Publications

Recent advances in vision-language models (VLMs) have improved image captioning for cultural heritage. However, inferring structured cultural metadata (e.g., creator, origin, period) from visual input remains underexplored. We introduce a multi-category, cross-cultural benchmark for this task and evaluate VLMs using an LLM-as-Judge framework that measures semantic alignment with reference annotations. To assess cultural reasoning, we report exact-match, partial-match, and attribute-level accuracy across cultural regions. Results show that models capture fragmented signals and exhibit substantial performance variation across cultures and metadata types, leading to inconsistent and weakly grounded predictions. These findings highlight the limitations of current VLMs in structured cultural metadata inference beyond visual perception.

Cross-cultural competence in large language models (LLMs) requires understanding and adapting Culture-Specific Items (CSIs) across varying cultural contexts. However, progress in evaluating this capability remains limited by the lack of high-quality CSI-annotated corpora with parallel cross-cultural sentence pairs. We introduce XCR-Bench, a Cross(X)-Cultural Reasoning Benchmark containing 4.1k parallel sentences and 1,098 CSIs across three reasoning tasks. XCR-Bench integrates Newmark’s CSI framework with Hall’s Triad of Culture, enabling evaluation across levels of cultural visibility, from observable practices to implicit social norms and values. Experiments on eight multilingual LLMs show that state-of-the-art models exhibit consistent weaknesses in identifying and adapting specific categories of CSIs, revealing a gap between surface-level recall and explicit cultural reasoning. Performance declines significantly on culturally sensitive categories and deeper cultural levels (p<0.005, 8/8 models), and adaptation quality varies systematically across target cultures and Bengali regional variants, indicating encoded regional and ethno-religious biases even within a single linguistic setting. We publicly release the corpus and code to support future research on cross-cultural NLP.

When a question has valid answers under different normative frameworks, a language model must decide which framework to use and whether it can answer correctly within it. We call this setting normative pluralism and study it in Islamic finance using a four-choice taxonomy that separates framework selection from within-framework correctness. This separation reveals the stereotype trap: a cultural cue steers a model toward one framework, but the model selects an incorrect answer within that framework. Across twelve models, two languages, and fifty demographic signals, cultural cues change framework selection and reveal substantial differences in accuracy, especially among non-frontier models. Under the strongest signal, large open-weight models select the Islamic framework 97% of the time. A two-choice evaluation would report near-perfect alignment, although 57–66% of those selections are incorrect. These findings motivate, but do not directly test, the competence-conditioned routing hypothesis: models may favor frameworks where they are more accurate, while cultural cues may expose framework-specific competence gaps.

Machine Translation (MT) is widely employed to address resource scarcity in low-resource languages by generating synthetic data from high-resource counterparts. While sentiment preservation in translation has long been studied, a critical but underexplored factor is the role of cultural alignment between source and target languages. In this paper, we hypothesize that semantic labels are drifted or altered during MT due to cultural divergence. Through a series of experiments across culturally sensitive and neutral domains, we establish three key findings: (1) MT systems, including modern Large Language Models (LLMs), induce label drift during translation, particularly in culturally sensitive domains; (2) unlike earlier statistical MT tools, LLMs encode cultural knowledge, and leveraging this knowledge can amplify label drift; and (3) cultural similarity or dissimilarity between source and target languages is a crucial determinant of label preservation. Our findings highlight that neglecting cultural factors in MT not only undermines label fidelity but also risks misinterpretation and cultural conflict in downstream applications.

A large number of studies rely on closed-style multiple-choice surveys to evaluate cultural alignment in Large Language Models (LLMs). In this work, we challenge this constrained evaluation paradigm and explore more realistic, unconstrained approaches. Using the World Values Survey (WVS) and Hofstede Cultural Dimensions as case studies, we demonstrate that LLMs exhibit stronger cultural alignment in less constrained settings, where responses are not forced. Additionally, we show that even minor changes, such as reordering survey choices, lead to inconsistent outputs, exposing the limitations of closed-style evaluations. Our findings advocate for more robust and flexible evaluation frameworks that focus on specific cultural proxies, encouraging more nuanced and accurate assessments of cultural alignment in LLMs.

