Publicação
Prometheus: a generic e-commerce crawler for the study of business markets and other e-commerce problems
| Resumo: | The continuous social and economic development has led over time to an increase in consumption, as well as greater demand from the consumer for better and cheaper products. Hence, the selling price of a product assumes a fundamental role in the purchase decision by the consumer. In this context, online stores must carefully analyse and define the best price for each product, based on several factors such as production/acquisition cost, positioning of the product (e.g. anchor product) and the competition companies strategy. The work done by market analysts changed drastically over the last years. As the number of Web sites increases exponentially, the number of E-commerce web sites also prosperous. Web page classification becomes more important in fields like Web mining and information retrieval. The traditional classifiers are usually hand-crafted and non-adaptive, that makes them inappropriate to use in a broader context. We introduce an ensemble of methods and the posterior study of its results to create a more generic and modular crawler and scraper for detection and information extraction on E-commerce web pages. The collected information may then be processed and used in the pricing decision. This framework goes by the name Prometheus and has the goal of extracting knowledge from E-commerce Web sites. The process requires crawling an online store and gathering product pages. This implies that given a web page the framework must be able to determine if it is a product page. In order to achieve this we classify the pages in three categories: catalogue, product and ”spam”. The page classification stage was addressed based on the html text as well as on the visual layout, featuring both traditional methods and Deep Learning approaches. Once a set of product pages has been identified we proceed to the extraction of the pricing information. This is not a trivial task due to the disparity of approaches to create a web page. Furthermore, most product pages are dynamic in the sense that they are truly a page for a family of related products. For instance, when visiting a shoe store, for a particular model there are probably a number of sizes and colours available. Such a model may be displayed in a single dynamic web page making it necessary for our framework to explore all the relevant combinations. This process is called scraping and is the last stage of the Prometheus framework. |
|---|---|
| Autores principais: | Dias, João Tiago Pereira |
| Assunto: | E-commerce Web mining Web page classification Machine learning Crawler Scraper Ciências Naturais::Ciências da Computação e da Informação |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | The continuous social and economic development has led over time to an increase in consumption, as well as greater demand from the consumer for better and cheaper products. Hence, the selling price of a product assumes a fundamental role in the purchase decision by the consumer. In this context, online stores must carefully analyse and define the best price for each product, based on several factors such as production/acquisition cost, positioning of the product (e.g. anchor product) and the competition companies strategy. The work done by market analysts changed drastically over the last years. As the number of Web sites increases exponentially, the number of E-commerce web sites also prosperous. Web page classification becomes more important in fields like Web mining and information retrieval. The traditional classifiers are usually hand-crafted and non-adaptive, that makes them inappropriate to use in a broader context. We introduce an ensemble of methods and the posterior study of its results to create a more generic and modular crawler and scraper for detection and information extraction on E-commerce web pages. The collected information may then be processed and used in the pricing decision. This framework goes by the name Prometheus and has the goal of extracting knowledge from E-commerce Web sites. The process requires crawling an online store and gathering product pages. This implies that given a web page the framework must be able to determine if it is a product page. In order to achieve this we classify the pages in three categories: catalogue, product and ”spam”. The page classification stage was addressed based on the html text as well as on the visual layout, featuring both traditional methods and Deep Learning approaches. Once a set of product pages has been identified we proceed to the extraction of the pricing information. This is not a trivial task due to the disparity of approaches to create a web page. Furthermore, most product pages are dynamic in the sense that they are truly a page for a family of related products. For instance, when visiting a shoe store, for a particular model there are probably a number of sizes and colours available. Such a model may be displayed in a single dynamic web page making it necessary for our framework to explore all the relevant combinations. This process is called scraping and is the last stage of the Prometheus framework. |
|---|