Hasil untuk "Nutrition. Foods and food supply"

Hunt Allcott, Rebecca Diamond, Jean-Pierre Dubé et al.

We study the causes of “nutritional inequality”: why the wealthy eat more healthfully than the poor in the United States. Exploiting supermarket entry and household moves to healthier neighborhoods, we reject that neighborhood environments contribute meaningfully to nutritional inequality. We then estimate a structural model of grocery demand, using a new instrument exploiting the combination of grocery retail chains’ differing presence across geographic markets with their differing comparative advantages across product groups. Counterfactual simulations show that exposing low-income households to the same products and prices available to high-income households reduces nutritional inequality by only about 10%, while the remaining 90% is driven by differences in demand. These findings counter the argument that policies to increase the supply of healthy groceries could play an important role in reducing nutritional inequality.

I. Creed, A. Bergström, C. Trick et al.

Weidi Luo, Xiaofei Wen, Tenghao Huang et al.

Large language models (LLMs) are increasingly deployed for everyday tasks, including food preparation and health-related guidance. However, food safety remains a high-stakes domain where inaccurate or misleading information can cause severe real-world harm. Despite these risks, current LLMs and safety guardrails lack rigorous alignment tailored to domain-specific food hazards. To address this gap, we introduce FoodGuardBench, the first comprehensive benchmark comprising 3,339 queries grounded in FDA guidelines, designed to evaluate the safety and robustness of LLMs. By constructing a taxonomy of food safety principles and employing representative jailbreak attacks (e.g., AutoDAN and PAP), we systematically evaluate existing LLMs and guardrails. Our evaluation results reveal three critical vulnerabilities: First, current LLMs exhibit sparse safety alignment in the food-related domain, easily succumbing to a few canonical jailbreak strategies. Second, when compromised, LLMs frequently generate actionable yet harmful instructions, inadvertently empowering malicious actors and posing tangible risks. Third, existing LLM-based guardrails systematically overlook these domain-specific threats, failing to detect a substantial volume of malicious inputs. To mitigate these vulnerabilities, we introduce FoodGuard-4B, a specialized guardrail model fine-tuned on our datasets to safeguard LLMs within food-related domains.

Aydin Kilic

ABSTRACT This article focused on the comprehensive evaluation of statistical criteria applied in common mathematical models selected for experimental cold drying data for thin‐layer food drying applications. In this context, Mackerel ( Trachurus trachurus ), known as a functional and sensitive food sample with its bioactive content, was selected as the experimental material for drying applications. For this purpose, four experimental groups (G5MM, G10MM, G15MM, G20MM) with different sample thicknesses (5, 10, 15, 20 mm) at 100 g were dried with 6 m/s air flow at 10°C for 24, 22, 20, and 14 h respectively. Twenty‐three common semi‐theoretic and empiric mathematical models were applied to the obtained drying values. For the comprehensive evaluation of the models, non‐linear regression analysis was performed using 13 different statistical criteria such as r , RSS , SST , SSE , R 2 , χ 2 , RMSE , residuals, RSSE , MBE , EF , SEE , and p . In this context, in the study where the relevant criteria were applied, for G20MM, Newton Lewis, Midilli‐Küçük, Balbay and Şahin, Page, for G15MM, Henderson & Pabis, Logarithmic (Asymptotic), Binomial, Verma et al., Modified Henderson, Simplified Fick diff., Balbay and Şahin model were concluded to be the most suitable. In addition, for G10MM, Logarithmic (Asymptotic), Demir et al., Binary, Verma et al., Balbay and Şahin, Thompson and Alibas models, and in the G05MM group, Logarithmic (Asymptotic), Demir et al., Binary, Verma et al., Thompson, Balbay‐Şahin and Alibas models were concluded to be the most suitable. According to the results obtained, it has been revealed that using only r , R 2 , χ 2 and RMSE equations instead of 13 statistical criteria in the evaluation of mathematical models gives significant and meaningful results.

