Deep Contextualized Pairwise Semantic Similarity for Arabic Language Questions

Abstract

Question semantic similarity is a challenging and active research problem that is very useful in many NLP applications, such as detecting duplicate questions in community question answering platforms such as Quora. Arabic is considered to be an under-resourced language, has many dialects, and rich in morphology. Combined together, these challenges make identifying semantically similar questions in Arabic even more difficult. In this paper, we introduce a novel approach to tackle this problem, and test it on two benchmarks; one for Modern Standard Arabic (MSA), and another for the 24 major Arabic dialects. We are able to show that our new system outperforms state-of-the-art approaches by achieving 93% F1-score on the MSA benchmark and 82% on the dialectical one. This is achieved by utilizing contextualized word representations (ELMo embeddings) trained on a text corpus containing MSA and dialectic sentences. This in combination with a pairwise fine-grained similarity layer, helps our question-to-question similarity model to generalize predictions on different dialects while being trained only on question-to-question MSA data.

Recent Developments in Arabic Conversational AI: A Literature Review

“…A few works in the literature have attempted to address this task for the Arabic language, and acceptable results have been achieved compared to other languages [44], [45], [46], [47], [48]. Due to the lack of relevant Arabic semantic corpora, rulebased approaches are currently used to achieve this task.…”

“…Al-Bataineh et al [47] proposed a new end-to-end prediction model to address QSS. They overcame the morphology and out-of-vocabulary (OOV) issues of Arabic using character-level CNN.…”

“…Unlike English, which has a variety of large datasets, Arabic conversational AI systems have been hindered by the deficiency of massive datasets. Arabic datasets are scarce, limited in size, and not diverse enough to be adaptable to the three discussed types of conversational AI systems [47]. Even with the massive Arabic content on the internet, obtaining large datasets is difficult because the datasets must be annotated.…”

“…However, they suffer from the lack of labeled datasets available for these dialects. Moreover, even if such datasets were available, it would be difficult to train such models [47].…”

A Panoramic Survey of Natural Language Processing in the Arab World

“…Performing such a task may employ a large set of NLP tools such as parsing, NER, co-reference resolution, and text semantic representation. There has been limited research on this problem [13,145] and existing commercial solutions such as Ujeeb.com are rudimentary.…”