LLM deception is often evaluated through direct markers such as fabricated claims, explicit lies, or strategic concealment. However, many real-world misleading communications do not depend on false statements, rather, they arise from selective treatment of true material facts: omitting adverse evidence, softening unfavorable details, emphasizing favorable details, or replacing precise qualifications with vague language. Existing benchmarks largely miss this subtler and arguably more dangerous failure mode. We introduce JANUS, a benchmark for measuring goal-conditioned pragmatic distortion in fact-grounded LLM outputs. Each scenario in our benchmark provides a fixed pool of favorable and adverse facts and compares a neutral condition against a goal-directed condition, such as increasing adoption, enrollment, approval, or support, despite potential harm to directly affected individuals or groups. Because all outputs are constrained to use the same fact pool, JANUS isolates misleading net impressions from hallucination and fabrication. JANUS contains 160 scenarios across 8 domains, with each scenario paired with neutral and goal-conditioned prompts and annotated material facts. Extensive experiments across 12 LLMs reveal consistent goal-conditioned distortions, demonstrating that current models remain sensitive to incentive and framing objectives and lack robust safeguards against selectively misleading communication. We publicly release our corpus and code for future research.

Large language models (LLMs) have been widely applied across various domains of finance. Since their training data are largely derived from human-authored corpora, LLMs may inherit a range of human biases. Behavioral biases can lead to instability and uncertainty in decision-making, particularly when processing financial information. However, existing research on LLM bias has mainly focused on direct questioning or simplified, general-purpose settings, with limited consideration of the complex real-world financial environments and high-risk, context-sensitive, multilingual financial misinformation detection tasks (MFMD). In this work, we propose MFMDScen, a comprehensive benchmark for evaluating behavioral biases of LLMs in MFMD across diverse economic scenarios. In collaboration with financial experts, we construct three types of complex financial scenarios: (i) role- and personality-based, (ii) role- and region-based, and (iii) role-based scenarios incorporating ethnicity and religious beliefs. We further develop a multilingual financial misinformation dataset covering English, Chinese, Greek, and Bengali. By integrating these scenarios with misinformation claims, MFMDScen enables a systematic evaluation of 22 mainstream LLMs. Our findings reveal that pronounced behavioral biases persist across both commercial and open-source models.

Current research on bias in language models (LMs) predominantly focuses on data quality, with significantly less attention paid to model architecture and temporal influences of data. Even more critically, few studies systematically investigate the origins of bias. We propose a methodology grounded in comparative behavioral theory to interpret the complex interaction between training data and model architecture in bias propagation during language modeling. Building on recent work that relates transformers to n-gram LMs, we evaluate how data, model design choices, and temporal dynamics affect bias propagation. Our findings reveal that: (1) n-gram LMs are highly sensitive to context window size in bias propagation, while transformers demonstrate architectural robustness; (2) the temporal provenance of training data significantly affects bias; and (3) different model architectures respond differentially to controlled bias injection, with certain biases (e.g. sexual orientation) being disproportionately amplified. As language models become ubiquitous, our findings highlight the need for a holistic approach, tracing bias to its origins across both data and model dimensions, not just symptoms, to mitigate harm.

The widespread adoption of language models highlights the need for critical examinations of their inherent biases, particularly concerning religion. This study systematically investigates religious bias in both language models and text-to-image generation models, analyzing both open-source and closed-source systems. We construct approximately 400 unique, naturally occurring prompts to probe language models for religious bias across diverse tasks, including mask filling, prompt completion, and image generation. Our experiments reveal concerning instances of underlying stereotypes and biases associated disproportionately with certain religions. Additionally, we explore cross-domain biases, examining how religious bias intersects with demographic factors such as gender, age, and nationality. This study further evaluates the effectiveness of targeted debiasing techniques by employing corrective prompts designed to mitigate the identified biases. Our findings demonstrate that language models continue to exhibit significant biases in both text and image generation tasks, emphasizing the urgent need to develop fairer language models to achieve global acceptability.

Despite the promise of large language models (LLMs) in finance, their capabilities for financial misinformation detection (FMD) remain largely unexplored. To evaluate the capabilities of LLMs in FMD task, we introduce the financial misinformation detection shared task featured at COLING FinNLP-FNP-LLMFinLegal-2024, FMD Challenge. This challenge aims to evaluate the ability of LLMs to verify financial misinformation while generating plausible explanations. In this paper, we provide an overview of this task and dataset, summarize participants’ methods, and present their experimental evaluations, highlighting the effectiveness of LLMs in addressing the FMD task. To the best of our knowledge, the FMD Challenge is one of the first challenges for assessing LLMs in the field of FMD. Therefore, we provide detailed observations and draw conclusions for the future development of this field.