Chanakan Khemthong, Sirinapa Thangsiri, Wimonphan Chathiran et al.

Tea, the world's most popular beverage, is prepared using different pre- and post-harvesting techniques, leading to its unique sensory characteristics and bioactive ingredients. However, knowledge about the impact of these processing methods on the phenolic compositions and health properties of commercially available ready-to-drink teas is limited. To fill this research lacuna, the phenolic compositions, antioxidant potentials, and inhibitory activities against the key enzymes relevant to non-communicable diseases including hyperlipidemia (lipase), type II diabetes (α-amylase, α-glucosidase, and dipeptidyl peptidase-IV), and Alzheimer's disease (acetylcholinesterase, butyrylcholinesterase, and β-secretase) of eleven commercially available ready-to-drink teas were investigated. The results indicated that pre-harvest shading led to higher contents of catechin, epicatechin gallate, chlorogenic acid, and 3,4-dihydroxybenzoic acid, while fermentation significantly increased rutin content, which, in turn, led to higher total phenolic contents. Antioxidants in shaded tea tended to follow a hydrogen atom transfer (HAT)-based mechanism, while the single electron transfer (SET)-based mechanism was a preferable reaction pathway for antioxidants in fermented teas. The ability to inhibit key enzymes was more pronounced in fermented teas than in shaded tea, potentially attributed to the biological activities of certain phenolics, either individually or working together as effective inhibitors. The addition of flavors (honey lemon, kyoho grape, watermelon, and roasted rice) led to higher contents of catechins than the original green tea (without added flavor), while the contents of other general phenolics varied, leading to similar or higher TPCs. Antioxidant potentials and enzyme inhibitions varied for different ready-to-drink tea varieties. The knowledge from this research will benefit the development of ready-to-drink teas with particular phenolics and health-related properties.

Kangzi Ren, Guilin Chen, Ziyi Zhang et al.

To investigate the regioselectivity of lipase-catalyzed synthesis of important sugar-laurate esters and their functionalities as emulsifiers and antimicrobial agents, glucose, galactose, mannose, maltose and trehalose were used. 6-O-lauryl glucose (Glu-L), 6-O-lauryl galactose (Gal-L), 6-O-lauryl mannose (Man-L), 6′-O-lauryl maltose (Mal-L) and 6-O-lauryl trehalose/6’-O-lauryl trehalose (Tre-L) were synthesized using lipase from Candida antarctica. The Glu-L and Man-L achieved the highest yields (65.49 % and 58.16 %, respectively). The synthesized esters produced smaller-particle-size (1.9–3.1 μm), but less stable emulsions than the commercial sucrose esters (4.2–8.1 μm). The zeta-potential data revealed Man-L and Gal-L had higher surface coverage than the Glu-L on the oil droplets, while the Mal-L and Tre-L had similar surface coverage. The Man-L, Mal-L and Tre-L demonstrated the superior foamability and foam stability. The Gal-L, Man-L, Mal-L, and Tre-L inhibited E.coli 12024 (MIC 1–4 mg/mL), while only Man-L, Mal-L, and Tre-L inhibited B.subtilis 5009 (MIC 0.5–2.0 mg/mL).

Tian Hu, Taotao Wang, Xiaojing Luo et al.

