RESEARCH ARTICLE

International Research Journal on Advanced Science Hub

2582-4376
www.rspsciencehub.com
Vol. 06, Issue 11 November

RSP Science Hub http://dx.doi.org/10.47392/IRJASH.2024.044

Study and Overview of Recipe Generators

Shravani Halaye1, Gauri Rane2, Khushbu Mahale3, Anjali Gupta4, Dr. Divya Tamma5
1,2,3,4
UG Scholar, Department of AI&DS, Rajiv Gandhi Institute of Tech., Mumbai, Maharashtra, India.
5
Assistant Professor, Department of AI&DS, Rajiv Gandhi Institute of Tech., Mumbai, Maharashtra, India.
Emails: [email protected], [email protected], [email protected],
[email protected], [email protected]

Article history Abstract

Received: 09 November 2024 This paper presents a literature survey focused on the development and
Accepted: 12 November 2024 effectiveness of recipe generators and nutrient tracking applications. The study
Published: 23 November 2024
explores the evolution of these tools, their underlying technologies, and their
Keywords: role in promoting healthier eating habits and personalized nutrition. Recipe
Ingredient substitution, Meal generators leverage algorithms to recommend meals based on user
planning, Nutrient tracking, preferences, dietary restrictions, and available ingredients, while nutrient apps
Personalized nutrition, Recipe provide detailed insights into food composition and help users monitor their
generator dietary intake. The survey examines various case studies to highlight successful
implementations, identifying key features such as ingredient substitution tools,
meal planning functionalities, and real-time nutrient tracking. Additionally, the
study investigates the challenges faced by these systems, including data
accuracy, user engagement, and personalization limitations. The findings offer
insights into how these digital solutions contribute to health management and
propose future directions for improving their functionality through
advancements in AI, user experience design, and interoperability with other
wellness platforms.

1. Introduction
As technology continues to penetrate everyday life, that suit the user's specific health conditions,
the way individuals plan meals and manage preferences, and restrictions. Recipe generators and
nutrition has undergone a digital transformation. nutrient tracking tools offer significant potential to
Recipe generators and nutrition apps have emerged enhance user’s dietary habits, promoting healthier
as essential tools, facilitating personalized meal lifestyles by providing access to dynamic meal
planning, ingredient tracking, and nutritional suggestions and diet-based recommendations.
management. These systems aim to help users make Automated recipe generation systems aim to
informed dietary choices by leveraging simplify the decision-making process by curating
technologies like artificial intelligence (AI), natural dishes based on user input, available ingredients,
language processing (NLP), and data analytics. and dietary preferences. Studies have explored the
With the global increase in lifestyle-related diseases use of AI-based recommendation models to suggest
such as diabetes, heart disease, and obesity, there is recipes aligned with users' personal tastes and
growing emphasis on creating tailored meal plans available pantry items. These systems are further

