Content-Based Filtering

In the view of massive content explosion in World Wide Web through diverse sources, it has become mandatory to have content filtering tools. The filtering of contents of the web pages holds greater significance in cases of access by minor-age people. The traditional web page blocking systems goes by the Boolean methodology of either displaying the full page or blocking it completely. With the increased dynamism in the web pages, it has become a common phenomenon that different portions of the web page holds different types of content at different time instances. This paper proposes a model to block the contents at a fine-grained level i.e. instead of completely blocking the page it would be efficient to block only those segments which holds the contents to be blocked. The advantages of this method over the traditional methods are fine-graining level of blocking and automatic identification of portions of the page to be blocked. The experiments conducted on the proposed model indicate 88% of accuracy in filtering out the segments.

In this era of the internet and with the easy availability of data at a very low cost, searching for information is growing at an exponential rate. So, it is now impossible to find the desired information without proper guidance. Here is the need of the recommendation system. The system will recommend which information relevant to the user according to their searching pattern. It will also explore hidden results with a minimal effect. The goal of this paper is to describe the famous recommendation systems which are used mostly and to explore what is the need of these kinds of systems and also what kind of technology has been used to provide better services to its users. Finally we would like to show how one recommendation system is different from another as per user need.

Reproducibility is an important requirement for scientific progress, and the lack of reproducibility for a large amount of published research can hinder the progress over the state-of-the-art. This concerns several research areas, and recommender systems are witnessing the same reproducibility crisis. Even solid works published at prestigious venues might not be reproducible for several reasons: data might not be public, source code for recommendation algorithms might not be available or well documented, and evaluation metrics might be computed using parameters not explicitly provided. In addition, recommendation pipelines are becoming increasingly complex due to the use of deep neural architectures or representations for multimodal side information involving text, images, audio, or video. This makes the reproducibility of experiments even more challenging. In this work, we describe an extension of an already existing open-source recommendation framework, called ClayRS, with the aim of providing the foundation for future reproducibility of recommendation processes involving images as side information. This extension, called ClayRS Can See, is the starting point for reproducing state-of-the-art recommendation algorithms exploiting images. We have provided our implementation of one of these algorithms, namely VBPR -Visual Bayesian Personalized Ranking from Implicit Feedback, and we have discussed all the issues related to the reproducibility of the study to deeply understand the main traps and pitfalls, along with solutions to deal with such complex environments. We conclude the work by proposing a checklist for recommender systems reproducibility as a guide for the research community.

In this paper, we present the results of an empirical evaluation investigating how recommendation algorithms are affected by popularity bias. Popularity bias makes more popular items to be recommended more frequently than less popular ones, thus it is one of the most relevant issues that limits the fairness of recommender systems. In particular, we define an experimental protocol based on two state-of-theart datasets containing users' preferences on movies and books and three different recommendation paradigms, i.e., collaborative filtering, content-based filtering and graph-based algorithms. In order to evaluate the overall fairness of the recommendations we use well-known metrics such as Catalogue Coverage, Gini Index and Group Average Popularity (ΔGAP). The goal of this paper is: (i) to provide a clear picture of how recommendation techniques are affected by popularity bias; (ii) to trigger further research in the area aimed to introduce methods to mitigate or reduce biases i...

Recommender Systems suggest items that are likely to be the most interesting for users, based on the feedback, i.e. ratings, they provided on items already experienced in the past. Time-aware Recommender Systems (TARS) focus on temporal context of ratings in order to track the evolution of user preferences and to adapt suggestions accordingly. In fact, some people’s interests tend to persist for a long time, while others change more quickly, because they might be related to volatile information needs. In this paper, we focus on the problem of building an effective profile for short-term preferences. A simple approach is to learn the short-term model from the most recent ratings, discarding older data. It is based on the assumption that the more recent the data is, the more it contributes to find items the user will shortly be interested in. We propose an improvement of this classical model, which tracks the evolution of user interests by exploiting the content of the items, besides ...

The growth of the Web is the most influential factor that contributes to the increasing importance of text retrieval and filtering systems. On one hand, the Web is becoming more and more multilingual, and on the other hand users themselves are becoming increasingly polyglot. In this context, platforms for intelligent information access as search engines or recommender systems need to evolve to deal with this increasing amount of multilingual information. This paper proposes a content-based recommender system able to generate cross-lingual recommendations. The idea is to exploit user preferences learned in a given language, to suggest item in another language. The main intuition behind the work is that, differently from keywords which are inherently language dependent, concepts are stable across different languages, allowing to deal with multilingual and cross-lingual scenarios. We propose four knowledge-based strategies to build concept-based representation of items, by relying on the knowledge contained in two knowledge sources, i.e. Wikipedia and BabelNet. We learn user profiles by leveraging the different concept-based representations, in order to define a cross-lingual recommendation process. The empirical evaluation carried out on two state of the art datasets, DBbook and Movielens, shows that concept-based approaches are suitable to provide cross-lingual recommendations, even though there is not a clear advantage of using one of the different proposed representations. However, it emerges that most of the times the approaches based on BabelNet outperform those based on Wikipedia, which clearly shows the advantage of using a native multilingual knowledge source.

