Face à la surabondance des ressources et de l'information sur le net, l'accès aux ressources pertinentes devient une tâche fastidieuse pour les usagers de la toile. Les systèmes de recommandation personnalisée comptent parmi les principales solutions qui assistent l'utilisateur en filtrant les ressources, pour ne lui proposer que celles susceptibles de l’intéresser. L’approche basée sur l’observation du comportement de l’utilisateur à partir de ses interactions avec le e-services est appelée analyse des usages. Le filtrage collaboratif et le filtrage basé sur le contenu sont les principales techniques de recommandations personnalisées. Le filtrage collaboratif exploite uniquement les données issues de l’analyse des usages alors que le filtrage basé sur le contenu utilise en plus les données décrivant le contenu des ressources. Un système de recommandation hybride combine les deux techniques de recommandation. L'objectif de cette thèse est de proposer une nouvelle technique d'hybridation en étudiant les bénéfices de l'exploitation combinée d'une part, des informations sémantiques des ressources à recommander, avec d'autre part, le filtrage collaboratif. Plusieurs approches ont été proposées pour l'apprentissage d'un nouveau profil utilisateur inférant ses préférences pour l’information sémantique décrivant les ressources. Pour chaque approche proposée, nous traitons le problème du manque de la densité des données et le problème du passage à l’échelle. Nous montrons également, de façon empirique, un gain au niveau de la précision des recommandations par rapport à des approches purement collaboratives ou purement basées sur le contenu / Face to the ongoing rapid expansion of the Internet, user requires help to access to items that may interest her or him. A personalized recommender system filters relevant items from huge catalogue to particular user by observing his or her behavior. The approach based on observing user behavior from his interactions with the website is called usage analysis. Collaborative Filtering and Content-Based filtering are the most widely used techniques in personalized recommender system. Collaborative filtering uses only data from usage analysis to build user profile, while content-based filtering relies in addition on semantic information of items. Hybrid approach is another important technique, which combines collaborative and content-based methods to provide recommendations. The aim of this thesis is to present a new hybridization approach that takes into account the semantic information of items to enhance collaborative recommendations. Several approaches have been proposed for learning a new user profile inferring preferences for semantic information describing items. For each proposed approach, we address the sparsity and the scalability problems. We prove also, empirically, an improvement in recommendations accuracy against collaborative filtering and content-based filtering

中國有句諺語:「民以食為天」，食物乃人類生活所不可缺的要素之一，而人們對於食物則有各自的偏好，而要從琳瑯滿目的食物中依照個人喜好來推薦則成為一門重要的課題。 隨著科技的進步，智慧型手機的出現為人類帶來了許多便利，也逐漸改變了人們的生活方式，群眾可以透過智慧型手機來記錄生活的點滴，記錄的方式正走向數位化，而如何利用這些累積下來的數位資料來做分析與推薦也成為熱門的研究目標。 本論文從味覺方面著手，將LifeLog的飲食記錄與味覺做結合，並透過大眾分類與群眾外包的方式，將味覺資料由智慧型手機使用者處獲得，並建構成包含餐廳、餐點名稱以及其對應味覺之資料庫。 本論文實作了一個程式Foodtaste，包含了記錄餐點味覺資料，查詢個人記錄，以及實作數種推薦的功能。本論文並提出了數個計算方法，透過LifeLog累積下來的味覺資料進行計算，來獲得每位使用者的個人口味偏好和味覺比例，並將這些資料與餐點的味覺比例計算來對餐廳進行個人化的餐點推薦。 / Foods and eating are the basic element of human's life, and people have their own favorite in choosing foods. Thus it is an important issue to make some recommendation for people in front of a dazzling array of foods. With the advances in technology, smartphones bring convenience to people and change their life style. One can use smartphones to record various things in his life. The ways of memories become digitalized, and how to use these digital data to analyze and give opinions becomes popular. Base on one’s taste, present study combined dietary records and food taste in Lifelog, using Folksonomy and Crowdsourcing to acquire data of specific food taste from smartphone users, and linked these data to restaurant’s name and the name of the meal in our database. We designed a smartphone application which called "Foodtaste". It provided users to record what they ate and how did it taste, looking up personal records, and several recommending methods. Our study also provides several methods in calculating cumulative data in Lifelog and acquiring the preference of one’s taste and ratio in variable foods from every user. Then we calculated these data to carry out personalized food recommendation.

Information Filtering and Recommender Systems have been used and has been implemented in various ways from various entities since the dawn of the Internet, and state-of-the-art approaches rely on Machine Learning and Deep Learning in order to create accurate and personalized recommendations for users in a given context. These models require big amounts of data with a variety of features such as time, location and user data in order to find correlations and patterns that other classical models such as matrix factorization and collaborative filtering cannot. This thesis researches, implements and compares a variety of models with the primary focus of Machine Learning and Deep Learning for the task of music recommendation and do so successfully by representing the task of recommendation as a multi-class extreme classification task with 100 000 distinct labels. By comparing fourteen different experiments, all implemented models successfully learn features such as time, location, user features and previous listening history in order to create context-aware personalized music predictions, and solves the cold start problem by using user demographic information, where the best model being capable of capturing the intended label in its top 100 list of recommended items for more than 1/3 of the unseen data in an offine evaluation, when evaluating on randomly selected examples from the unseen following week. / Informationsfiltrering och rekommendationssystem har använts och implementeratspå flera olika sätt från olika enheter sedan gryningen avInternet, och moderna tillvägagångssätt beror påMaskininlärrning samtDjupinlärningför att kunna skapa precisa och personliga rekommendationerför användare i en given kontext. Dessa modeller kräver data i storamängder med en varians av kännetecken såsom tid, plats och användardataför att kunna hitta korrelationer samt mönster som klassiska modellersåsom matris faktorisering samt samverkande filtrering inte kan. Dettaexamensarbete forskar, implementerar och jämför en mängd av modellermed fokus påMaskininlärning samt Djupinlärning för musikrekommendationoch gör det med succé genom att representera rekommendationsproblemetsom ett extremt multi-klass klassifikationsproblem med 100000 unika klasser att välja utav. Genom att jämföra fjorton olika experiment,så lär alla modeller sig kännetäcken såsomtid, plats, användarkänneteckenoch lyssningshistorik för att kunna skapa kontextberoendepersonaliserade musikprediktioner, och löser kallstartsproblemet genomanvändning av användares demografiska kännetäcken, där den bästa modellenklarar av att fånga målklassen i sin rekommendationslista medlängd 100 för mer än 1/3 av det osedda datat under en offline evaluering,när slumpmässigt valda exempel från den osedda kommande veckanevalueras.