OPEN ACCESS 336

Shravani Halaye et al 2024, Vol. 06, Issue 11 November
enhanced by nutrient tracking apps, which offer scoring recipes selected for future iterations.
users a platform to monitor calorie intake, Finally, the recipes are transformed into coherent
macronutrients, and specific nutrients such as text formats by traversing recipe trees and merging
vitamins or minerals. Despite the promise of these instructions. The findings reveal varying levels of
tools, challenges remain. Developing effective recipe quality, with a focus on validity, edibility,
ingredient substitution algorithms that provide and creativity, while also highlighting the potential
suitable alternatives without compromising taste or for future enhancements in recipe generation
nutrition is still an area of active research. through improved data on food combinations. This
Additionally, user engagement is critical for the paper [2] also uses AutoChef, an autonomous recipe
success of these platforms, as users need to feel generator designed specifically for the Indian
empowered by the recommendations to build long- diaspora, filling a gap in recipe generation research.
term healthy habits. Another challenge lies in Using a genetic algorithm inspired by natural
achieving interoperability across platforms, such as evolution, the system selects and refines the best
linking recipe apps with wearable fitness devices or recipes to create new, innovative culinary
health tracking systems. Furthermore, ensuring data combinations. The methodology evaluates recipes
privacy and security is paramount, especially as based on tastiness, creativity, validity, and edibility.
users share personal health information to generate Results show that users rated the recipes highly for
customized meal recommendations. The objective creativity, while validity and edibility received
of this literature survey is to provide an in-depth moderate scores, indicating that the generated
exploration of current advancements in AI-powered recipes are generally understandable and
recipe generation and nutrition tracking acceptable. In addition to the genetic algorithm, the
applications. It seeks to identify key technologies, authors implemented several machine learning
trends, and gaps within the field, highlighting both techniques, particularly in the realm of Natural
the opportunities and limitations of existing Language Processing (NLP). They utilized neural
systems. By synthesizing research findings from network language models and N-gram models to
various sources, this study aims to offer insights analyze ingredient combinations and cooking
into how recipe recommendation systems and instructions effectively. The study also incorporated
nutrition apps are evolving to meet the growing an ensemble approach that combined classifiers
demand for personalized dietary tools. such as Naive Bayes, Multinomial Logistic
2. Literature Survey Regression, and Random Forest. This ensemble
[1] Explores the evaluation of cooking recipes method achieved an impressive accuracy of 79% in
through parameters such as correctness and predicting the cuisine based on ingredient sets,
innovation, utilizing survey results to gauge showcasing the effectiveness of the algorithms
understandability and willingness to cook. It used. The paper highlights the importance of
introduces AutoChef, an evolutionary algorithm ingredient relationships and the mapping of these
that generates valid and innovative recipes by relationships to improve recipe generation. The
leveraging existing culinary knowledge, employing paper [3] explores the classification of cuisines
techniques like natural language processing and through structured recipes, emphasizing the
genetic programming. Recipes are structured as significance of cooking processes and utensils
trees, with nodes representing cooking actions and alongside ingredients. The authors applied various
ingredients, enabling the evolution of recipes over classification techniques, including traditional
generations. The system employs a fitness function machine learning models and advanced neural
to evaluate ingredient combinations, cooking networks. The methodology of the study focused on
methods, and instruction clarity, ensuring that only classifying cuisines using the RecipeDB dataset,
high-quality recipes are selected for further which comprises 118,071 recipes, 20,280 unique
refinement. Knowledge is extracted from user- ingredients, 256 unique processes, and 69 unique
generated online recipes, providing a foundation for utensils. Given the dataset's high sparsity ratio of
creating new, valid, and appealing dishes. 99.50%, effective preprocessing was crucial. The
AutoChef manages a population of recipes, pairing preprocessing steps involved tokenization,
and evaluating old and new versions, with higher- lemmatization, and the removal of digits and

