Elamur Research⌕
☰
home / papers / 2022 / Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation

Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation

Paper · 2022 · Ons Meddeb, Mohsen Maraoui and Mounir Zrigui

Summary

The paper addresses recommending Arabic books and documents when user ratings are sparse. Pure collaborative filtering struggles with users or items that have few ratings, and most review-aware recommenders target English. Arabic text adds rich morphology and heavy lexical ambiguity. The authors build a hybrid recommender with two parallel towers, one for users and one for items. Each tower combines two signals:

  • dense ID embeddings learned from the rating matrix;
  • the texts of the reviews a user wrote or an item received, encoded with a compact fine-tuned AraBERT and then a GRU.

The user and item vectors are joined in a shared layer, and an MLP predicts the rating in place of a dot product. Tested on BRAD1.0 (Goodreads book reviews in Arabic), the model reaches an RMSE of 0.855. This beats PMF, SVD++, J-NCF and the CNN-based DeepCoNN. The authors read this as evidence that contextual embeddings capture Arabic review semantics better than static word vectors. Only one dataset and one metric are reported, and there is no ablation separating the AraBERT, GRU and MLP contributions.

Key points

  • Data is BRAD1.0: 4,993 Arabic books of various genres, 76,530 users and 510,600 interactions, each with a review and a 1–5 rating.
  • The AraBERT encoder is small (4 layers, 4 heads, hidden size 256) and fine-tuned during training. Each review is fed as a single segment of 256 tokens.
  • The GRU stack uses 256/128/64 units. The interaction MLP has hidden layers of 128/64/8 and K=16 latent factors, which the authors found to be the best setting.
  • RMSE results: PMF 1.355, SVD++ 1.215, J-NCF 1.050, DeepCoNN 0.905, proposed model 0.855.
  • Planned future work is a bidirectional recurrent layer, so that both preceding and following words shape a word's representation.

Interesting ideas

  • I1. Users and items get identically built towers that differ only in input, so one architecture learns profiles for both sides of a relation; this keeps the design simple and symmetric.
  • I2. Each side fuses an identity-based signal from the rating matrix with a content-based signal from text, so sparse identity data is backed up by descriptive evidence.
  • I3. The baselines form a ladder by information used: ratings only, then non-linear interaction, then reviews, then contextual reviews. A feature table states exactly what each method sees, which makes it clearer where gains come from.
  • I4. A deliberately compact transformer (4 layers, 256 hidden) is fine-tuned end-to-end inside a larger pipeline rather than used as a frozen feature extractor. This shows a light encoder can be one trainable component among others.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (world): An entity profile could be assembled the same way from every clause where the entity acts and every clause where it is acted upon, using one shared procedure for both views.
  • I2 (learn): When pooling silver annotations, each witness parser's track record (an identity signal) could be combined with features of the sentence at hand (a content signal) to weigh disagreements.
  • I3 (learn): Evaluation reports could carry a similar matrix marking which knowledge sources each Mova variant uses (VerbNet, WordNet, LLM teacher), turning comparisons into crude ablations.
  • I4 (tree): gparse could test small, jointly fine-tuned encoder configurations to cut GPU cost while staying trainable together with the parser head.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Ons Meddeb, Mohsen Maraoui and Mounir Zrigui. 2022. Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation. Proceedings of the 36th Pacific Asia Conference on Language, Information and Computation.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{meddeb2022contextual,
  title = {{Contextual-Boosted Deep Neural Collaborative Filtering Approach for Arabic Textual Documents Recommendation}},
  author = {Ons Meddeb, Mohsen Maraoui and Mounir Zrigui},
  booktitle = {Proceedings of the 36th Pacific Asia Conference on Language, Information and Computation},
  year = {2022},
  url = {https://aclanthology.org/2022.paclic-1.54/}
}