Progress in mental health research remains constrained by the accessibility of adequate, high-quality data. Annotating mental health data requires a lot of resources and expert monitoring. In this study, we explore the utility of active learning with Human-in-the-Loop labeling approach to reduce the annotation task for identifying signs of depression in people’s social media posts. The data for this study was collected from the #WorldMentalHealthDay trend on Twitter, which is a popular mental health campaign to raise awareness of mental illness across the global population. From the pool of unlabeled data, we initially labeled a small portion of data to train an LSTM model with GloVe embedding; thereafter, the entire pool was labeled using uncertainty sampling, labeling the least confident data in each cycle. Along with the methodology, we present a high-quality dataset of 3659 samples, with a notable proportion of 22% of tweets indicating symptoms of depression. We also analyze the language usage of depressed and non-depressed individuals on social media by dissecting the semantic structure of tweets. The quality of the dataset was validated by establishing strong baseline results with state-of-the-art models and word-embedding techniques.

Mental health research through data-driven methods has been hindered by a lack of standard typology and scarcity of adequate data. In this study, we leverage the clinical articulation of depression to build a typology for social media texts for detecting the severity of depression. It emulates the standard clinical assessment procedure Diagnostic and Statistical Manual of Mental Disorders (DSM-5) and Patient Health Questionnaire (PHQ-9) to encompass subtle indications of depressive disorders from tweets. Along with the typology, we present a new dataset of 40191 tweets labeled by expert annotators. Each tweet is labeled as ’non-depressed’ or ’depressed’. Moreover, three severity levels are considered for ’depressed’ tweets: (1) mild, (2) moderate, and (3) severe. An associated confidence score is provided with each label to validate the quality of annotation. We examine the quality of the dataset via representing summary statistics while setting strong baseline results using attention-based models like BERT and DistilBERT. Finally, we extensively address the limitations of the study to provide directions for further research.

The influence of social media is one of the most dominating characteristics of the current era, and this has led cyberbullying to grow into a more serious social issue. As a result, automated cyberbullying detection systems need to be an integral part of almost all social media platforms. Past studies on this domain have primarily focused on hand-picked features and traditional machine learning approaches for cyberbullying detection from user comments on social media. Recently, transformers have been proved to be quite effective in various language-related tasks; however, their effectiveness has not been extensively explored in this particular domain. In this study, we evaluate the individual performance of several well-known transformer-based architectures and aim to contribute to the development of automated cyberbullying detection systems by proposing our own transformer-based ensemble framework. Our proposed framework is evaluated on a balanced and an imbalanced dataset, both of which are constructed from a large collection of Twitter comments and are publicly available. Our proposed architecture outperforms all the baseline models, as well the individual standalone classifier networks that are used in our ensemble, obtaining an average F1-score of 95.92% on the balanced dataset and an average F1-score of 87.51% on the imbalanced dataset. We further investigate the cases where our proposed architecture misclassifies samples from both datasets, preventing it from achieving a perfect score.

People can easily reveal their aggressive remarks on social media platforms using the anonymity it provides. During the COVID-19 pandemic, the usage of social media has been increased several times according to surveys and people are vulnerable to cyber attacks now more than ever. Prevention of cyberbullying needs careful monitoring and identification. Most of the existing works on cyberbullying detection employed traditional machine learning classifiers with handcrafted features, and deep learning-based models have made their way in this domain very recently. Categorizing cyberbullying based on traits is a complex task and needs contextual consideration. In this work, we have proposed a new approach to detect cyberbullying on social media platforms using a neural ensemble method of transformer-based architectures with attention mechanism. Our proposed architecture is trained on one balanced and one imbalanced dataset and outperforms the given ML and DNN baselines by a significant margin in both cases. We achieved an average F1-score of 95.59% for five classes and 90.65% for six classes on the Fine-Grained Cyberbullying Dataset (FGCD), and 87.28% on Twitter parsed dataset. Our in-depth results provide great insights into the effectiveness of transformer-based models in cyberbullying detection and paves the way for future researches to combat this serious online issue.

