USAGE LICENSE Creative Commons Attribution 4.0 International Public License FILE CONTENTS dataset├─ foods folder containing files of food dataset (starting data from giallozafferano.it)│ ├─ CSV folder containing files in .csv format│ │ ├─ categories.csv │ │ ├─ foodDataset.xlsx each of other .csv files derived from sheet of this excel file│ │ ├─ ingredients.csv │ │ ├─ ingredientsClasses.csv │ │ ├─ ingredientsMetaclasses.csv │ │ ├─ preparations.csv │ │ └─ recipes.csv │ └─ TXT folder containing files in .txt format│ ├─ scorpored values folder containing values in textData scorpored by type of data│ │ ├─ category-cost-difficulty.txt │ │ ├─ ingredients.txt │ │ ├─ names.txt │ │ ├─ preparations.txt │ │ └─ preparationTime.txt │ └─ textData.txt .txt version of dataset/foods/CSV/foodDataset.xlsx file├─ survey_answers folder containing results of the surveys on food preferences of the dataset│ ├─ sorts folder containing results of the three surveys' questions where users sort foods│ │ ├─ ID folder containing the sorts identifying foods by their ID│ │ │ ├─ sort1.csv the three csv files contain the survey id, and then the food ordered by the user│ │ │ ├─ sort2.csv │ │ │ └─ sort3.csv │ │ └─ Name folder containing the sorts identifying foods by their names (in italian)│ │ ├─ sort1.csv the three csv files contain the survey id, and then the food ordered by the user│ │ ├─ sort2.csv │ │ └─ sort3.csv │ ├─ answers.csv results of the surveys│ ├─ labels.txt labels of the samples in samples.txt │ └─ samples.txt couples of food in pairwise comparison form [1] extracted from the sorts├─ survey_of_return folder containing the results of survey of returns│ ├─ theories folder containing the theories used to build the survey of returns│ │ ├─ 10 │ │ ├─ ... folders relative to the i-th user theories│ │ └─ 43 │ │ ├─ global-indirectPCA-8PC2STD.las .las file from which the relative theory is returned using ILASP│ │ ├─ global-indirectPCA-8PC2STD.txt .txt file containing the returned theory by ILASP of relative case│ │ ├─ global-indirectPCA-17PC2STD.las │ │ ├─ global-indirectPCA-17PC2STD.txt │ │ ├─ global-noPCA.las │ │ ├─ global-noPCA.las │ │ ├─ local-indirectPCA-8PC2STD.las │ │ └─ local-indirectPCA-8PC2STD.las │ ├─ 10.csv .csv files containing the survey of return results relative to the i-t user│ ├─ ...│ └─ 43.csv├─ readme.md└─ readme.txt * Each sample has the form and has label 1 if IDfood1 is preferred over IDfood2, -1 if IDfood2 is preferred over IDfood1, 0 if there is indifference relationship over IDfood1 and IDfood2 The preference sorts have been translated in two phases: search of inconsistencies among the three sorts in order to get the "certain" preference relationships. search of transitivity of preferences among the "certain" couples, giving indifference relationship to those for which no transitivity has been found. RECIPES DATASET DESCRIPTION the description refer to dataset/foods/CSV/foodDataset.xlsx Feature Name Description Name italian name of the recipe ID ID associated to the recipe Link link of where the food data has been get Category Name name of the category (Starter, Complete Meal, First Course, Second Course, Savoury Cake) Category ID ID associated to the category Cost cost indicator of the food, discrete interval from 1 to 5 Difficulty difficuly indicator of the food, discrete interval from 1 to 4 Preparation Time integer positive number that indicates preparation time of the food expressed in minutes Ingredient english name of an ingredient of the recipe Ingredient ID ID associated to the ingredient Weight weight that the ingredient has in the composition of the interested recipe ... the columns of ingredients repeats for 18 times, leaving empty spaces when the recipe has no ingredients other than those already entered Preparation english name of a preparation performed on the recipe Preparation ID ID associated to the preparation Weight weight that the preparation has in the composition of the interested recipe ... the columns of preparations repeats for 5 times similarly to ingredients in other sheet of the file are reportet all the ingredients, divided in classes and metaclasses, preparations and categories NOTE: in dataset/foods/TXT/textData.txt ingredients and preparation has been vectorized as follow: each element of the ingredient vector represent the weight of the ingredient class in the recipe. The weight of an ingredient class in a recipe is collected by sum up the weight of the ingredients owned by that particular ingredient class in the recipe. each element of the preparation vector represent the weight of the preparation in the recipe. PREFERENCES DATASET DESCRIPTION In the file dataset/survey_answers/answers.csv are reported the 54 user's answers to the surveys, formatted in the following format: Column Name Description UserID ID associated to the user Survey ID ID of the survey which user has answered Answer ID ID of the answer respect to the survey Gender Gender of the user AgeRange Age Range of the user Region Italian Region of the user Food1 Rating of the first recipe proposed to the user ... the following are rating of the other food proposed to the user Food21 rating of the twenty-first food proposed to the user Sort-Name1 first sort of three requested to the user during the survey, in which food are identified by name Sort-Name2 second sort of three requested to