Samsung Ranks 1st in the Top Semantic Evaluation International Competition

Samsung R&D Institute China–Beijing (SRC-B) took part in the SemEval (International Workshop on Semantic Evaluation) 2022 Task 5 MAMI (Multimedia Automatic Misogyny Identification) Reading Comprehension of Abstract Meaning and achieved 1st place in both subtasks. SRC-B’s related paper was also accepted by the said conference.

SRC-B’s team members

SemEval is an ongoing series of computational semantics systems evaluations organized by the Association for Computational Linguistics (ACL). ACL is one of the most influential semantic evaluation competitions. According to Google Scholar, articles published in SemEval are second only to those of ACL, the Empirical Methods in Natural Language Processing, or the North American Chapter of the Association for Computational Linguistics in terms of influence in the field of computational linguistics (also belong to Samsung R&D’s recommended conferences).

Subtask A results: A basic task on misogynous meme identification, where a meme should be categorized either as misogynous or not misogynous

Subtask B results: An advanced task where the type of misogyny should be recognized among potential overlapping categories, such as stereotype, shaming, objectification, and violence

Task 5 is a multimodal classification task targeting the identification of misogynous memes (i.e., widely spread social image data, characterized by pictorial content with an overlaying text introduced a posteriori by a human). The task involves using the meme’s image and text, making it a multimodal understanding problem.

A total of 83 teams from universities and companies participated in the MAMI task. The SRC-B team submitted the PBR system, an ensemble system of Pre-training models, Boosting method, and Rule-based adjustment. Text information is fused into the system using our late sequential fusion scheme. The system managed to win 1st place on subtask 1 and 2 in the leaderboard.