Sarcasm is a form of figurative language that serves as a humorous tool for mockery and ridicule. We present a novel architecture for sarcasm generation with emoji from a non-sarcastic input sentence in English. We divide the generation task into two sub tasks: one for generating textual sarcasm and another for collecting emojis associated with those sarcastic sentences. Two key elements of sarcasm are incorporated into the textual sarcasm generation task: valence reversal and semantic incongruity with context, where the context may involve shared commonsense or general knowledge between the speaker and their audience. The majority of existing sarcasm generation works have focused on this textual form. However, in the real world, when written texts fall short of effectively capturing the emotional cues of spoken and face-to-face communication, people often opt for emojis to accurately express their emotions. Due to the wide range of applications of emojis, incorporating appropriate emojis to generate textual sarcastic sentences helps advance sarcasm generation. We conclude our study by evaluating the generated sarcastic sentences using human judgement.

Sarcasm can be defined as saying or writing the opposite of what one truly wants to express, usually to insult, irritate, or amuse someone. Because of the obscure nature of sarcasm in textual data, detecting it is difficult and of great interest to the sentiment analysis research community. Though the research in sarcasm detection spans more than a decade, some significant advancements have been made recently, including employing unsupervised pre-trained transformers in multimodal environments and integrating context to identify sarcasm. In this study, we aim to provide a brief overview of recent advancements and trends in computational sarcasm research for the English language. We describe relevant datasets, methodologies, trends, issues, challenges, and tasks relating to sarcasm that are beyond detection. Our study provides well-summarized tables of sarcasm datasets, sarcastic features and their extraction methods, and performance analysis of various approaches which can help researchers in related domains understand current state-of-the-art practices in sarcasm detection.

Data augmentation involves generating synthetic samples that resemble those in a given dataset. In resource-limited fields where high-quality data is scarce, augmentation plays a crucial role in increasing the volume of training data. This paper introduces a Bangla Text Data Augmentation (BDA) Framework that uses both pre-trained models and rule-based methods to create new variants of the text. A filtering process is included to ensure that the new text keeps the same meaning as the original while also adding variety in the words used. We conduct a comprehensive evaluation of the framework’s effectiveness in Bangla text classification tasks. Our framework achieved significant improvement in F1 scores across five distinct datasets, delivering performance equivalent to models trained on 100% of the data while utilizing only 50% of the training dataset. Additionally, we explore the impact of data scarcity by progressively reducing the training data and augmenting it through BDA, resulting in notable F1 score enhancements. The study offers a thorough examination of BDA’s performance, identifying key factors for optimal results and addressing its limitations through detailed analysis.

Large Language Models (LLMs) have emerged as one of the most important breakthroughs in NLP for their impressive skills in language generation and other language-specific tasks. Though LLMs have been evaluated in various tasks, mostly in English, they have not yet undergone thorough evaluation in under-resourced languages such as Bengali (Bangla). To this end, this paper introduces BenLLM-Eval, which consists of a comprehensive evaluation of LLMs to benchmark their performance in the Bengali language that has modest resources. In this regard, we select various important and diverse Bengali NLP tasks, such as text summarization, question answering, paraphrasing, natural language inference, transliteration, text classification, and sentiment analysis for zero-shot evaluation of popular LLMs, namely, GPT-3.5, LLaMA-2-13b-chat, and Claude-2. Our experimental results demonstrate that while in some Bengali NLP tasks, zero-shot LLMs could achieve performance on par, or even better than current SOTA fine-tuned models; in most tasks, their performance is quite poor (with the performance of open-source LLMs like LLaMA-2-13b-chat being significantly bad) in comparison to the current SOTA results. Therefore, it calls for further efforts to develop a better understanding of LLMs in modest-resourced languages like Bengali.

This paper introduces a novel informal Bangla word embedding for designing a cost-efficient solution for the task "Violence Inciting Text Detection" which focuses on developing classification systems to categorize violence that can potentially incite further violent actions. We propose a semi-supervised learning approach by training an informal Bangla FastText embedding, which is further fine-tuned on lightweight models on task specific dataset and yielded competitive results to our initial method using BanglaBERT, which secured the 7th position with an f1-score of 73.98%. We conduct extensive experiments to assess the efficiency of the proposed embedding and how well it generalizes in terms of violence classification, along with its coverage on the task’s dataset. Our proposed Bangla IFT embedding achieved a competitive macro average F1 score of 70.45%. Additionally, we provide a detailed analysis of our findings, delving into potential causes of misclassification in the detection of violence-inciting text.