This paper is to present an overview of Collaborative Filtering (CF) recommender system and show the major CF challenges. In general, the recommendation systems are the best way to help users to overcome the information overload issue. The CF approach is one of the most widely used and most successful methods in the recommendation system, such as e-commerce. This paper introduced a brief description about recommender's approaches which are: content-Based, collaborative filtering and hybrid approach. Next, defined the main challenges which have clearly impact on the performance and accuracy of CF recommender system. The major finding of this paper is the CF main problems: Data sparsity, Cold-star, and Scalability. By presenting of these challenges the quality of recommendations can be improved by proposing new methods. The paper ends with conclusion summarizes the limitations of the existing methods and recommendations.

Various entities (e.g., parents, employers) that provide users (e.g., children, employees) access to web content wish to limit the content accessed through those computers. Available filtering methods are crude in that they too often block bacceptableQ content while failing to block bunacceptableQ content. This paper presents a general and flexible classification method based on statistical techniques applied to text material, that we call, Filtering by Statistical Classification (FSC). According to each individual entity's expressed opinions about what content in a training data set is or is not acceptable, FSC constructs a customized model to represent each individual entity's preferences. FSC then uses this customized model to examine new web content and to block unwanted content. The empirical results suggest that our method has greater predictive power than do a variety of existing approaches.

In 2015, global real estate was worth $217 trillion, which is approximately 2.7 times the global GDP; it also accounts for roughly 60% of all conventional global resources, making it one of the key factors behind any country’s economic growth and stability. The accessibility of spatial big data will help real estate investors make better judgement calls and earn additional profit. Since location is deemed necessary for real estate and consequent decision-making, digital maps have become a prime resource for real estate purchases, planning and development. Personalisation can assist in making judgments by identifying user desires and inclinations, which can then be recorded or captured as a user performs some interactions with a digital map. A personalised real estate portal can use this information to suggest properties, assist homeowners and provide valuable real estate analytics. This article presents a novel framework for recommending real estate to users. By monitoring user inte...

The quick increment in the online data content has made it extremely troublesome for individuals to discover data that is pertinent to their requirements and interests. Proposal framework is an intense apparatus that gives a potential answer for this test by offering a mechanized component to search out applicable and new data. Individuals depend on client surveys on the web when they require data about item, motion picture, video, news, eatery, and so on. In any case, fundamental issue is giving suggestion to client shape huge information environment. Huge information is a rising innovation, as the name recommends is about taking care of huge measure of information, technique and preparing the information inside middle of the road passed time. An enormous information is gathered and handled to settle on some choice furthermore used to portray any sort of information which might be organized, semi-organized, unstructured and if the information develops. In this work our primary commitment is to prescribe online recordings to client in view of client intrigue. In past framework, recordings were positioned in view of the viewer of that video yet in our proposed framework recordings are positioned and prescribed to client in view of remarks, check and likes. Here client can give remarks just in the event that he/she has watched that specific video, generally clients are not qualified to give remarks on that video. By the assistance of this element we rank recordings in remarks and likes based. The upgrade of our proposition is the opinion grouping of client remarks, with two conceivable marks: negative (neg) and positive (pos). The given technique is utilized to enhance the versatility and effectiveness, if information develops.

Cet article résume le déroulement du projet proposé par la société KeeSeeK dans le cadre de la Semaine d'Étude Maths-Entreprises (SEME) de Strasbourg organisée en novembre 2018. Ce projet porte sur la recommandation d'offres d'emploi dans un moteur de recherche en ligne. Nous nous plaçons dans le cas où le moteur de recherche doit proposer des offres d'emploi susceptibles de maximiser les clics des candidats. Ces offres ont été préalablement sélectionnées et l'utilisateur (possédant le moteur de recherche) souhaite améliorer sa sélection. Pour s'intéresser à ce problème, nous avons proposé une version revisitée d'un algorithme d'apprentissage par renforcement (Thompson-sampling) lorsque le nombre d'offres d'emploi disponibles est important (∼ 100). Mots-clés. Allocation dynamique, bandit manchot multi bras, système de recommandations, partial pooling et données réelles.

