows in the geospatial field.

Concurrent phonetic transcription, lexical stress assignment and syllabification with deep neural networks

Abstract This paper evaluates four different sequence-to-sequence deep neural network architectures aimed to jointly solve the tasks of: phonetic transcription, lexical stress assignment and syllabification. These text processing tasks are considered essential components for high quality text-to-speech or automatic speech recognition systems, with the phonetic transcription being the most frequently used in these types of applications. Although each of the tasks has been individually and extensively analyzed in the scientific literature, there are few studies which target a concurrent solution for them. In general, the lexical stress assignment and syllabification are used as augmenting input features to the phonetic transcription model and not considered as target features. The proposed network architectures include recurrent, convolution and attention neural layers and were evaluated on hand-checked English and Romanian datasets. The accuracy of the models was evaluated in terms of accuracy for the concurrent prediction of all three tasks, as well as by discarding the syllabification or lexical stress predictions. The best results were obtained with a combination of convolution and attention layers, where the accuracy of the joint prediction for the three tasks was of 58.96% for English and 86.64% for Romanian. The same model for English obtains an accuracy of 59.70% when syllables are discarded and 64% when the prediction of lexical stress is ignored. With the same best performing model for Romanian an accuracy of 88.83% without syllables and 93.84% without lexical stress is obtained.

Spatial loess distribution in the eastern Carpathian Basin: a novel approach based on geoscientific maps and data

ABSTRACTGeo- and palaeoecological studies in the Carpathian Basin require a detailed knowledge of the distribution of aeolian sediments. Existing maps are not detailed enough and erroneous as a result of the basic input data and scale used. Here we present a map showing the detailed distribution of loess sediments in the Carpathian Basin at the border of Hungary and Romania using a Geographic Information System and the vectorized, statistically analysed geological map of Hungary (scale 1:300,000) and the Romanian pedological map (scale 1:500,000). Both, the Hungarian and the Romanian data sets were combined and transferred into a common loess sediment classification system resulting in a seamless cross-border map showing the loess distribution in the Carpathian Basin at a scale of about 1:500,000.

The measurement equivalence of personality measures across high- and low-stake test taking settings

Traditionally, organizational psychologists rely on personality measures in order to make personnel selection decisions. However, the extent to which the measurement of personality traits is varies across high-stake test-taking context is not entirely understood. Drawing on a Romanian dataset, we explored different levels of invariance between high and low stake contexts for a celebrated five-factor personality measure. Results revealed that the measurement of five-factor personality traits is non-equivalent across high and low test-taking contexts. Additionally, significant differences in latent means were observed for Neuroticism, Extraversion, Agreeableness and Conscientiousness.

Age driven influence of role‐models on entrepreneurship in a transition economy

PurposeThe purpose of this paper is to examine the differential impact of entrepreneurial role models over entrepreneurial activities at different stages of an individual's life cycle in Romania.Design/methodology/approachTo attain this objective, the paper adopts a socio‐institutional approach to entrepreneurship. The authors carry out a rare‐events logit model using a robust Romanian dataset from 2006 comprising 1,449 individuals.FindingsThe results indicate that the effect of entrepreneurial role models on entrepreneurial activities differs across age groups. The results indicate that for older individuals the positive effect of role models over entrepreneurship is diluted. In contrast, younger individuals, who have been mainly exposed to a market‐based economy, have a more positive societal valuation of entrepreneurial examples, which increases the positive effect that the knowledge of an entrepreneur has on their attitudes towards entrepreneurial activities.Originality/valueThe main contribution of the study lies in the identification of the specific effect that role models have over individuals' level of implication in entrepreneurial activities. This gives ammunition to the argument that, rather than uniform entrepreneurship support programmes, the effective implementation of policies aimed at fostering entrepreneurship should be designed according to the specific profile of the targeted group of beneficiaries. In this case, the implementation of role models as an effective entrepreneurial promotion tool should take into account the differential impact that entrepreneurial examples have over individuals' behaviour towards entrepreneurship.

Automated classification of landform elements using object-based image analysis

This paper presents an automated classification system of landform elements based on object-oriented image analysis. First, several data layers are produced from Digital Terrain Models (DTM): elevation, profile curvature, plan curvature and slope gradient. Second, relatively homogenous objects are delineated at several levels through image segmentation. These object primatives are classified as landform elements using a relative classification model, built both on the surface shape and on the altitudinal position of objects. So far, slope aspect was not used in classification. The classification has nine classes: peaks and toe slopes (defined by the altitudinal position or the degree of dominance), steep slopes and flat/gentle slopes (defined by slope gradients), shoulders and negative contacts (defined by profile curvatures), head slopes, side slopes and nose slopes (defined by plan curvatures). Classes are defined using flexible fuzzy membership functions. Results are visually analyzed by draping them over DTMs. Specific fuzzy classification options were used to obtain an assessment of output accuracy. Two implementations of the methodology are compared using (1) Romanian datasets and (2) Berchtesgaden National Park, Germany. The methodology has proven to be reproducible; readily adaptable for diverse landscapes and datasets; and useful in respect to providing additional information for geomorphological and landscape studies. A major advantage of this new methodology is its transferability, given that it uses only relative values and relative positions to neighboring objects. The methodology introduced in this paper can be used for almost any application where relationships between topographic features and other components of landscapes are to be assessed.

Changes in seasonal mean maximum air temperature in Romania and their connection with large‐scale circulation

AbstractThis paper investigates the temporal and spatial variability of the seasonal mean of maximum air temperature in Romania and its links with the large–scale atmospheric circulation. The Romanian data sets are represented by time series at 14 stations. The large‐scale parameters are represented by the observed sea‐level pressure (SLP) and geopotential height at 500 hPa (Z500). The period analysed was 1922–98 for winter and 1960–98 for all seasons. Before analysis, the original temperature data were tested to detect for inhomogeneity using the standard normal homogeneity test. Empirical orthogonal functions (EOFs) were used to analyse the spatial and temporal variability of the local and large‐scale parameters and to eliminate noise from the original data set. The time series associated with the first EOF pattern of the SLP and mean maximum temperature in Romania were analysed from trend and shifts point of view using the Pettitt and Mann–Kendall tests respectively. The covariance map computed using the Z500 and the seasonal mean of maximum temperature in Romania were used as additional methods to identify the large‐scale circulation patterns influencing the local variability.Significant increasing trends were found for winter and summer mean maximum temperature in Romania, with upward shifts around 1947 and 1985 respectively. During autumn, a decreasing trend with a downward shift around 1969 was detected. These changes seem to be real, since they are connected to similar changes in the large‐scale circulation. So, the intensification of the southwesterly circulation over Europe since 1933 overlapped with the enhancement of westerly circulation after the 1940s could be the reason for the change in winter mean maximum temperature. The slight weakening of the southwesterly circulation during autumn could be one of the reasons for the decrease in the regime of the mean maximum temperature for autumn seasons. Additionally, the covariance map technique reveals the influence of the North Atlantic oscillation in winter, East Atlantic Jet in summer and Scandinavian (or Euroasia‐1) circulation pattern in autumn upon mean maximum air temperature. Copyright © 2002 Royal Meteorological Society.