International Research Journal on Advanced Science Hub (IRJASH) 337

Study and Overview of Recipe Generators 2024, Vol. 06, Issue 11 November
symbols to minimize noise, resulting in a refined and interpretation. These findings highlight the
dataset containing 20,400 distinct entities. These potential of NLP methods in recipe generation and
entities were subsequently vectorized using TF-IDF the importance of context in ingredient substitution.
and word embedding techniques to prepare them for [5]Presents a novel AI system designed to enhance
analysis. The study employed a combination of personalized nutrition management, particularly for
statistical and sequential classification models. The patients with chronic diseases. The primary
statistical models included Logistic Regression, objective of the project is to facilitate daily
Naive Bayes, Support Vector Machines (SVM), and assessments of nutrient intake, enabling tailored
Random Forest, while the sequential models dietary adjustments to meet individual nutritional
comprised Long Short-Term Memory (LSTM) needs. The methodology encompasses several key
networks and Transformers, specifically BERT and stages, starting with data collection, where patients
RoBERTa. The classification of cuisines through are referred by healthcare professionals, and
structured recipes emphasizes the importance of nutritionists compile essential information such as
ingredients, cooking processes, and utensils in medical history and biological assessments. This
defining culinary identity. By treating recipes as data is crucial for determining dietary restrictions
sequentially structured data, the methodology based on factors like height, weight, age, and
enhances classification accuracy and provides physical activity. The system employs advanced
deeper insights into the relationships among various image processing techniques, including
culinary features. This comprehensive approach classification, segmentation, and detection, to
leads to more meaningful results in culinary identify food items from images. The dataset used
research and applications. [4]The project for training the system includes thousands of
investigates the application of natural language images of various food items, which are essential
processing (NLP) techniques to assist in cooking by for accurate classification and nutrient estimation.
suggesting alternative ingredients and generating Subsequently, machine learning algorithms are
new recipes. The dataset was collected from utilized to estimate the nutrient content of these
Spoonacular, comprising 3,433 recipes across 15 identified food items, enhancing the accuracy of
cuisine styles, with an average of 229 recipes per nutrient intake assessments. The system aims for an
cuisine. During data preprocessing, the original accuracy improvement of over 90% compared to
cooking instructions were preserved, and a new traditional methods, which often yield biased
feature called "process" was created to pair cooking results due to reliance on self-reported data. A
activities with ingredients, enhancing the significant focus of the project is on improving
understanding of ingredient usage in various accuracy, as traditional dietary assessment methods
contexts. The project employed the Skip-gram often suffer from biases due to reliance on self-
model to obtain word embeddings for ingredients, reported data, particularly among patients with
measuring similarity through cosine similarity of cognitive impairments. The AI-based approach
their embedding vectors. Two language models aims to mitigate these issues by reducing human
were developed: the traditional N-gram model, error and providing real-time analysis of food
which often produces repetitive text, and the Long images. Initial results suggest that the AI system has
Short-Term Memory (LSTM) model, which the potential to replace conventional food
generates more coherent recipes with recognizable assessment methods, with ongoing efforts directed
steps and ingredients. The study found that recipes towards refining its accuracy and usability for better
generated by the LSTM model included an average patient health outcomes. The paper [6] discusses the
of seven recognizable cooking steps and eight development of a system for Food Nutritional
ingredients, while those generated by the N-gram Detection, Visualization, and Recommendation
model included five cooking steps and three aimed at health monitoring using image processing
ingredients. Additionally, the project employs word techniques. It addresses the challenges of
embedding methods to suggest alternative maintaining a healthy lifestyle in a busy world by
ingredients based on similarity, emphasizing the enabling users to identify food items through
need for further data collection and collaboration images, view their calorie content, and receive
with culinary experts to enhance recipe generation health recommendations. The system is structured

International Research Journal on Advanced Science Hub (IRJASH) 338

Shravani Halaye et al 2024, Vol. 06, Issue 11 November
into two main sections: an admin section for and conference proceedings related to food
managing food data and a user-friendly Android recognition, calorie measurement, and nutrition
application that allows users to upload food images. monitoring using technology, particularly focusing
Key processes involved in the system include on mobile devices and wearable systems. Key
converting images to greyscale, segmentation, topics highlighted include artificial ripening
histogram calculation, feature extraction, and detection, personalized healthcare
classification using algorithms such as Convolution recommendations, food recognition systems
Neural Networks (CNN). Additionally, the system tailored for diabetic patients, and calorie
incorporates a crowdsourcing approach to keep the management tools. The studies utilize advanced
food database updated, ensuring that users have methods such as deep learning, image processing,
access to the latest nutritional information. Overall, and augmented reality to enhance dietary
the project aims to assist users in managing their assessment and monitoring, further emphasizing the
calorie intake and making healthier food choices. importance of technology in promoting healthier
The paper also references various research studies eating habits. [7-9]

2.1 Tables
Table 1 List of Recipes Combinations
Recipe ID Continent Cuisine Recipe
2610 African Middle Eastern ['water', 'red lentil', 'rom tomato', ,smooth', 'stir', 'heat']
3957 Asian Southeast Asian ['olive oil', 'onion', 'garlic', 'ginger' 'stir', 'add', 'cook 'season', 'garnish', 'pot']
Indian ['coconut milk', 'milk', 'white sugar', 'basmati rice' 'stir', 'cook', 'saucepan',
4153 Asian
Subcontinent 'bowl']
Latin
79897 Mexican ['beef', 'chunky salsa', 'mushroom', 'garlic', 'heat', 'simmer', 'serve'. 'skillet']
American
['oven buttermilk biscuit', 'onion', 'cream', ..., 'spread', 'sprinkle', 'bake',
138976 European Deutschland
'pan']
North
149191 Canadian ['raisin', 'fig', 'water', 'date', 'butter' 'chill', 'cut', 'bowl', 'processor', 'pan']
American