BackgroundThe C-reactive protein-albumin-lymphocyte (CALLY) index, a novel inflammation-immune-nutritional biomarker, has not been comprehensively evaluated for mortality risk prediction in older populations. Here, we investigate the relationship between the CALLY index and all-cause mortality in Chinese adults aged ≥ 60 years.MethodsData were obtained from the 2014 to 2018 wave of the Chinese Longitudinal Healthy Longevity Survey (CLHLS). Upon applying a natural logarithmic transformation to the CALLY index, the lnCALLY was stratified into tertiles. Kaplan-Meier analysis and the log-rank test were employed to assess the cumulative survival probability across lnCALLY-stratified older adults populations. Cox proportional hazards regression was utilized to investigate the association between lnCALLY and all-cause mortality. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were conducted to evaluate the predictive capacity of lnCALLY for all-cause mortality. Restricted cubic splines (RCS) with four knots were applied to explore the potential non-linear dose-response association of lnCALLY with all-cause mortality. Subgroup analyses and sensitivity analyses were conducted to ensure validity.ResultsA total of 1,738 older adults participants were included in this cohort. Over a median follow-up of 3.3 years, 580 deaths (33.3%) occurred. The multivariable Cox regression demonstrated that the highest lnCALLY tertile was associated with a 40% reduced mortality risk compared to the lowest tertile [adjusted hazard ratio (HR) = 0.60, 95% confidence interval (CI): 0.49–0.73]. Kaplan-Meier curves revealed significantly higher survival probabilities in individuals with elevated lnCALLY (P < 0.001). Time-dependent ROC analysis showed that the AUC of lnCALLY for predicting all-cause mortality at 1-, 2-, and 3-year were 0.751, 0.746, and 0.762, respectively. RCS demonstrated an approximate “L”-shaped negative correlation between lnCALLY and all-cause mortality (Poverall < 0.001, Pnon–linearity = 0.364). Subgroup and sensitivity analyses confirmed robustness, with no significant interactions observed across demographic or clinical strata.ConclusionThese findings suggest that the CALLY index serves as a practical prognostic biomarker for monitoring survival in older populations, underscoring the interplay of inflammation, immunity, and nutrition in aging-related mortality.

Fengwang Xue, Qingmei Fang, Kuangyang Yu et al.

BackgroundLong-term or high-dose glucocorticoid administration can markedly impair immune responses, mask clinical indicators of pulmonary infections, and increase the susceptibility to refractory pneumonia, leading to heightened mortality risk. The Prognostic nutritional index (PNI), derived from peripheral lymphocyte count and serum albumin (ALB) levels, serves as a reliable indicator for evaluating nutritional and immune statuses across various clinical populations, including oncology patients, individuals with cardiovascular disorders, and perioperative patients. However, the predictive value of PNI in pneumonia patients receiving glucocorticoids, especially within the Chinese population, has not been sufficiently investigated. This observational analysis aimed to explore the correlation between PNI levels and all-cause mortality (ACM) in patients undergoing prolonged glucocorticoid therapy for pneumonia.MethodsA retrospective cohort study was conducted utilizing data extracted from the Dryad database. Kaplan–Meier curves, multivariable Cox regression, restricted cubic splines (RCS), and subgroup analyses were used to assess the association between PNI and ACM in patients with pneumonia who received glucocorticoids.ResultsThe study incorporated a total of 639 pneumonia patients who received glucocorticoid therapy. The ACM rates were 22.5% at 30 days and rose to 26.0% at 90 days. Multivariable Cox regression showed that, after full adjustment for potential confounders, every 2-unit decrease in PNI was associated with a 10% higher 30-day mortality hazard (HR = 1.10, 95% CI = 1.05–1.15, p < 0.001) and a 9% higher 90-day mortality hazard (HR = 1.09, 95% CI = 1.04–1.14, p < 0.001). Compared with patients with PNI ≥ 43, patients with PNI < 43 had a 118% increased risk of 30-day mortality (HR = 2.18, 95% CI = 1.28–3.81, p = 0.005) and a 96% increased risk of 90-day mortality (HR = 1.96, 95% CI = 1.20–3.19, p = 0.008). Further validation using RCS analysis revealed a robust inverse relationship between PNI scores and ACM, and subgroup analyses revealed no significant interactions.ConclusionAmong pneumonia patients receiving glucocorticoid therapy, a decreased PNI was associated with an increased risk of 30-day and 90-day mortality, particularly in those with a PNI < 43.

Yun-Shi Dai, Peng-Fei Dai, Stéphane Goutte et al.