The constant growth of online real estate information has emphasized the need for the creation and improvement of intelligent recommendation systems to help mitigate the difficulties associated with user decision making. This systematic review, following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines and criteria, investigates current approaches and models used in real estate recommendation systems, with a focus on papers published in 2019 and 2024. The review identifies four main techniques: content-based filtering, collaborative filtering, knowledge-based systems, and hybrid approaches. Key findings indicate a preference for deep learning models, specifically convolutional neural network and long-short term memory (CNN-LSTM) architectures, and highlight the most used property characteristics: price, number of rooms, size, and location. The research addresses several important challenges, including the cold start problem, data sparsity, and the importance of adaptive learning in dynamic markets. Potential future research fields are outlined, with a focus on hybrid model architectures, attention mechanisms, and explainable artificial intelligence (AI). This review provides a comprehensive overview of the field, enabling scholars and practitioners to improve the accuracy and user experience of real estate recommendation systems.

Les systèmes de recommandation (SR) sont largement utilisés de nos jours dans de nombreux domaines pour générer les items d'intérêt. Récemment, ils ont été appliqués dans le domaine de l'apprentissage amélioré par la technologie (Technology Enhanced Learning TEL) afin de permettre la recommandation des ressources d'apprentissage pertinentes pour aider les enseignants et les apprenants. Dans cet article, nous proposons une nouvelle technique de recommandation qui combine un algorithme de filtrage collaboratif flou avec un algorithme basé contenu pour améliorer la recommandation, en utilisant les préférences des apprenants et l'importance de la connaissance, pour recommander des items avec différents contextes, afin de réduire l'effet du problème célèbre de la stabilité des SR par rapport à la dynamicité du profil de l'apprenant. Les évaluations empiriques montrent que la technique proposée est réalisable et efficace. .

The use of Recommendation Systems in virtually all online services nowadays makes people interact with them more and more, especially when considering the domain of intelligent cities. However, these systems have accumulated criticism over time because of their overemphasis on similarity, which ultimately produces recommendations that are often obvious and redundant and puts users in a "filter bubble", limiting their experiences. As a solution, the diversification of recommendation arises, which implies a problem of how to balance the accuracy and diversity since the increase of diversity decreases the accuracy and vice versa. Existing works diversify using post-filtering or modifications in the recommendation algorithm, being the first approach the most common. Therefore, this article proposes an outline hybrid approach that seeks to maximize diversity with minimum losses of accuracy applied in the domain of intelligent cities, in order to mitigate the "filter bubble" effect. Resumo. O emprego de Sistemas de Recomendação em praticamente todos os serviços on-line, na atualidade, faz com que as pessoas interajam com eles cada vez mais, principalmente quando se considera o domínio de cidades inteligentes. Entretanto, esses sistemas têm acumulado críticas ao longo do tempo devido à sua ênfase demasiada em similaridade, que acaba produzindo recomendações que são muitas vezes óbvias e redundantes e coloca os usuários em uma "bolha de filtro", limitando suas experiências. Como solução, surge a diversificação de recomendação, que implica em um problema de como balancear a acurácia e diversidade já que o aumento da diversidade diminui a acurácia e vice-versa. Trabalhos existentes buscam diversificar utilizando pós-filtragem ou com base no próprio algoritmo de recomendação sendo, a primeira abordagem, a mais comum. Portanto, neste artigo é proposto, a partir de um vasto estudo bibliográfico, um esboço de abordagem híbrida que busca maximizar a diversidade com perdas mínimas de acurácia, aplicada no domínio de cidades inteligentes, visando atenuar o efeito de "bolha de filtro". ___________ Artigo publicado em 22/05/2019 Cadernos de Informática vol. 10, n. 1 -Modelos e Sistemas para Cidades Inteligentes experiência do usuário e possibilitando-o interagir com diferentes opiniões, pontos de vista, dentre outros. Como trabalhos futuros pode-se citar: a realização de uma ampla e contundente revisão sistemática, a partir deste estudo inicial, que permitirá a escolha dos algoritmos a serem implementados. Também, a realização dos experimentos com e sem usuários, conforme relatado na seção de proposta. O presente trabalho foi realizado com apoio da Coordenação de Aperfeiçoamento de Pessoal de Nível Superior -Brasil (CAPES) -Código de Financiamento 001, e pelo Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq).

Recommender Systems took an importante place in Internet to item recommendation for users. However, many Point-of-Interest(POI) Recommender Systems work with only one contexto for each user but user can be in many contexts at the same time. This work proposes an adaptive model for creation of multiples contexts for same user and thus, Recommender System can make better recommendations.

Prior knowledge of the social aspects of prospective destinations can be very influential in making travel destination decisions, especially in instances where social concerns do exist about specific destinations. In this paper, we describe the implementation of an ontology-enabled Hybrid Destination Recommender System (HDRS) that leverages an ontological description of five specific social attributes of major Nigerian cities, and hybrid architecture of content-based and case-based filtering techniques to generate personalised top-n destination recommendations. An empirical usability test was conducted on the system, which revealed that the dependability of recommendations from Destination Recommender Systems (DRS) could be improved if the semantic representation of social attributes information of destinations is made a factor in the destination recommendation process.

