Mohamed Ramzi Haddad

Towards a new model for context-aware recommendation

With current growth of internet sales and content consumption, several research efforts focus on recommendation and personalization approaches as a solution to information overload. In this paper, we first propose a new context-aware recommendation model that inspires from consumption psychology researches. Then, two different techniques for generating recommendations from the proposed model are detailed and evaluated. The first is based on logistic regression and the second uses enumeration in order to calculate the probability a customer purchases a given item. We Also study and evaluate three strategies of recommender systems hybridization based on weighting and selection to eliminate the problems the underlying techniques have when applied solely. At the end, we conclude with some ideas for further development and research.

A personalised semantic and spatial information retrieval system based on user's modelling and accessibility measure

Search personalisation is a multi-criteria decision problem whose objective is to filter relevant information based on a set of criteria such as needs, interests and content semantics. Hereby, different users could enter the same query into a search system, but their information needs can be very different. Web personalisation can be seen as an interdisciplinary field whose objective is to facilitate the interaction between web content and users needs. It includes per definition several research domains from social to information sciences. The personalised search focuses on integrating users contexts, needs and relevancy criteria in the information retrieval process in order to help them finding the right content. This paper presents users network modelling for personalised spatial and semantic information retrieval. The idea is to provide a user with personalised results based on his model and on the neighbour users models. The spatial personalisation search is based on a measure of spatial accessibility, whose objective is to predict and evaluate location relevancy, accessibility and associations at the user level. This measure favours delivery of location-based and personalised recommendations. Our experiments confirm the effectiveness of our proposal by pointing out the improvement of the personalised search results when compared to a baseline web search.

Large-Scale Real-Time News Recommendation Based on Semantic Data Analysis and Users’ Implicit and Explicit Behaviors

Online news portals constantly produce a huge amount of content about different events and topics. In such data streams scenarios, delivering relevant recommendations that best suit each user’s interests is a challenging task. Indeed, tight-time constraints and highly dynamic conditions in these environments make traditional batch recommendation approaches ineffective. In this paper, we present a scalable news recommendation system that takes into account data semantics, trending topics, users’ behaviors and the usage context in order to (1) model news articles, (2) infer users’ preferences and (3) provide real-time suggestions. In fact, our proposal is based on the semantic analysis of news articles’ content in order to extract relevant keywords and referenced named entities. This information is then used to model users’ interests by analyzing their attitudes while interacting with the available content. Moreover, our proposition accounts for the temporal variance of a news article’s utility by considering its freshness, popularity and attractiveness. To prove our proposition’s quality, scalability and efficiency in real-time data streaming environments, it was evaluated during the CLEF-NEWSREEL challenge connecting recommender systems to an active large-scale news delivery platform. Experiment results show that our system produces high quality and reliable performances in such dynamic environments.

Distributed Collaborative Filtering for Batch and Stream Processing-Based Recommendations

Nowadays, user actions are tracked and recorded by multiple websites and e-commerce platforms, allowing them to better understand their preferences and support them with specific and accurate content suggestions. Researches have proposed several recommendation approaches and addressed several challenges such as data sparsity and cold start. However, the low-scalability problem remains a major challenge when handling large volumes of user actions data. This issue becomes more challenging when it comes to real-time applications. Such constraint requires a new class of low latency recommendation approaches capable of incrementally and continuously update their knowledge and models at scale as soon as data arrives. In this paper, we focus on the user-centered collaborative filtering as one of the most adopted recommendation approaches known for its lack of scalability. We propose two distributed and scalable implementations of collaborative filtering addressing the challenges and the requirements of batch offline and incremental online recommendation scenarios. Several experiments were conducted on a distributed environment using the MovieLens dataset in order to highlight the properties and the advantages of each variant.

PerceptRank: A Real-Time Learning to Rank Recommender System for Online Interactive Platforms

In highly interactive platforms with continuous and frequent content creation and obsolescence, other factors besides relevance may alter users’ perceptions and choices. Besides, making personalized recommendations in these application domains imposes new challenges when compared to classic recommendation use cases. In fact, the required recommendation approaches should be able to ingest and process continuous streams of data online, at scale and with low latency while making context dependent dynamic suggestions. In this work, we propose a generic approach to deal jointly with scalability, real-time and cold start problems in highly interactive online platforms. The approach is based on several consumer decision-making theories to infer users’ preferences. In addition, it tackles the recommendation problem as a learning-to-rank problem that exploits a heterogeneous information graph to estimate users’ perceived value towards items. Although the approach is addressed to streaming environments, it has been validated in both offline batch and online streaming scenarios. The first evaluation has been carried out using the MovieLens dataset and the latter targeted the news recommendation domain using a high-velocity stream of usage data collected by a marketing company from several large scale online news portals. Experiments show that our proposition meets real world production environments constraints while delivering accurate suggestions and outperforming several state-of-the-art approaches.

Towards a Generic Architecture for Recommenders Benchmarking

With current growth of internet sales and content consumption, more research efforts are focusing on developing recommendation and personalization algorithms as a solution for the choice overload problem. In this paper, we first enumerate several state-of-the-art recommendation algorithms in order to highlight their main ideas and methodologies. Then, we propose a generic architecture for recommender systems benchmarking. Using the proposed architecture, we implement and evaluate several variants of existing recommendation algorithms and compare their results to our unified recommendation model. The experiments are conducted on a real world dataset in order to assess the genericity of our recommendation model and its quality. At the end, we conclude with some ideas for further development and research.

Towards an Integration of Space and Accessibility in Web Personalization

Web personalization can be seen as an interdisciplinary domain that facilitates interaction between web content and user needs. One of the peculiarities of Web information is that a significant part of the data is georeferenced although this is not completely taken into account by current search and personalization engines. This paper introduces a spatial personalization approach based on a user modeling technique and a measure of spatial accessibility. We develop a personalized accessibility measure whose objective is to predict and evaluate location relevancy, accessibility and associations at the user level. This measure favors delivery of location-based and personalized recommendations.