With escalating macroeconomic uncertainty, the risk interlinkages between energy and food markets have become increasingly complex, posing serious challenges to global energy and food security. This paper proposes an integrated framework combining the GJRSK model, the time-frequency connectedness analysis, and the random forest method to systematically investigate the moment connectedness within the energy-food nexus and explore the key drivers of various spillover effects. The results reveal significant multidimensional risk spillovers with pronounced time variation, heterogeneity, and crisis sensitivity. Return and skewness connectedness are primarily driven by short-term spillovers, kurtosis connectedness is more prominent over the medium term, while volatility connectedness is dominated by long-term dynamics. Notably, crude oil consistently serves as a central transmitter in diverse connectedness networks. Furthermore, the spillover effects are influenced by multiple factors, including macro-financial conditions, oil supply-demand fundamentals, policy uncertainties, and climate-related shocks, with the core drivers of connectedness varying considerably across different moments and timescales. These findings provide valuable insights for the coordinated governance of energy and food markets, the improvement of multilayered risk early-warning systems, and the optimization of investment strategies.

D Bhanu Prakash, D K K Vamsi

In recent years, the study on the impact of competition on additional food provided prey-predator systems have gained significant attention from researchers in the field of mathematical biology. In this study, we consider an additional food provided prey-predator model exhibiting Holling type-IV functional response and the intra-specific competition among predators. We prove the existence and uniqueness of global positive solutions for the proposed model. We study the existence and stability of equilibrium points and further explore the codimension-$1$ and $2$ bifurcations with respect to the additional food and competition. We further study the global dynamics of the system and discuss the consequences of providing additional food. Later, we do the time-optimal control studies with respect to the quality and quantity of additional food as control variables by transforming the independent variable in the control system. Making use of the Pontraygin maximum principle, we characterize the optimal quality of additional food and optimal quantity of additional food. We show that the findings of these dynamics and control studies have the potential to be applied to a variety of problems in pest management.

Moh. Hairus Zaman, Diah Wahyuningsih, Ris Yuwono Yudo Nugroho

Price uncertainty in food commodities can create uncertainty for farmers and potentially negatively impact the level of farmer household well-being. On the other hand, the agriculture sector in the province of East Java has greatly contributed to East Java's economy. This paper analyses the response of farmer welfare through farmer exchange values amidst fluctuation shock of food needed prices and inflation level in the east java province. The research method of this paper employs the impulse response function of the Bayesian Vector Autoregressive (BVAR) model by using time series secondary data from May 2017 until December 2023. This paper finds that the shock that happens to aggregate food prices can increase farmer exchange values even though the shock to the inflation level has reduced farmer exchange values and increased aggregate food prices.

David Simchi-Levi, Konstantina Mellou, Ishai Menache et al.

Supply Chain Management requires addressing a variety of complex decision-making challenges, from sourcing strategies to planning and execution. Over the last few decades, advances in computation and information technologies have enabled the transition from manual, intuition and experience-based decision-making, into more automated and data-driven decisions using a variety of tools that apply optimization techniques. These techniques use mathematical methods to improve decision-making. Unfortunately, business planners and executives still need to spend considerable time and effort to (i) understand and explain the recommendations coming out of these technologies; (ii) analyze various scenarios and answer what-if questions; and (iii) update the mathematical models used in these tools to reflect current business environments. Addressing these challenges requires involving data science teams and/or the technology providers to explain results or make the necessary changes in the technology and hence significantly slows down decision making. Motivated by the recent advances in Large Language Models (LLMs), we report how this disruptive technology can democratize supply chain technology - namely, facilitate the understanding of tools' outcomes, as well as the interaction with supply chain tools without human-in-the-loop. Specifically, we report how we apply LLMs to address the three challenges described above, thus substantially reducing the time to decision from days and weeks to minutes and hours as well as dramatically increasing planners' and executives' productivity and impact.

