"Do not deceive me anymore!" interpretation through model design and visualization for instagram counterfeit seller account detection. (December 2022)

• Record Type: Journal Article
• Authors: Park, Jeongeun; Gu, Jinmo; Kim, Ha Young
• Is Part Of: Computers in human behavior. Volume 137(2022)
• Publication Date: 2022-12
• Subjects: Deep learning -- User experience -- Instagram -- Social problem -- Multimodal ensemble model -- Explainable artificial intelligence; Interactive computer systems -- Periodicals; Man-machine systems -- Periodicals; 004.019
• DOI: 10.1016/j.chb.2022.107418
• Languages: English
• ISSNs: 0747-5632

Abstract

Abstract: Counterfeit goods sold online have amounted to $330 billion in losses worldwide. Of these, luxury brand losses account for $30.3 billion. The social and economic effects of counterfeit goods have become more severe with the shift of counterfeit sales channels to social networking services (SNSs) since the first decade of the 2000s when these networks became active. In particular, the number of counterfeit sellers on Instagram increased by more than 171% in 2019. This study endeavors to detect counterfeit seller accounts in the SNS environment through a deep learning method that implements an algorithm to differentiate legitimate and counterfeit seller accounts. We designed a model to classify counterfeit seller accounts and general accounts using image and textual data collected from Instagram. As a result, the proposed model obtained a final account detection accuracy of 100% and demonstrated the possibility of use as a counterfeit seller account detector. Moreover, the elements that influenced the results and the parts that the model focused on were identified using a visual analysis method to improve the interpretation capability of the algorithm and explain the results. The difference between human and machine judgments was analyzed based on the visualization. Highlights: Proposed a deep learning model to detect counterfeit seller accounts on Instagram. Image and text classifiers can be used individually according to the purpose. A visual analysis method is used to increase the interpretability of the model. Human and machine judgments are compared to distinguish counterfeit seller accounts.