Table 1 shows the recipe ID, continent, cuisine, and water, red lentil, rom tomato, smooth, stir, and heat.
ingredients required for each recipe. The recipes are Similarly, the recipe with ID 3957, from Southeast
from African, Asian, Latin American, European, Asian cuisine, requires olive oil, onion, garlic,
and North American cuisines. The ingredients ginger, stir, add, cook, season, garnish, and pot.
required for each recipe are listed as a string of This table provides a concise overview of the
words, with each word representing a specific different recipes and their corresponding
ingredient. For example, the recipe with ID 2610, ingredients.
from African cuisine, requires the ingredients

Table 2 The Performance of the Models

Model Accuracy (%) Loss Precision Recall F1 Score
Logistic Regression 57.70 1.51 0.56 0.57 0.56
Naïve Bayes 51.64 7.14 0.50 0.51 0.50
SVM (Linear) 56.60 2.97 0.54 0.56 0.54
Random Forest 50.37 2.32 0.48 0.50 0.49
LSTM 53.61 1.65 0.53 0.54 0.53
Transformers (BERT) 68.71 0.21 0.58 0.60 0.57
RoBERTa 73.30 0.10 0.67 0.71 0.69

Table 2 shows the accuracy, loss, precision, recall, include Logistic Regression, Naïve Bayes, SVM,
and F1 score of different models used for Random Forest, LSTM, Transformers (BERT) and
information extraction from a recipe. The models RoBERTa. The models were evaluated on a dataset

International Research Journal on Advanced Science Hub (IRJASH) 339

Study and Overview of Recipe Generators 2024, Vol. 06, Issue 11 November
of recipes and their performance is presented in the In [1] , the process starts with Datasets, which are
table. RoBERTa achieved the highest accuracy fed into the Recipe Analysis stage. This stage is
(73.30%) and F1 score (0.69), indicating it's the composed of Entity Recognition, Recipe Analysis,
most effective model for this task. and Adjacency Matrices, which work together to
2.2 Figures identify and map the relationships between
ingredients and actions. The output of this analysis,
combined with User Input, is passed to the
Evolutionary Algorithm, which generates New
Recipes based on both data-driven insights and user
preferences. This architecture enables the
automated creation of novel recipes, leveraging
evolutionary algorithms to optimize combinations
and outcomes.
3. Results and Discussion
3.1 Results
The studies under review have shown the
usefulness of AI-based recipe generators, as well as
nutrition apps, while they have also highlighted a
few limitations. Particularly, numerous distinct
methods, including evolutionary algorithms,
genetic programming, and NLP models, were used
to generate new and applicable recipes. Tasks in
these studies indicate the need to manage multiple
parameters on creativity, validity, edibility, and
Figure 1 Information Extraction from a Recipe tastiness, implying that both data-driven and
heuristic methods are essential to user satisfaction.
Figure 1 shows Information extraction from a 3.2 Discussions
recipe [1] It illustrates the design flow for automatic The most widely used method in these studies
digitization of a recipe, breaking down the recipe seems to be the genetic algorithm, which evolves
into three main components: Ingredients, Actions, the recipes over many generations, concentrating on
and Recipe Instructions. The left column lists the the improvement of ingredient combinations and
ingredients, while the middle column specifies the the technology to prepare the dish. This is the case
actions applied to these ingredients. The right in AutoChef, which considers genes coded in trees
column combines the ingredients and actions into to be recipes [1]. The fitness functions guarantee
step-by-step instructions. This structured process that the best recipes are only chosen for further
enables the systematic extraction of recipe data, refinement on the scales of flavor and ease of
which can be used for digital applications such as understanding of the recipe. Cuisine classification
automated cooking guidance or smart kitchen and ingredient prediction tasks are some of the tasks
systems. Figure 2 represents the technique used by that had cross discipline algorithms such as Support
the authors and outlines the workflow of the vector machines (SVM), Naive bayes classifiers,
AutoChef system, which automates the generation Random forests, and neural networks (particularly,
of new recipes through an evolutionary algorithm. LSTMs and Transformers). A significant
improvement in model performance is observed
with Transformers which include BERT and
RoBERTa as these performed better than other
traditional classifiers with accuracy of over 73%.
[3]Other ways in which the systems were improved
include Natural Language Processing techniques
such as Skip-gram, and N-gram models which were
Figure 2 Autochef System Overview

