Learning over No-Preferred and Preferred Sequence of Items for Robust Recommendation
Authors: Aleksandra Burashnikova, Yury Maximov, Marianne Clausel, Charlotte Laclau, Franck Iutzeler, Massih-Reza Amini
JAIR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Furthermore, we provide a convergence analysis of both algorithms and demonstrate their practical efficiency over six large-scale collections with respect to various ranking measures and computational time. [...] Furthermore, we provide empirical evaluation over six large publicly available datasets showing that both versions of SAROS are highly competitive compared to the state-of-the-art models in terms of quality metrics and, that are significantly faster than both the batch and the online versions of the algorithm. |
| Researcher Affiliation | Academia | Aleksandra Burashnikova EMAIL Skolkovo Institute of Science and Technology Moscow, 125349, Russia, and University Grenoble-Alpes Grenoble, CS 40700 38058, France Yury Maximov EMAIL Theoretical Division, T-5 Los Alamos National Laboratory, USA Los Alamos, NM 87545, USA Marianne Clausel EMAIL Department of Statistics University of Lorraine Nancy, F-54506, France Charlotte Laclau EMAIL Hubert Curien Laboratory Telecom Saint-Etienne Saint-Etienne, CS 82301-42023, France Franck Iutzeler EMAIL Jean Kuntzmann Laboratory University Grenoble Alpes Grenoble, F-38000, France Massih-Reza Amini EMAIL Grenoble Informatics Laboratory University Grenoble Alpes Grenoble, CS 40700 38058, France |
| Pseudocode | Yes | Algorithm 1 SAROSb: Sequenti Al Rec Ommender System for implicit feedback |
| Open Source Code | Yes | 3. The source code is available at https://github.com/Sasha Burashnikova/SAROS. |
| Open Datasets | Yes | Datasets. We report results obtained on five publicly available datasets, for the task of personalized Top-N recommendation on the following collections : ML-1M (Harper and Konstan, 2015) and Netflix (Bennett and Lanning, 2007b) datasets... We get Outbrain dataset from of the Kaggle challenge (Outbrain Inc., 2016)... Pandor(Sidana et al., 2018b) is another publicly available dataset for online recommendation (Sidana et al., 2018a) provided by Purch1... Rec Sys 16 is a sample based on historic XING data... Kasandr(Sidana et al., 2017a) dataset (Sidana et al., 2017b) contains 15,844,717 interactions... |
| Dataset Splits | Yes | To construct the training and the test sets, we discarded users who did not interact over the shown items and sorted all interactions according to time-based on the existing time-stamps related to each dataset. Furthermore, we considered 80% of each user s first interactions (both positive and negative) for training, and the remaining for the test. Table 3 presents the size of the training and the test sets as well as the percentage of positive feedback (preferred items) for all collections ordered by increasing training size. [...] Finally, we have used 10% of the most recent interactions of users in the training set as validation set for hyperparameter tuning (see below). |
| Hardware Specification | No | All subsequently discussed components were implemented in Python3 using the Tensor Flow library (Abadi et al., 2016). and computed on Skoltech CDISE HPC cluster ZHORES (Zacharov et al., 2019). |
| Software Dependencies | No | All subsequently discussed components were implemented in Python3 using the Tensor Flow library (Abadi et al., 2016). |
| Experiment Setup | Yes | Hyper-parameters of different models and the dimension of the embedded space for the representation of users and items; as well as the regularisation parameter over the norms of the embeddings for all approaches were found using grid search on the validation set. We fixed b and B, used in SAROSb, to respectively the minimum and the average number of blocks found on the training set of each corresponding collection. [...] For more details regarding the exact values for the parameters, see the Table 4. Table 4: Hyperparameter values of SAROSb. |