The analysis of consumer sentiment, as expressed through reviews, can provide a wealth of insight regarding the quality of a product. While the study of sentiment analysis has been widely explored in many popular languages, relatively less attention has been given to the Bangla language, mostly due to a lack of relevant data and cross-domain adaptability. To address this limitation, we present BanglaBook, a large-scale dataset of Bangla book reviews consisting of 158,065 samples classified into three broad categories: positive, negative, and neutral. We provide a detailed statistical analysis of the dataset and employ a range of machine learning models to establish baselines including SVM, LSTM, and Bangla-BERT. Our findings demonstrate a substantial performance advantage of pre-trained models over models that rely on manually crafted features, emphasizing the necessity for additional training resources in this domain. Additionally, we conduct an in-depth error analysis by examining sentiment unigrams, which may provide insight into common classification errors in under-resourced languages like Bangla.

Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field’s evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks and examine benchmark saturation, contamination, and reporting mismatches. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference, and identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable.

The art of mathematical reasoning stands as a fundamental pillar of intellectual progress and is a central catalyst in cultivating human ingenuity. Researchers have recently published a plethora of works centered around the task of solving Math Word Problems (MWP), a crucial stride towards general AI. These existing models are susceptible to dependency on shallow heuristics and spurious correlations to derive the solution expressions. In order to ameliorate this issue, in this paper, we propose a framework for MWP solvers based on the generation of linguistic variants of the problem text. The approach involves solving each of the variant problems and electing the predicted expression with the majority of the votes. We use DeBERTa (Decoding-enhanced BERT with disentangled attention) as the encoder to leverage its rich textual representations and enhanced mask decoder to construct the solution expressions. Furthermore, we introduce a challenging dataset, ParaMAWPS, consisting of paraphrased, adversarial, and inverse variants of selectively sampled MWPs from the benchmark MAWPS dataset. We extensively experiment on this dataset along with other benchmark datasets using some baseline MWP solver models. We show that training on linguistic variants of problem statements and voting on candidate predictions improve the mathematical reasoning and robustness of the model.

We evaluated contemporary Islamic lifestyle applications supporting religious practices and motivation among Muslims. We reviewed 11 popular applications using self-determination theory and the technology-as-experience framework to assess their support for motivation and affective needs. Most applications lack features that foster autonomy, competence, and relatedness. We also interviewed ten devoted Muslim application users to gain insights into their experiences and unmet needs. Our findings indicate that existing applications fall short in providing comprehensive learning, social connections, and scholar consultations. We propose design implications based on our results, including guided religious information, shareability, virtual community engagement, scholarly question-answering, and personalized reminders. We aim to inform the design of Islamic lifestyle applications that better facilitate ritual practices, benefitting application designers and Muslim communities. Our research provides valuable insights into the untapped potential for lifestyle applications to act as religious companions supporting Muslims’ spiritual journey.

Motion-based gaming can aptly substitute an individual’s daily exercise requirements while being a great source of entertainment. Conversion of sedentary-based games into motion-based form can reach more gamers who are otherwise avoiding daily exercises. Toward that end, there is scope for research on the factors to be considered to retain a greater number of players by recreating the games they are already accustomed to playing. Hence, we developed the motion-based forms of two popular sedentary games, Flappy Bird, and Temple Run. We then conducted a study incorporating both of the games with a group of participants. By analyzing the user experience through feedback and observation, we determined the key factors involved in converting conventional games into motion-based form. Our study shows that if specific considerations are made, motion-based forms of sedentary games have the potential to be more or similarly appealing and can achieve greater retention when compared to their sedentary counterpart.