There are changes taking place in Indonesia's educational system, particularly in universities. The learning approach used in the transformation program is student-centered. A firm foundation in literacy is required for the application of this learning. A library is one of the amenities that students require. The library is an ideal resource for pupils to enhance their critical thinking skills. It's not always simple to find books in the library, though. Students could find it challenging to locate the books they seek because there are so many collections already in existence. One method for doing book searches is through the use of a recommendation system. Using content based filtering is one recommendation system algorithm. This study suggests a content based filtering algorithm-based book recommendation system to facilitate students' search for informatics book titles. TF-IDF and Cosine Similarity are used in a similarity search to find phrases and assign weights to them. The content-based filtering algorithm's research findings might suggest books depending on user-specified parameters. 92,5% accuracy is the average for this method.

In this research, we propose an intelligent job recommendation system that leverages BERT (Bidirectional Encoder Representations from Transformers) and WordNet for semantic job matching. Traditional job recommendation methods often rely on keyword-based matching, which fails to capture the true meaning of job descriptions and candidate profiles. Our system enhances job recommendation accuracy by utilizing deep learning and natural language processing (NLP) techniques to analyze the contextual relationships between job postings and applicant skills. By incorporating BERT for text embeddings and WordNet for semantic similarity, the system improves candidate-job matching, providing more relevant and personalized job suggestions. Experimental results show that our approach significantly outperforms conventional keyword-based methods in accuracy and efficiency.
Keywords: Job Recommendation System, BERT, WordNet, Deep Learning, NLP, Semantic Matching, AI-driven Recruitment.

Prior knowledge of the social aspects of prospective destinations can be very influential in making travel destination decisions, especially in instances where social concerns do exist about specific destinations. In this paper, we describe the implementation of an ontology-enabled Hybrid Destination Recommender System (HDRS) that leverages an ontological description of five specific social attributes of major Nigerian cities, and hybrid architecture of content-based and case-based filtering techniques to generate personalised top-n destination recommendations. An empirical usability test was conducted on the system, which revealed that the dependability of recommendations from Destination Recommender Systems (DRS) could be improved if the semantic representation of social attributes information of destinations is made a factor in the destination recommendation process.

We propose and evaluate the performance of a number of methods for automatic recording of TV programs for digital video servers, which estimate the user's preference over TV programs based on her/his past viewing behavior and automatically record a selected number of TV programs believed to be of interest to the user. Our methods combine the so-called content-based filtering and social (or collaborative) filtering methods and are based on a certain class of on-line learning algorithms known as the 'specialist' algorithms, recently developed in the field of computational learning theory. We empirically evaluated the performance of content-based part of the proposed methods using preference data on TV programs consisting of scores given by people on actual TV programs. The results are largely encouraging and indicate in particular that our methods are practical in terms of both the precision in predicting the user's preference and computational complexity.

This study investigates the difficulty of improving product recommendations in e-commerce systems by tackling the common problem of poor diversity in suggestions. We present a novel approach that uses a Siamese network architecture and ResNet for feature extraction to recommend visually similar elements while incorporating diversity through a cluster-based mechanism. The Siamese network is used to compare product pairs, allowing it to recommend both comparable and dissimilar items from distinct clusters. The model was evaluated using a variety of evaluation metrics, resulting in an accuracy of 88.5%, a precision of 90.2%, a recall of 87.1%, and an F1 score of 88.6%. Our results demonstrate that our strategy maintains a high level of relevance in suggestions while efficiently incorporating variety, hence improving the overall user experience in e-commerce applications.

On-line monitoring is essential for observing and improving the reliability and performance of large-scale distributed (LSD) systems. In an LSD environment, large numbers of events are generated by system components during their execution and interaction with external objects (e.g. users or processes). These events must be monitored to accurately determine the run-time behavior of an LSD system and to obtain status information that is required for debugging and steering applications. However, the manner in which events are generated in an LSD system is complex and represents a number of challenges for an on-line monitoring system. Correlated events axe generated concurrently and can occur at multiple locations distributed throughout the environment. This makes monitoring an intricate task and complicates the management decision process. Furthermore, the large number of entities and the geographical distribution inherent with LSD systems increases the difficulty of addressing traditi...

This paper presents a distributed client-server architecture for the personalized delivery of textual news content to mobile users. The user profile consists of two separate models, that is, the long-term interests are stored in a skeleton profile on the server and the short-term interests in a detailed profile in the handset. The user profile enables a high-level filtering of available news content on the server, followed by matching of detailed user preferences in the handset. The highest rated items are recommended to the user, by employing an efficient ranking process. The paper focuses on a two-level learning process, which is employed on the client side in order to automatically update both user profile models. It involves the use of machine learning algorithms applied to the implicit and explicit user feedback. The system's learning performance has been systematically evaluated based on data collected from regular system users.