International Research Journal on Advanced Science Hub (IRJASH) 340

Shravani Halaye et al 2024, Vol. 06, Issue 11 November
used to aid in generating more sensible ingredient [5]. K. Azzimani, H. Bihri, A. Dahmi, S.
substitutions and recipes that made more sense.[4] Azzouzi and M. E. H. Charaf, "An AI Based
Conclusion Approach for Personalized Nutrition and
AI-based recipe generators and nutrition systems Food Menu Planning," 2022 IEEE 3rd
are transforming culinary innovation and International Conference on Electronics,
personalized health management through advanced Control, Optimization and Computer
techniques like evolutionary algorithms, genetic Science (ICECOCS), Fez, Morocco, 2022,
programming, NLP, and machine learning models. pp. 1-5,
Platforms like AutoChef highlight the potential for doi:10.1109/ICECOCS55148.2022.998309
creative recipe generation, though challenges 9.
remain in balancing creativity with practicality. [6]. P. R. Kaushik, P. H. M, R. S. Srinivas, S.
Systems integrating nutritional tracking with real- Puri and A. M, "Automated Recipe
time assessments offer promising solutions for Generation using Ingredient Classification
healthcare applications. These AI-driven tools have based on an Image from a Real-Time Photo
the potential to revolutionize personalized cooking Station," 2023 4th International Conference
and nutrition, promoting healthier dietary choices for Emerging Technology (INCET),
and better user engagement. Continuous innovation Belgaum, India, 2023, pp.1-6,
will be key to addressing current limitations and doi:10.1109/INCET57972.2023.10170563.
unlocking their full potential. [7]. R. Rewane and P. M. Chouragade, "Food
References Nutritional Detection, Visualization and
[1]. H. Jabeen, J. Weinz and J. Lehmann, Recommendation for Health Monitoring
"AutoChef: Automated Generation of using Image Processing," 2019 3rd
Cooking Recipes," 2020 IEEE Congress on International Conference on Trends in
Evolutionary Computation (CEC), Electronics and Informatics (ICOEI),
Glasgow, UK, 2020, pp. 1-7, doi: Tirunelveli, India, 2019, pp. 556-559, doi:
10.1109/CEC48606.2020.9185605. 10.1109/ICOEI.2019.8862550.
[2]. S. Chaudhary, B. Soni, A. Sindhavad, A. [8]. P. Chhikara, D. Chaurasia, Y. Jiang, O.
Mamaniya, A. Dalvi and I. Siddavatam, Masur and F. Ilievski, "FIRE: Food Image
"ChefAI.IN: Generating Indian Recipes to REcipe generation," 2024 IEEE/CVF
with AI Algorithm," 2022 International Winter Conference on Applications of
Conference on Trends in Quantum Computer Vision (WACV), Waikoloa, HI,
Computing and Emerging Business USA, 2024, pp. 8169-8179, doi:
Technologies (TQCEBT), Pune, India, 10.1109/WACV57701.2024.00800.
2022, pp. 1-6, doi: [9]. R. R. Hariadi, W. N. Khotimah and E. A.
10.1109/TQCEBT54229.2022.10041463. Wiyono, "Design and implementation of
[3]. Prashant T. Sharma, U. Upadhyay and G. food nutrition information system using
Bagler, "Classification of Cuisines from SURF and FatSecret API," 2015
Sequentially Structured Recipes," 2020 International Conference on Advanced
IEEE 36th International Conference on Data Mechatronics, Intelligent Manufacture, and
Engineering Workshops (ICDEW), Dallas, Industrial Automation (ICAMIMIA),
TX, USA, 2020, pp. 105-108, doi: Surabaya, Indonesia, 2015, pp. 181-183,
10.1109/ICDEW49219.2020.00008. doi: 10.1109/ICAMIMIA.2015.7508027.
[4]. Y. Pan, Q. Xu and Y. Li, "Food Recipe
Alternation and Generation with Natural
Language Processing Techniques," 2020
IEEE 36th International Conference on Data
Engineering Workshops (ICDEW), Dallas,
TX, USA, 2020, pp. 94-97, doi:
10.1109/ICDEW49219.2020.000-1.

International Research Journal on Advanced Science Hub (IRJASH) 341