Tisha Ghosh

The wastage of perishable items has led to significant health and economic crises, increasing business uncertainty and fluctuating customer demand. This issue is worsened by online food delivery services, where frequent and unpredictable orders create inefficiencies in supply chain management, contributing to the bullwhip effect. This effect results in stockouts, excess inventory, and inefficiencies. Accurate demand forecasting helps stabilize inventory, optimize supplier orders, and reduce waste. This paper presents a Third-Party Logistics (3PL) supply chain model involving restaurants, online food apps, and customers, along with a deep learning-based demand forecasting model using a two-phase Long Short-Term Memory (LSTM) network. Phase one, intra-day forecasting, captures short-term variations, while phase two, daily forecasting, predicts overall demand. A two-year dataset from January 2023 to January 2025 from Swiggy and Zomato is used, employing discrete event simulation and grid search for optimal LSTM hyperparameters. The proposed method is evaluated using RMSE, MAE, and R-squared score, with R-squared as the primary accuracy measure. Phase one achieves an R-squared score of 0.69 for Zomato and 0.71 for Swiggy with a training time of 12 minutes, while phase two improves to 0.88 for Zomato and 0.90 for Swiggy with a training time of 8 minutes. To mitigate demand fluctuations, restaurant inventory is dynamically managed using the newsvendor model, adjusted based on forecasted demand. The proposed framework significantly reduces the bullwhip effect, improving forecasting accuracy and supply chain efficiency. For phase one, supply chain instability decreases from 2.61 to 0.96, and for phase two, from 2.19 to 0.80. This demonstrates the model's effectiveness in minimizing food waste and maintaining optimal restaurant inventory levels.

P. Pingali

There is a growing disconnect between agricultural policy and contemporary nutritional challenges, the persistent problem of micronutrient malnutrition and child stunting, as well as the emerging challenges of overweight and obesity. Diversification of production systems and the market supply of enhanced diversity will only happen when the current distortions to farm and market level incentives are corrected. Data on the diet transition that is taking place across the developing world is presented and the growing divergence between staple crop demand and supply trends discussed. The reasons for the low producer response to rising demand for non-staple food, such as vegetables, are examined. Finally, the paper presents the main elements of a crop neutral agricultural policy, one that creates a level playing field which allows farmers to respond to market signals rather than a policy that is biased toward a particular set of crops.

EFSA Panel on Plant Health (PLH), Claude Bragard, Paula Baptista et al.

Abstract Following a request from the European Commission, the EFSA Panel on Plant Health performed a quantitative risk assessment of Leucinodes orbonalis (Lepidoptera: Crambidae), the eggplant fruit and shoot borer, for the EU. The assessment focused on potential pathways for entry, climatic conditions favouring establishment, spread and impact. Options for risk reduction are discussed but effectiveness was not quantified. L. orbonalis is a key pest of eggplant (aubergine/brinjal) in the Indian subcontinent and occurs throughout most of southern Asia with records mostly from India and Bangladesh. The main pathway of entry is fruit of solanaceous plants, primarily exotic varieties of eggplant, Solanum melongena and turkey berry, S. torvum. The trade in both commodities from Asia is small but nevertheless dwarfs the trade in other Solanum fruits from Asia (S. aethiopicum, S. anguivi, S. virginianum, S. aculeatissimum, S. undatum). Other Solanum fruits were therefore not further assessed as potential pathways. The trade in eggplant from Asia consists of special fruit types and caters mostly to niche markets in the EU, while most eggplant consumed in Europe is produced in southern European and northern African countries, where L. orbonalis does not occur. Using expert knowledge elicitation (EKE) and pathway modelling, the Panel estimated that approximately 3–670 infested fruit (90% certainty range, CR) of S. melongena or fruit bunches of S. torvum enter into regions of the EU that are suitable for L. orbonalis establishment each year. Based on CLIMEX modelling, and using two possible thresholds of ecoclimatic index (EI) to indicate uncertainty in establishment potential, climates favouring establishment occur mostly in southern Europe, where, based on human population, approximately 14% of the imported produce is distributed across NUTS2 regions where EI ≥ 30; or 23% of the produce is distributed where EI ≥ 15. Escape of adult moths occurs mostly from consumer waste. By analysing results of different scenarios for the proportion of S. melongena and S. torvum in the trade, and considering uncertainties in the climatic suitability of southern Europe, adult moth emergence in areas suitable for establishment is expected to vary between 84 individuals per year and one individual per 40 years (based on 90% CR in different scenarios). In the baseline scenario, 25% of the solanaceous fruit from Asia is S. torvum, 75% is S. melongena and EI ≥ 30 is required for establishment. After accounting for the chances of mating, host finding and establishment, the probability of a mated female establishing a founder population in the EU is less than 1 in 100,000 to about 1 event per 622 years (90% CR in baseline scenario). The waiting time until the first establishment is then 622 to more than 100,000 years (CR). If such a founder population were established, the moth is estimated to spread at a rate of 0.65–7.0 km per year after a lag phase of 5–92 years. The impact of the insect on the production of eggplant is estimated to be 0.67%–13% (CR) if growers take no specific action against the insect and 0.13%–1.9% if they do take targeted actions. Tomato (S. lycopersicum) and potato (S. tuberosum) are hosts of L. orbonalis, but the insect does not develop to maturity in tomato fruit, and it does not feed on potato tubers under field conditions; hence, damage to potato can only occur due to feeding on shoots. Tomato and potato are not preferred hosts; nevertheless, impact can occur if populations of L. orbonalis are high and preferred hosts are not available. The Panel did not assess this damage due to insufficient information.

