Recommendation systems have become complex algorithms that analyze user data and preferences, to provide personalized recommendations. These systems play an important role in helping users make decisions in many diferent fields. Recently, a significant diference in restaurant selection preferences has been observed among individuals. To this end, this paper presents an innovative system named a restaurant recommendation system based on collaborative filtering using machine learning algorithms and the multi-criteria method (CRMS) that uses machine learning and the Analytical Hierarchy Process (AHP), to enhance dining experiences through personalized restaurant recommendations. The CRMS provides a framework for comparing user preferences with restaurant attributes, allowing data-driven decisions to be made in choosing restaurants, taking into account other users' ratings and location. The primary goal of this system is to develop an application that attempts to recommend restaurants that match the user's preferences using the multi-criteria Ahp method, as well as their geographical location and user ratings using machine learning algorithms based on collaborative filtering. The results of the proposed system show that it helps users find restaurants according to their preferences and aspirations to provide better recommendations.
Recommender systems are vital tools for enhancing user experiences by ofering personalized and
relevant information. These intelligent tools employ sophisticated algorithms and techniques to analyze
user preferences, historical behavior, and contextual data, with the goal of providing personalized
suggestions and recommendations. Moreover, these systems analyze users data and preferences in order
to match users with suitable products, services, or content, leading to increased customer satisfaction
and engagement [
In the remainder of the paper, Section 2 presents previous studies of food and restaurant recommendation systems. Then Section 3, Methods and materials, which details the components and algorithms used in the proposed approach. Section 4 describes the experimental results including the system implementation . Section 5 describes deployment of website and the final section concludes the research and presents future directions for the work.
Restaurant recommendation systems play a pivotal role in enhancing individuals’ well-being by tailoring recommendations based on their food preferences and convenience. Our review categorizes related works into four sections: Traditional Collaborative Filtering (CF) Systems (Section 2.1), Content-Based Recommendation Systems (Section 2.2), Hybrid Recommendation Systems (Section 2.3), and MultiCriteria Decision Making (MCDM) Systems (Section 2.4).
Daniel and Amalia.[
Gupta et al. [
Shirisha et al. [
Amari et al.[
We noticed that the realm of dining recommendation systems, research by (Danielle and Amalia, Abdul-Jabbar, Gurung, Karabila et al. and Kouahla et al.) highlights the eficacy of traditional collaborative filtering (CF) techniques such as user-based collaborative filtering, matrix factorization, and sentiment analysis, tailored for the restaurant domain. However, traditional CF systems often fall short in considering multiple criteria and user preferences beyond ratings, leading to the emergence of content-based recommendation systems emphasized by (Gupta et al., Kosem and Prihandi., Parihar and Pérez-Almaguer et al.) accommodating user preferences and dietary requirements. To address these limitations, hybrid recommendation systems integrating CF, content-based filtering, sentiment analysis, and AI algorithms, as demonstrated by(Sherisha et al., Nandan and Gupta and Keya et al.) deliver personalized and diverse restaurant recommendations catering to users’ preferences and health conditions. Additionally, multi-criteria decision-making (MCDM) systems, exemplified by works from (Amari et al., Al-Wedyan and Shu et al., ) leverage approaches like the Analytic Hierarchy Process (AHP) and linguistic weighting to rank restaurant locations or create customized meal plans, considering factors such as dietary requirements, user preferences, and location-based preferences.
The CRMS restaurant recommendation system integrates collaborative filtering with AHP and ML algorithms, surpassing traditional CF and content-based systems by considering diverse criteria and user preferences.
For the development of our restaurant recommendation system, we utilized the extensive restaurant dataset from ’Restaurant Data with Consumer Ratings’, a widely recognized platform ofering diverse datasets for academic and non-commercial use. This comprehensive dataset encompasses a rich array of information, including user reviews, business details, ratings, and more, providing a robust foundation for our system’s development and evaluation.
The proposed approach is illustrated in Figure 1 and consists of the following steps. CRMS system has two main phases: (1) user profile: where personal data such as ID name and user name are collected and preferences profile are calculated using the AHP method and (2) Filtering: where recommendations are given to the user’s based on their profile and location.
Creating a user profile in the first step in the personalized recommendation process. It Builds a rough description of the user, by taking into account not only their preferences and geographic location, but also the ratings of other similar users.
We obtain a global profile for each user which composed of three profiles: 1. Demographic Profile: It represents the initial user profile section created during registration, comprising the user’s identifier and personal details like name. This data, collected through a registration form, constitutes personal . 2. Preference Profile: It is formed from user input on various restaurant criteria including food, dress code, company, budget, transportation, and area preferences. Users must accurately specify their preferences. 3. Location Profile: It represents the dynamic segment of the profile, crucial for defining the user’s current location, primarily conveyed through latitude and longitude coordinates.
This phase is as a pivotal mechanism for recommending restaurants to the user’s. It involves calculating the similarity between the user’s location and potential restaurants, factoring in distance and proximity. Subsequently, machine learning algorithms, including Singular Value Decomposition (SVD), Nonnegative Matrix Factorization (NMF), and Slope One, leverage the user’s historical ratings to refine recommendations. By analyzing past ratings , these algorithms generate personalized recommendations, enhancing the overall user experience with precision and relevance.
The goal of collaborative filtering is to predict its rating by collecting ratings from other users. Matrix decomposition is a common technique used in collaboration-based filtering, which distinguishes items and users through factor vectors inferred from item classification patterns. The concept behind matrix factorization is straightforward. This involves decomposing the user and restaurant classification matrix into two parts. Example of user ratings and system recommendations for preference-based filtering method. User ratings for a group of restaurants are filtered, based on the top 10 recommendations generated by the system based on the user’s preferences. The recommendations are sorted in descending order based on their expected scores, with the highest scores listed at the top.