Email is an increasingly important and ubiquitous means of communication, both facilitating contact between individuals and enabling rises in the productivity of organizations. However, the relentless rise of automatic unauthorized emails, a.k.a. spam is eroding away much of the attractiveness of email communication. Most of the attention dedicated to spam detection has focused on the content of the emails or on the addresses or domains associated with spam senders. Although, methods based on these -easily changeable -identifiers work reasonably well, they miss on the fundamental nature of spam as an opportunistic relationship, very different from the normal mutual relations between senders and recipients of legitimate email. Here we present a comprehensive graph theoretical analysis of email traffic that captures these properties quantitatively. We identify several simple metrics that serve both to distinguish between spam and legitimate email and to provide a statistical basis for models of spam traffic.

The growing complexity and variety of patient information require the creation of intelligent systems to assist halthcare professionals in making informed choices. This research presents an AI-powered online platform created to detect and prioritize comparable patient cases based on demographic and clinical factors like age, gender, blood type, health conditions, and treatments. Utilizing Natural Language Processing (NLP) methods such as TF-IDF vectorization and cosine similarity, along with statistical feature engineering, the system provides precise and meaningful insights.

Information filtering (IF) systems usually filter data items by correlating a vector of terms (keywords) that represent the user profile with similar vectors of terms that represent the data items (eg documents). The terms that represent the data items can be determined by (human) experts (eg authors of documents) or by automatic indexing methods. In this study we employ an artificial neural-network (ANN) as an alternative method for both filtering and term selection, and compare its effectiveness to “traditional” methods. In an earlier ...

The best knowledge management systems are recommender systems, which let consumers filter out irrelevant data and provide tailored recommendations based on their past historical data and related products they are looking for online. A recommender system provides suggestions to customers in various circumstances. Online book sellers today engage in a number of competitive activities. One of the more powerful techniques for increasing earnings and keeping customers is the recommendation system. Books that will attract buyers must be recommended by the book recommendation system. This study proposes a system for recommending books that combines association rule mining, collaborative filtering and content filtering.

We present an intelligent agent designed to compile a daily news program for individual users. Based on feedback from the user, the system automatically adapts to the user’s preferences and interests. In this paper we focus on the system's user modeling component. First, we motivate the use of a multi-strategy machine learning approach that allows for the induction of user models that consist of separate models for long-term and short-term interests. Second, we investigate the utility of explicitly modeling information that the system has already presented to the user. This allows us to address an important issue that has thus far received virtually no attention in the Information Retrieval community: the fact that a user's information need changes as a direct result of interaction with information. We evaluate the proposed algorithms on user data collected with a prototype of our system, and assess the individual performance contributions of both

Scientific events bring together a large number of researchers and are composed of different types of sessions, which can cause an overload of attention and difficulty in deciding which sessions to participate. To lessen such problems, Recommender Systems can assist the user by offering options that are appropriate for each attendee. This paper presents a proposal for recommending sessions of academic/scientific events based on social elements. The recommendations are supported by the academic events' co-authoring network to improve the quality of session recommendation based on the users' previous publications. For authors/participants who do not have publications in previous editions of the corresponding event, the recommendations will be generated through the Collaborative Filtering approach.

Information retrieval (IR) and recommender systems (RS) have been employed for addressing search tasks executed during literature review and the overall scholarly communication lifecycle. Majority of the studies have concentrated on algorithm design for improving the accuracy and usefulness of these systems. Contextual elements related to the scholarly tasks have been largely ignored. In this paper, we propose a framework called the Scientific Paper Recommender and Retrieval Framework (SPRRF) that combines aspects of user role modeling and user-interface features with IR/RS components. The framework is based on eight emergent themes identified from participants feedback in a user evaluation study conducted with a prototype assistive system. 119 researchers participated in the study for evaluating the prototype system that provides recommendations for two literature review and one manuscript writing tasks. This holistic framework is meant to guide future studies in this area.

Esse artigo a apresenta o resultado de uma revisão bibliográfica sobre explicação de recomendação com diversificação. Constatou-se com base nela que nenhuma pesquisa propôs ainda estudar como gerar explicação de recomendação com diversificação. Foram encontrados apenas trabalhos que indicam a necessidade de haver explicação de recomendação com diversificação. A partir dessa constatação de necessidade de pesquisa propõese, como trabalho futuro, investigar e desenvolver uma abordagem de explicação de recomendação com diversificação. Essa abordagem terá que gerar explicações que sejam interpretáveis e persuasivas de modelos

This research investigates the performances of the Markov Blanket (MB) and Tree Augmented Naïve-Bayes Network (TAN) of the Bayesian Network structure of the IRIS dataset. For evaluation purposes, the performances of the TAN, and MB classifiers were measured using statistical indices. Experimental results strongly suggested that the TAN is better than MB on training dataset and vise vasa in the test dataset. In the other hand, time computational complexity of both the classifiers was found to be equal. The result obtained in this research is of significance to researchers intending to use Bayesian Network to create a classifier for enhancing the performance of biometrics systems.