Non-Functional Requirements (NFRs) play a critical role in determining the overall quality and user satisfaction of software systems. Accurately identifying and classifying NFRs is essential to ensure that software meets performance, usability, and reliability expectations. However, manual identification of NFRs from documentation is time-consuming and prone to errors, necessitating automated solutions. Before implementing any automated solution, a robust and comprehensive dataset is essential. To build such a dataset, we collected NFRs from various Project Charters and Open Source Software Documentation. This enhanced the technical depth and usability of an already existing NFR dataset. We categorized NFRs into sub-classes and identified needs using widely used Large Language Models to facilitate automation. After classifying the NFRs, we compared the classification results of the selected LLMs: RoBERTa, CodeBERT, Gemma-2, Phi-3, Mistral-8B, and Llama-3.1-8B using various evaluation metrics, including precision, recall, F1-score, and lime scores. Among these models, Gemma-2 achieved the best results with a precision of 0.87, recall of 0.89, and F1-score of 0.88, alongside a lime hit score of 78 out of 80. Phi-3 closely followed with a precision of 0.85, recall of 0.87, F1-score of 0.86, and the highest lime hit score of 79.

Sharing cooking recipes is a great way to exchange culinary ideas and provide instructions for food preparation. However, categorizing raw recipes found online into appropriate food genres can be challenging due to a lack of adequate labeled data. In this study, we present a dataset named the Assorted, Archetypal, and Annotated Two Million Extended (3A2M+) Cooking Recipe Dataset that contains two million culinary recipes labeled in respective categories with extended named entities extracted from recipe descriptions. This collection of data includes various features such as title, NER, directions, and extended NER, as well as nine different labels representing genres including bakery, drinks, non-veg, vegetables, fast food, cereals, meals, sides, and fusions. The proposed pipeline named 3A2M+ extends the size of the Named Entity Recognition (NER) list to address missing named entities like heat, time or process from the recipe directions using two NER extraction tools. 3A2M+ dataset provides a comprehensive solution to the various challenging recipe-related tasks, including classification, named entity recognition, and recipe generation. Furthermore, we have demonstrated traditional machine learning, deep learning and pre-trained language models to classify the recipes into their corresponding genre and achieved an overall accuracy of 98.6%. Our investigation indicates that the title feature played a more significant role in classifying the genre.

The term "asteroid family" refers to a collection of asteroids that share similar proper orbital elements such as semi-major axis, eccentricities, and orbital inclinations. Detecting small asteroid families has proved to be a challenge for a long time because of their extremely low sample size. In general, standalone machine learning classifiers tend to exhibit a bias towards classes with larger sample sizes, resulting in the inadequate classification of asteroid families with limited data. In this paper, a two-step supervised model was proposed for the effective classification of the asteroid families, especially for the tiny, small, and lower groups of medium asteroid families. The proposed model uses two-step classification in an attempt to resolve the challenges that come with the imbalanced dataset where at first a binary classification of small and large families was done with an XGB (Extreme Gradient boosting) classifier and then in the second stage Random Forest classifier was used alongside previously identified binary features to classify asteroid families. The proposed model performed better with higher F1 scores for tiny and small asteroid families compared to other algorithms tested in this work. It also achieved a perfect F1 score for 90 families, among 112 families which were tested. As for the lower group of medium sized asteroid families, it performed slightly worse compared to the previously used machine learning algorithms. Along with this, four dataset imbalance handling techniques have been employed in this work and compared to the proposed algorithm.

Cooking recipes allow individuals to exchange culinary ideas and provide food preparation instructions. Due to a lack of adequate labeled data, categorizing raw recipes found online to the appropriate food genres is a challenging task in this domain. Utilizing the knowledge of domain experts to categorize recipes could be a solution. In this study, we present a novel dataset of two million culinary recipes labeled in respective categories leveraging the knowledge of food experts and an active learning technique. To construct the dataset, we collect the recipes from the RecipeNLG dataset. Then, we employ three human experts whose trustworthiness score is higher than 86.667% to categorize 300K recipe by their Named Entity Recognition (NER) and assign it to one of the nine categories: bakery, drinks, non-veg, vegetables, fast food, cereals, meals, sides and fusion. Finally, we categorize the remaining 1900K recipes using Active Learning method with a blend of Query-by-Committee and Human In The Loop (HITL) approaches. There are more than two million recipes in our dataset, each of which is categorized and has a confidence score linked with it. For the 9 genres, the Fleiss Kappa score of this massive dataset is roughly 0.56026. We believe that the research community can use this dataset to perform various machine learning tasks such as recipe genre classification, recipe generation of a specific genre, new recipe creation, etc.