Research Reports

Permanent URI for this collectionhttp://dl.cerist.dz/handle/CERIST/34

Browse

Author: search.filters.author.Lebib, Fatma Zohra

Search Results

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Two-tower neural network for personalizing service recommendation in cloud environment
    (CERIST, 2025-02) Lebib, Fatma Zohra; Meziane, Abdelkrim
    Effective cloud service recommendation necessitates a deep understanding of both user preferences and the diversity of cloud services. This paper proposes a novel two-tower neural network architecture to address this challenge. By leveraging the power of neural networks, we create concise representations of users and cloud services, enabling highly personalized recommendations. Our two-tower architecture is designed to efficiently scale to the massive scale of cloud environments. Each tower, dedicated to users and cloud services respectively, integrates relevant features and interaction data to generate dense embeddings. Cloud services are then ranked based on their similarity to the user’s embedding, ensuring accurate and tailored suggestions. The proposed two-tower model was evaluated on the WSdream dataset for both regression and classification tasks. Experimental results consistently demonstrated superior performance compared to state-of-the-art recommendation techniques, including k-nearest neighbors and matrix factorization.
  • Thumbnail Image
    Item
    Knowledge Discovery from Log Data Analysis in a Multi-source Search System based on Deep Cleaning
    (CERIST, 2019-07) Lebib, Fatma Zohra; Mellah, Hakima; Meziane, Abdelkrim
    In a multi-source search system, understanding users’ interests and behaviour is essential to improve the search and adapt the results according to each user profile. The interesting information characterizing the users can be hidden in large log files, whereas it must be discovered, extracted and analyzed to build an accurate user profile. This paper presents an approach which analyzes the log data of a multi-source search system using the web usage mining techniques. The aim is to capture, model and analyze the behavioural patterns and profiles of users interacting with this system. The proposed approach consists of two major steps, the first step “pre-processing” eliminates the unwanted data from log files based on predefined cleaning rules, and the second step “processing” extracts useful data on user’s previous queries. In addition to the conventional cleaning process that removes irrelevant data from the log file, such as access of multimedia files, error codes and accesses of Web robots, deep cleaning is proposed, which analyzes the queries structure of different sources to further eliminate unwanted data. This allows to accelerate the processing phase. The generated data can be used for personalizing user-system interaction, information filtering and recommending appropriate sources for the needs of each user.
  • Thumbnail Image
    Item
    Selection of Information Sources using a Genetic Algorithm
    (CERIST, 2017-01-02) Lebib, Fatma Zohra; Drias, Habiba; Mellah, Hakima
    We address the problem of information sources selection in a context of a large number of distributed sources. We formulate the sources selection problem as a combinatorial optimization problem in order to yield the best set of relevant information sources for a given query. We define a solution as a combination of sources among a huge predefined set of sources. We propose a genetic algorithm to tackle the issue by maximizing the similarity between a selection and the query. Extensive experiments were performed on databases of scientific research documents covering different domains such as computer science and medicine. The results based on the precision measure are very encouraging.