Muhammad Fuad Anshori, Yunus Musa, Muh Farid et al.

Implementing a cropping intensity program with rice cultivation four times a year (CI 400) can be achieved using early maturing varieties of rice. However, this development needs to pay attention to the adaptability of the varieties planted to ensure successful implementation. The adaptability approach is a combination of assessing stability and productivity potential. This concept has been developed and applied in several studies, including research on rice. However, this approach is considered less comprehensive because it is non-parametric and only focuses on one stability analysis. Therefore, a systematic integration of various stability analyses, including index methods, is needed to comprehensively assess adaptability, particularly for early-maturing rice in South Sulawesi. This region is characterized by a dynamic climate zone and is one of the top four highest rice producers in Indonesia. Meanwhile, this study aims to develop a comprehensive adaptability index and select the best early-maturing rice varieties, especially in South Sulawesi. The investigation was conducted in Bone, Soppeng, and Gowa over two seasons using a nested randomized complete block design, with organized replications in each environment (location-season). Additionally, there was a significant focus on the application of five early-maturing and two check rice varieties, with each factor repeated three times at each location, totaling 126 experimental units. The results showed that the adaptability index, by combining stability rank accumulation with yield min max standardization, was effective at assessing the yield potential and stability of early-maturing rice varieties in supporting CI 400. Inpari 13 had the best index value at 0.55, followed by Cakrabuana at 0.31; hence both were recommended as adaptive early-maturing rice varieties, especially in South Sulawesi.

Iolanda Cioffi, Daniela Martini, Cristian Del Bo’ et al.