References

32 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Bai T., Wen J., Zhang J., and Zhao W. X. 2017. A neural collaborative filtering model with interaction-based neighborhood, ACM on Conference on Information and Knowledge Management (ICKM): 1979–1982. find
  2. Bao Y., Fang H., Zhang J. 2014. TopicMF: simultaneously exploiting ratings and reviews for recommendation, 28 th AAAI Conference on Artificial Intelligence, Quebec, Canada: 2-8. find
  3. Bellagha M. L., and Zrigui M. 2020. Speaker naming in tv programs based on speaker role recognition, IEEE/ACS 17 th International Conference on Computer Systems and Applications (AICCSA), Antalya, Turkey: 1-8. find
  4. Chen C., Zhang M., Liu Y., and Ma S. 2018. Neural attentional rating regression with review-level explanations, International World Wide Web Conferences Steering Committee (WWW): 1583– 1592. find
  5. Chen W., Cai F., Chen H., and Rijke M. D. 2019. Joint neural collaborative filtering for recommender systems, ACM Transactions on Information Systems, 1(1): 1-30. find
  6. Devlin J., Chang M., Lee K., and Toutanova K. 2018. Bert: pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. ↗
  7. Dhelim S., Aung N., Bouras M., Ning H., and Cambria E. 2022. A survey on personality-aware recommendation systems, Artificial Intelligence Review, volume 55: 2409-2454. find
  8. Duan R., Jiang C., and Jain H. 2022. Combining review-based collaborative filtering and matrix factorization: A solution to rating’s sparsity problem, Decision Support Systems, volume 156:113748. find
  9. Fang J., Zhang X., Hu Y., Xu Y., Yang M., and Liu J. 2020. Probabilistic latent factor model for collaborative filtering with Bayesian inference, 25 th International Conference on Pattern Recognition (ICPR), Milan, Italy: 73-80. find
  10. Haffar N., Ayadi R., Hkiri E., and Zrigui M. 2021. Temporal ordering of events via deep neural networks, 16 th International Conference on Document Analysis and Recognition (ICDAR), Lausanne, Switzerland: 762-777. find
  11. Hazar M. J., Maraoui M., and Zrigui M. 2022. Recommendation system based on video processing in an E-learning platform, Journal of Human University (Natural Sciences), 49(6): 49-61. find
  12. He X., Liao L., Zhang H., Nie L., and Chua T.S. 2017. Neural collaborative filtering, 26 th International Conference on World Wide Web (WWW), Perth Australia: 173–182 find
  13. Jian P. K., Patel A., Kuari S., and Pamula R. 2022. Predicting airline customers’ recommendations using qualitative and quantitative contents of online reviews, Multimedia and Tools, volume 81: 6979-6994. find
  14. King G. L. L. 2014. Ratings meet reviews, a combined approach to recommend, 8 th ACM Conference on Recommender Systems (RecSys), Silicon Valley, CA, USA: 105–112. find
  15. Lee D., and Seung H. S. 2001. Algorithms for non-negative matrix factorization, Advances in Neural Information Processing Systems (NIPS), Vancouver, BC, Canada: 556–562. find
  16. Mahmoud A., and Zrigui M. 2017. Semantic similarity analysis for paraphrase identification in Arabic texts, 31 st Pacific Asia Conference on Language, Information and Computation (PACLIC), Philippine: 274-281.
  17. Mahmoud A., and Zrigui M. 2021a. BLSTM-API: Bi-LSTM recurrent neural-based approach for Arabic paraphrase identification, Arabian for Science and Engineering, volume 46: 4163-4174. find
  18. Mahmoud A., and Zrigui M. 2021b. Semantic similarity analysis for corpus development and paraphrase detection in Arabic, International Arab Journal of Information Technology (IAJIT), volume 18: 1-7. find
  19. McAuley J., and Leskovec J. 2013. Hidden factors and hidden topics: understanding rating dimensions with review text, 7 th ACM Conference on Recommender systems, Hong Kong, China: 165– 172. find
  20. Meddeb O., Maraoui M., Aljawarneh S. 2017. Hybrid modelling of an off line Arabic handwriting recognition system: results and evaluation, International Journal Intelligent Enterprise, 4(½): 168–189. find
  21. Meddeb O., Maraoui M., and Zrigui M. 2021a. Deep learning based semantic approach for Arabic textual documents recommendation, IEEE International Conference on INnovations in Intelligent SysTems and Applications (INISTA), Kocaeli, Turkey: 1-6. find
  22. Meddeb O., Maraoui M., and Zrigui M. 2021b. Arabic text documents recommendation using joint deep representations learning, Procedia Computer Science, 192(1): 812-821. find
  23. Meddeb O., Maraoui M., and Zrigui M. 2021c. Personalized smart learning recommendation system for Arabic users in smart campus, International Journal of Web-based learning and Teaching Technologies, 51(5): 1734-1747. find
  24. Mnih A., and Salakhutdinov R. 2007. Probabilistic matrix factorization, Advances in Neural Information Processing Systems, volume 20: 1257-1264. find
  25. Saeed R. M. K., Rady S., and Gharib T. F. 2021. Optimizing sentiment classification for Arabic opinion texts, Cognitive Computation, volume 13: 164–178. find
  26. Sghaier M. A., and Zrigui M. 2020. Rule-based machine translation from Tunisian dialect to modern Arabic standard, Procedia Computer Science, volume 196: 310-319 find
  27. Slimi A., Zrigui M., and Nicolas H. 2022. MuLER: Multiplet-loss for emotion recognition, International Conference on Multimedia Retrieval (ICMR), Newark, NJ, USA: 435-442. find
  28. Sun C., Qiu X., Xu Y., and Huang X. 2019. How to fine-tune bert for text classification?: arXiv preprint arXiv:1905.05583. ↗
  29. Trigui A., Mars A., Ben Jannet M. A., Maroui M., and Zrigui M. 2022. Foreign accent classification for Arabic speech learning: 1-5. find
  30. Wang X., He X., Nie L., and Chua T. 2017. Item silk road: Recommending items from information domains to social users, 40 th International ACM SIGIR Conference on Research and Development in Information Retrieval: 185–194. find
  31. Zhang Y., Ai Q., Chen X., and Croft W. 2017. Joint representation learning for topn recommendation with heterogeneous information sources. ACM on Conference on Information and Knowledge Management (ICKM): 1449–1458. find
  32. Zheng L., Noroozi V., and Yu P. S. 2017. Joint deep modeling of users and items using reviews for recommendation, 10 th ACM International Conference on Web Search and Data Mining (WSDM): 425-434. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/meddeb-2022-contextual-boosted-deep-neural-collaborative/