Huge amount of information available on internet makes it difficult for the user to get the exact search results according to his preferences. In this paper, we attempt to solve this problem to certain extent by extending the NUTCH open source search engine using personalized information of user. The user's information will be extracted from the social networking sites like Facebook. The search keywords given by user will be input to the NUTCH search engine. The results returned by NUTCH search engine will be further refined using our own Profile Biasing Algorithm.

Recommender systems are in the center of network science, and they are becoming increasingly important in individual businesses for providing efficient, personalized services and products to users. Previous research in the field of recommendation systems focused on improving the precision of the system through designing more accurate recommendation lists. Recently, the community has been paying attention to diversity and novelty of recommendation lists as key characteristics of modern recommender systems. In many cases, novelty and precision do not go hand in hand, and the accuracy--novelty dilemma is one of the challenging problems in recommender systems, which needs efforts in making a trade-off between them. In this work, we propose an algorithm for providing novel and accurate recommendation to users. We consider the standard definition of accuracy and an effective self-information--based measure to assess novelty of the recommendation list. The proposed algorithm is based on it...

Recommender systems have been accompanied by many applications in both academia and industry. Among different algorithms used to construct a recommender system, collaborative filtering methods have attracted much attention and been used in many commercial applications. Incorporating the time into the recommendation algorithm can greatly enhance its performance. In this paper, we propose a novel time-aware model-based recommendation system. We show that future ratings of a user can be inferred from his/her rating history. We assume that there is cascade of information between the items such that rating an item can lead to other items being rated. There is indeed a hidden network structure among the items and each user tracks a sequence of items in this network. The dependencies between the items are modeled based on statistical diffusion models and the parameters are obtained through maximum-likelihood estimation. We show that under some mild assumptions, the estimation task becomes a convex optimization problem. A major advantage of the proposed method over classical recommender systems is the ability to include novel items in its recommendation lists besides providing accurate recommendations. The proposed model also results in personalized and diverse recommendations. Experimental evaluations show that the model can be trained based on the ratings of a limited number of users. Furthermore, the proposed model outperforms classical recommendation algorithms in terms of both accuracy and novelty.

Cet article formalise le Filtrage Collaboratif comme un problème de consensus d'ordonnancements. Lorsque les seules informations disponibles sur les utilisateurs sont la liste des produits qu'ils ont achetés ou l'historique des liens qu'ils ont suivis, ce formalisme permet de tenir compte de l'ordre des achats (ou des clics). Un modèle fondé sur un processus de Dirichlet ainsi que sur le modèle de Plackett-Luce est introduit pour gérer le problème de consensus d'ordonnancements correspondant. Enfin, l'article propose un algorithme de chaîne de Markov Monte Carlo à deux niveaux afin d'inférer les paramètres du modèle de consensus d'ordonnancements. Des résultats expérimentaux montrent que l'algorithme proposé est capable de retrouver le mode de chaque exemple dans un délai raisonnable, et ce même pour les configurations comportant jusqu'à 1.000.000 de produits et des historiques d'utilisateurs de taille 1.000. À titre de comparaison, les approches en consensus d'ordonnancement fondées sur des modèles statistiques se limitent de façon standard à des historiques d'utilisateurs de quelques dizaines de produits. Mots-clés : Apprendre à Ordonner, Systèmes de Recommandation, Apprentissage Non Supervisé, Méthodes Statistiques 1. (Meilȃ & Chen, 2010) considère un modèle avec plus de paramètres : λ n'est plus une constante, mais dépend de k.

This study proposed a Collaborative Filtering (CF) algorithm that generates and recommends a list of alternate keywords based on the keywords originally set by a user when searching documents from an online search engine. The proposed CF algorithm incorporates the similarities among keywords associated with the documents and provides crucial recommendations that may not be considered by the user. The purpose of the proposed CF algorithm is to reduce the efforts in identifying appropriate keywords set to allocate the desired documents more efficiently. The traditional search algorithms strive to provide closely-matched results to the keywords specified by the user. However, the most exceptional search engine would not provide good quality results if the original keywords selected by the user were not suitable. Therefore, the proposed system aims at suggesting a set of alternative keywords generated based on a user’s original keywords to help a user in his/her subsequent search activi...