CRMS System implemented a collaborative-based recommender system using three matrix factorizationbased algorithms. In particular, we implemented and compared three algorithms: Non-negative Matrix Factorization (NMF), Singular Value Decomposition (SVD), and Slope One. NMF decomposes when implicit feedback is taken into account. Implicit comments include information about other users’ rating history, which provides details about the items for which the user has expressed a preference, either explicitly or implicitly. Algorithm 1 presents the main steps of Restaurant Recommendation System. Algorithm 1 Restaurant Recommendation Algorithm 1: Input: User preferences, restaurant attributes, ratings data 2: Output: Top 10 recommended restaurants for each user 3: Construct pairwise comparison matrix for criteria 4: Calculate priority weights from pairwise comparison matrix 5: Check consistency of pairwise comparisons 6: Calculate overall priorities for each alternative 7: Analyze sensitivity to assess robustness of results 8: Sort list of restaurants and select top 10 9: Predict user ratings using machine learning algorithms (e.g., SVD, NMF, Slope One) 10: Split data into training and testing sets 11: Train selected algorithms (SVD, NMF, Slope One) on training data 12: Aggregate predicted ratings to generate ranked list of recommended restaurants 13: Evaluate trained models using metrics such as MAE, RMSE 14: Return top_list_10 list of restaurants
To handle the computational requirements for modeling large datasets, we used Google Colab to take advantage of shared resources and ensure consistent performance. All algorithms have been implemented in Python using popular libraries such as NumPy and pandas. PyCharm was used as an integrated development environment (IDE) for managing and deploying packages, and additional libraries including arbitrary math and os.path were used. A website is created using the Laravel framework and JavaScript.
To evaluate the performance of our system, we used the AHP method. first, we calculated the similarity scores between characteristics restaurants with users preferences. The results showed that our algorithm gives an average similarity score of 0.60 for the top recommended restaurants. Also, we calculated error comparison and analysis tests using the two measures RMSE (Root Mean Squared Error) and MAE (Mean Absolute Error) for the evaluation of the final results of the prediction. We compared our method with Sun et al.’s Method. Evaluation metrics are: RMSE measures the square root of the average of the squares of the diferences between actual values and predicted values ^:
MAE computes the average absolute diferences between actual values and predicted values ^: Based on the provided information, we can summarize the performance results of diferent recommendation algorithms in terms of RMSE (Root Mean Square Error) and MAE (Mean Absolute Error)in the table 1 : RMSE = ⎷⎸⎸ 1 ∑=︁1 ( − ^)2
MAE = 1 ∑︁ | − ^|
=1
RMSQ-MF CRMS-SVD CRMS-NMF CRMS-SlopOne
According to the results, the CRMS-SVD method exhibits the lowest RMSE, while the CRMS-SlopOne method achieves the lowest MAE. The choice between these algorithms may depend on the specific application requirements and priorities. As shown in the table 1. Compared to other methods, the SVD algorithm gives the best results, with RMSE = 0.5777, MAE = 0.6821. To this end, we create a learning model based on this method that construct and provide RMSE and MAE values for the recommendation system.
After developing and evaluating the system, we deployed it for use by users. This involves integrating the system into a website that users can browse to find the recommendation. website boasts an intuitive interface crafted specifically to recommend closest restaurants, facilitating their dining experiences by ofering tailored restaurant suggestions aligned with their preferences. Our system goes beyond to accurately predict user ratings. In addition, it recommends restaurants closest to users, the proposed system ensure that users receive recommendations precisely attuned to their preferences, enhancing their overall dining satisfaction and well-being.
1. Users log in using their email and password. 2. When logging in, users are required to enter personal data such as ID, Name. 3. The user answers questions to find out his preferences 4. The system calculates the user’s priority preferences for choosing restaurants 5. Predict user ratings 6. Get his location from the GPS device 7. Taking advantage of the data collected, our system creates personalized recommendations for suitable restaurants that match the user’s needs, location and rating . 8. Recommendations take into account all this user profile .
(b) the user input form where they can fill in their preferences.
(c) Recommendation results of the best restaurants
This web application provides users with a complete solution to find suitable restaurant options, with their preferences.
This paper presents a personalized approach to food and restaurant recommendations for individuals based on their health status, especially those with diabetes and obesity. We ofer an innovative recommendation system designed to enhance the user experience by looking at individual profiles, preferences, and applying the AHP method to generate personalized recommendations based on the user’s current location and predictive rating using machine learning algorithms. This approach takes into account the user’s diabetes status and optimal nutrient intake. It provides a user-friendly interface and requires users to enter relevant personal details. The system calculates the nutritional content. Overall, the program aims to support individuals in making healthy food choices when dining out. The system skillfully evaluates and prioritizes restaurant options based on multi-criteria, including user ratings and other relevant factors.
As a future work, we want to develop mechanisms for dynamically updating user profiles based on user feedback, behavior, or changes in preferences over time. This adaptive approach ensures that recommendations stay relevant and reflective of evolving user preferences.
During the preparation of this work, the author(s) used GPT-4 in order to correct grammatical errors, typos, and other writing mistakes. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content. [17] Z. Shu, R. A. Carrasco, M. Sánchez-Montañés, J. P. García-Miguel, A multi-criteria decision support model for restaurant selection based on users’ demand level: The case of dianping. com, Information Processing & Management 61 (2024) 103650.