Item matching based on image and text analysis
| dc.contributor.author | Smagul O. | |
| dc.date.accessioned | 2025-04-14T07:01:01Z | |
| dc.date.available | 2025-04-14T07:01:01Z | |
| dc.date.issued | 2024 | |
| dc.description.abstract | In recent years, the e-commerce sector has experienced exponential growth as more retailers and brands have opened online sites to reach a worldwide consumer base. The rise in online marketplaces and comparison sites, where consumers may evaluate and buy goods from different vendors or companies, is another effect of the digital transformation. But this has also resulted in a proliferation of sellers and products, making it challenging for buyers to identify the ideal items and for vendors to connect with their intended clientele. Product matching has emerged as a critical problem for efficient decision-making in the retail and supply chain management sectors in response to these challenges. Purpose of the product matching is to match identical or nearly same products across various sources across, such as different e-commerce websites, based on features and their attributes. It can help retailers to find in demand products and boost sales. On the other hand, customers can take benefit from technologies by finding products they needed, compare across different aggregators, and make decisions to purchase. However, product matching is a challenging task. In most cases same products have different names, descriptions and image cards. With its recent breakthroughs in object detection and image categorization, deep learning has come a long way. A convolutional neural network may identify identical products based on the input picture and text, which might be a label or tag. In this work, three pretrained deep convolutional models: MobileNet-V2, VGG-19, and ResNet-50, are implemented to find the most identical products based on text and image data. The best model for similarity detection based on text and image has been found using a variety of performance assessment techniques, including cosine similarity, Levenshtein distance, and custom metric score. The study concludes that the Mobilenet model is the best suitable model for handling these laborious tasks and supporting the development of a digital strategy plan. | |
| dc.identifier.citation | Smagul O / Item matching based on image and text analysis / 2024 / Computer Science - 7M06102 | |
| dc.identifier.uri | https://repository.sdu.edu.kz/handle/123456789/1705 | |
| dc.language.iso | en | |
| dc.publisher | Faculty of Engineering and Natural Science | |
| dc.title | Item matching based on image and text analysis | |
| dc.type | Other |