This study proposed a Collaborative Filtering (CF) algorithm that generates and recommends a list of alternate keywords based on the keywords originally set by a user when searching documents from an online search engine. The proposed CF algorithm incorporates the similarities among keywords associated with the documents and provides crucial recommendations that may not be considered by the user. The purpose of the proposed CF algorithm is to reduce the efforts in identifying appropriate keywords set to allocate the desired documents more efficiently. The traditional search algorithms strive to provide closely-matched results to the keywords specified by the user. However, the most exceptional search engine would not provide good quality results if the original keywords selected by the user were not suitable. Therefore, the proposed system aims at suggesting a set of alternative k e y w o r d s g e n e r a t e d b a s e d o n a u s e r ' s o r i g i n a l k e y w o r d s t o help a user in his/her subsequent search activities. The proposed system is expected to improve search accuracy and effectiveness for a novice user in the field of interest. For experience users, with the fast-growing availability of information online, who may not be aware of most the updated/critical keywords, the proposed system is also expected to improve search efficiency. Furthermore, the proposed system is flexible and can easily be integrated with other search algorithms to improve search results.

Titre : Système de recommandation sémantique enrichi. Application au domaine du e-marketing. Mots clés : système de recommandation, interprétation d'information, extraction de connaissances, outil d'annotation sémantique, apprentissage automatique Résumé : Cette thèse est ancrée dans les champs des sciences de l'information et de la communication, de l'ingénierie des connaissances et appliquée au domaine du e-marketing et plus précisément aux enquêtes mystères. Une enquête mystère est une visite en point de vente effectuée par un enquêteur. La visite en point de vente a pour but de mesurer la qualité de l'accueil, la qualité du conseil et le respect des consignes de vente. Les rapports produits par les enquêtes mystères et les études de satisfaction sont des notes, des classements et des verbatims issus de questions ouvertes et fermées. Les commanditaires des enquêtes utilisent ces rapports pour définir des plans d'actions pour améliorer la qualité et l'efficacité du service, allant des changements organisationnels jusqu'à la mise en place de formations spécifiques pour augmenter les compétences du personnel. Ainsi, l'enjeu de cette thèse est d'exploiter les technologies du web sémantique et de l'apprentissage automatique pour mettre en place un système de recommandation capable d'analyser et de partiellement interpréter les données collectées. Pour ce faire, nous nous intéressons à l'élaboration et l'expérimentation de nouvelles approches d'annotation sémantique de données. La mise en place de ce système d'interprétation nécessitera une base de connaissances qui sera alimentée par ces annotations (métadonnées). Nous utilisons aussi les technologies de l'apprentissage automatique pour enrichir le traitement de la recommandation, pour améliorer la pertinence de la prédiction et pour classifier automatiquement les items à recommander pour mieux guider l'expert dans sa prise décision finale sur les axes à améliorer au sein des points de vente. Title : Enriched semantic recommendation system. Application to the e-marketing domain.

Recommenders personalize the web content by typically using collaborative filtering to relate users (or items) based on explicit feedback, e.g., ratings. The difficulty of collecting this feedback has recently motivated to consider implicit feedback (e.g., item consumption along with the corresponding time). In this paper, we introduce the notion of consumed item pack (CIP) which enables to link users (or items) based on their implicit analogous consumption behavior. Our proposal is generic, and we show that it captures three novel implicit recommenders: a user-based (CIP-U), an item-based (CIP-I), and a word embedding-based (DEEPCIP), as well as a state-ofart technique using implicit feedback (FISM). We show that our recommenders handle incremental updates incorporating freshly consumed items. We demonstrate that all three recommenders provide a recommendation quality that is competitive with stateof-the-art ones, including one incorporating both explicit and implicit feedback.

Recommenders personalize the web content by typically using collaborative filtering to relate users (or items) based on explicit feedback, e.g., ratings. The difficulty of collecting this feedback has recently motivated to consider implicit feedback (e.g., item consumption along with the corresponding time). In this paper, we introduce the notion of consumed item pack (CIP) which enables to link users (or items) based on their implicit analogous consumption behavior. Our proposal is generic, and we show that it captures three novel implicit recommenders: a user-based (CIP-U), an item-based (CIP-I), and a word embedding-based (DEEPCIP), as well as a state-of-the-art technique using implicit feedback (FISM). We show that our recommenders handle incremental updates incorporating freshly consumed items. We demonstrate that all three recommenders provide a recommendation quality that is competitive with state-of-the-art ones, including one incorporating both explicit and implicit feedback.