Plant-based diets represent a valid strategy to improve human health and increase food sustainability. The availability of legume-based products, a good source of proteins and fibers, could help consumers to promote healthy dietary patterns. The aim of this study was to examine the impact of different legume-based pastas on energy intake and appetite in healthy volunteers. Four ad libitum (protocol 1) and iso-caloric pre-load meals (protocol 2) were tested using a randomized repeated measure design. The test meals consisted of lentils pasta (LP), chickpeas pasta (CP); durum wheat pasta (DWP) and gluten free pasta (GFP), served with tomato sauce. Protocol 1: the ad libitum lunch meal was consumed then EI registered. Protocol 2: subjective appetite was assessed by visual analogue scale before and after the pre-load meal for 2 h until an ad libitum buffet was served to assess EI. Twenty (age: 39.2 ± 8.41 years; BMI: 23.4 ± 3.4 kg/m2) and 40 (age: 42.6 ± 8.7 years; BMI: 23.8 ± 4.2 kg/m2) healthy subjects were respectively recruited for each protocol. ANCOVA analysis showed an overall effect of meals and sex on EI within meal and at the subsequent meal, resulting in a lower EI after LP compared to DWP (p < 0.05). Appetite sensations were significantly influenced solely after the pre-load meal, where repeated measures ANCOVA showed increased post-prandial satiety after LP and CP (p < 0.05) compared to DWP in females, whereas a reduction in desire to eat and higher fullness was found following LP compared to the other meals in both sexes (p < 0.05). Overall, lentil-based pasta seemed to acutely affect EI both within and at the subsequent meal, especially in females. Consumption of legume-based pasta might enhance legume intake by modulating appetite feelings and increasing food sustainability. However, further studies are needed to support these results in the long-term and considering different target populations.

Xinyue Pan, Jiangpeng He, Fengqing Zhu

Food image classification is the fundamental step in image-based dietary assessment, which aims to estimate participants' nutrient intake from eating occasion images. A common challenge of food images is the intra-class diversity and inter-class similarity, which can significantly hinder classification performance. To address this issue, we introduce a novel multi-modal contrastive learning framework called FMiFood, which learns more discriminative features by integrating additional contextual information, such as food category text descriptions, to enhance classification accuracy. Specifically, we propose a flexible matching technique that improves the similarity matching between text and image embeddings to focus on multiple key information. Furthermore, we incorporate the classification objectives into the framework and explore the use of GPT-4 to enrich the text descriptions and provide more detailed context. Our method demonstrates improved performance on both the UPMC-101 and VFN datasets compared to existing methods.

Takuya Sera, Izumi Kuwata, Yuki Taya et al.

This study explores a new method in food development by utilizing AI including generative AI, aiming to craft products that delight the senses and resonate with consumers' emotions. The food ingredient recommendation approach used in this study can be considered as a form of multimodal generation in a broad sense, as it takes text as input and outputs food ingredient candidates. This Study focused on producing "Romance Bread," a collection of breads infused with flavors that reflect the nuances of a romantic Japanese television program. We analyzed conversations from TV programs and lyrics from songs featuring fruits and sweets to recommend ingredients that express romantic feelings. Based on these recommendations, the bread developers then considered the flavoring of the bread and developed new bread varieties. The research included a tasting evaluation involving 31 participants and interviews with the product developers. Findings indicate a notable correlation between tastes generated by AI and human preferences. This study validates the concept of using AI in food innovation and highlights the broad potential for developing unique consumer experiences that focus on emotional engagement through AI and human collaboration.

Fnu Mohbat, Mohammed J. Zaki

In the rapidly evolving landscape of online recipe sharing within a globalized context, there has been a notable surge in research towards comprehending and generating food recipes. Recent advancements in large language models (LLMs) like GPT-2 and LLaVA have paved the way for Natural Language Processing (NLP) approaches to delve deeper into various facets of food-related tasks, encompassing ingredient recognition and comprehensive recipe generation. Despite impressive performance and multi-modal adaptability of LLMs, domain-specific training remains paramount for their effective application. This work evaluates existing LLMs for recipe generation and proposes LLaVA-Chef, a novel model trained on a curated dataset of diverse recipe prompts in a multi-stage approach. First, we refine the mapping of visual food image embeddings to the language space. Second, we adapt LLaVA to the food domain by fine-tuning it on relevant recipe data. Third, we utilize diverse prompts to enhance the model's recipe comprehension. Finally, we improve the linguistic quality of generated recipes by penalizing the model with a custom loss function. LLaVA-Chef demonstrates impressive improvements over pretrained LLMs and prior works. A detailed qualitative analysis reveals that LLaVA-Chef generates more detailed recipes with precise ingredient mentions, compared to existing approaches.