the user during the survey, in which food are identified by name Sort-Name3 third sort of three requested to the user during the survey, in which food are identified by name Sort-ID1 first sort of three requested to the user during the survey, in which food are identified by ID Sort-ID2 second sort of three requested to the user during the survey, in which food are identified by ID Sort-ID3 third sort of three requested to the user during the survey, in which food are identified by ID Class-Citrus Rating of the user about the ingredient class "Citrus" ... the following are rating of the other ingredient classes Class-Yogurt Rating of the user about the ingredient class "Yogurt" Metaclass-Meat Rating of the user about the ingredient metaclass "Meat" ... the following are rating of the other ingredient metaclasses Metaclass-Vegetables Rating of the user about the ingredient metaclass "Vegetables" Preparation-Boiling Ratibg of the user about the preparation "Boiling" ... the following are rating of the other preparations Preparation-Stewing Ratibg of the user about the preparation "Stewing" Difficulty rating of the user respect to recipes which preparation difficulties are generally higher PreparationTime rating of the user respect to recipes which preparation time is generally higher Cost rating of the user respect to recipes which cost is generally higher ParticularCase1-Ingredient/Preparation1 first ingredient/preparation of a particular combination of ingredients/preparations for which previous expressed preferences are no longer valid [2] ParticularCase2-Ingredient/Preparation1 second ingredient/preparation of particular combination of ingredients/preparations for which previous expressed preferences are no longer valid [2] ParticularCase3-Ingredient/Preparation1 third ingredient/preparation of particular combination of ingredients/preparations for which previous expressed preferences are no longer valid [2] ParticularCase1-rating rating of the particular combination of ingredients/preparations expressed with previously three columns ... following are other particular combination of ingredients/preparation Note that the Sort-Name and Sort-ID columns are also reported in form of .csv files, respectevely in the folders dataset/survey_answers/sorts/Name and dataset/survey_answers/sorts/ID. In the file dataset/survey_answer/samples.txt are reported 54 user's orderings in the form of pairwise comparison. Thus for each row correspond the ordering of a user. The ordering is written in the form of pairwise comparison, so each element of the ordering are paired with all others (avoiding simmetries). For instance, given the recipes as their ID: 1; 2; 3 becomes: 1,2; 1,3; 2,3 The file is written following the .csv format. In the file dataset/survey_answer/lables.txt are written the corresponding labels of the couples. The label has value 1 if the first element of the couple is preferred over the second, -1 if the second element is preferred over the first, 0 if there is uncertainity about which of the element is preferred. SURVEY OF RETURN In the folder survey of return are reported the .csv files containing the result of survey of return of each user and the theories used to create the respective surveys. The used theories are those obtained after training ILASP both as global and local approximator. In global approximator we considered both the cases in which PCA is involved and not involved (considering only the indirect case with PC=8, 17 when involved). In local approximator we considered only the case in which PCA is involved indirectly with PC = 8 (considering std for gaussian noise equal to 0.1). More detail are reported in the "HOW TO CITE" correlated paper. In the folder are also returned the .las files containing the file used to train ILASP (version 4.2.0). NOTES [1] Each sample has the form  and has label 1 if IDfood1 is preferred over IDfood2, -1 if IDfood2 is preferred over IDfood1, 0 if there is indifference relationship over IDfood1 and IDfood2. The preference sorts have been translated into pairwise comparison in two phases: search of inconsistencies among the three sorts in order to get the "certain" preference relationships search of transitivity of preferences among the "certain" couples, giving indifference relationship to those for which no transitivity has been found. More details about these two phases and a pseudo-code can be found in the article attached with this dataset [2] During the survey we proposed to the user to indicate metaclasses if the particular combination applies for all classes owned by that specific metaclass, or to indicate classes if the particular combination apply only for that specific class. To Identify classes from metaclasses, since their name could be misleading, we denoted classes with the prefix "---". We left this notation also in the final dataset, since could be useful to easily recognize classes from metaclasses withouth looking to the files regarding them. HOW TO CITE D.Fossemò, F.Mignosi, L.Raggioli, M.Spezialetti, F.A.D'Asaro. Using Inductive Logic Programming to globally approximate Neural Networks for preference learning: challenges and preliminary results. Proceedings of BEWARE-22, co-located with AIxIA 2022, November 28-December 2, 2022, University of Udine, Udine, Italy. CEUR Workshop proceedings. Vol. 3319. 67:83. 2023.