Never before have so many information sources been available. Most are accessible on-line and some exist on the Internet alone. However, this large information quantity makes interesting articles hard to find. Modern Personal Digital Assistants (PDAs), mobile phones, and the advent of ubiquitous computing will further complicate matters. Away from the desktop, the time to select important articles might be even harder to find. Strategies to select relevant information are sorely needed. One such strategy is content-based filtering, coupled with User Profiles. Our prototype uses a Bayesian classifier to select articles of interest to a specific user, according to his profile. The articles are extracted from web pages and displayed in a zoomable interface-based browser on a PDA. Interests may change over time, making it important to keep the profile up to date. The system monitors the users' reading behaviors, from which it infers their interest in particular articles and updates the profile accordingly. Results show that, from the start, most articles are correctly classified. An initial profile opposite to the user's actual interests can be reversed in less than ten days, showing the robustness of our approach. A user's interest in an article is inferred with a high degree of accuracy (over 90%).

Les systemes de recommandation sont des outils servant a suggerer aux utilisateurs des items pouvant les interesser. De tels systemes requierent la definition d'un algorithme prenant en compte le domaine d'application. Cet algorithme est ensuite execute pour chaque utilisateur du systeme afin de lui generer des recommandations, et ce, sans prendre en compte ses particularites et ses besoins specifiques.L'objet de cette these consiste a proposer une nouvelle approche de recommandation hybride combinant plusieurs algorithmes de recommandation afin d'obtenir une recommandation plus precise. De plus, l'approche proposee repose sur la structure de l'ontologie donnee en entree du systeme, ce qui la rend reutilisable, facilement adaptable et applicable a tous les domaines (musique, publications scientifiques, films, etc.).Nous nous sommes egalement interessees a la detection du type de recommandations auxquelles l'utilisateur repond le mieux afin d'adapter l...

This research paper represents the techniques and approaches which are used in the movie recommendation system. As we are very well aware about the fact that extracting meaningful data from the homogenous amount of raw data is a challenging problem and recommendation systems helps us in this situation. Recommendation system plays a very important role in our day-today life as it provides suggestions based on some data sets to users for certain resources such as movies, books, songs, etc. Recommendation systems are very fruitful for various organizations as large amount of data is being collected from various customers and after extracting the data's it provides best suggestions. Movie recommendation systems is helping peoples who are fond of watching movies by providing suggestions for what movie to go with without going through the large set of movies data. To reduce the human efforts by providing suggestions of movies based on the user interest is our main moto. Recommendation system is based on three approaches: first one is Collaborative Filtering; second one is Content based and third one is hybrid-based Approach.

Web mining has been widely used to discover knowledge from various sources in the web. One of the important tools in web mining is mining of web user's behavior that is considered as a way to discover the potential knowledge of web user's interaction. Nowadays, Website personalization is regarded as a popular phenomenon among web users and it plays an important role in facilitating user access and provides information of users' requirements based on their own interests. Extracting important features about web user behavior plays a significant role in web usage mining. Such features are page visit frequency in each session, visit duration, and dates of visiting a certain pages. This paper presents a method to predict user's interest and to propose a list of pages based on their interests by identifying user's behavior based on fuzzy techniques called fuzzy clustering method. Due to the user's different interests and use of one or more interest at a time, user's interest may belong to several clusters and fuzzy clustering provide a possible overlap. Using the resulted cluster helps extract fuzzy rules. This helps detecting user's movement pattern and using neural network a list of suggested pages to the users is provided.

Pencarian pada database yang biasa dilakukan mahasiswa hanya mampu mencari judul yang sesuai berdasarkan kata kunci yang diinputkan, misalnya, jika kata kunci yang dimasukkan adalah "sistem cerdas" maka akan ditampilkan semua dokumen yang mengandung kata "sistem cerdas" namun sistem tidak bisa mengukur mana dokumen yang paling mirip. Untuk dapat melakukan pencarian berdasar substansi yang paling mirip, terdapat teknologi yang disebut information Text Retrieval. Dalam penelitian ini akan dikembangkan suatu sistem temu kembali informasi judul tugas akhir dan perhitungan kemiripan dokumen menggunakan vector space model. Sistem secara otomatis akan melakukan indexing secara offline dan temu kembali (retrieval) secara real time. Proses retrieval dimulai dengan mengambil query dari pengguna, menerapkan stop word removal sehingga dihasilkan keyword yang compaq tetapi dapatmewakili query tersebut, kemudian sistem menghitung kemiripan antarakeyword dengan daftar dokumen yang diwakili oleh term-term di dalam index. Dokumen akan ditampilkan diurutkan berdasarkan dokumen yang paling mirip.Dari hasil pengujian terlihat ketika keyword "android" dimasukkan maka akan tampil empat dokumen yang diurutkan sesuai tingkat kemiripannya, yaitu docId 3 dengan tingkat kemiripan 0.9512, docId 4 dengan tingkat kemiripan 0.5020, docId 2 dengan tingkat kemiripan 0.2671, docId 8 dengan tingkat kemiripan 0.1522.