Detail publikace

Graph neural network for website element detection

Originální název

Graph neural network for website element detection

Anglický název

Graph neural network for website element detection

Jazyk

en

Originální abstrakt

Websites are a mixture of structured HTML tags, unstructured natural language and styling, which gives a wide range of possibilities how a website can look like. The paper introduces a website node detector based on the so-called graph neural networks - a new kind of neural networks, which are not working just with tensors like traditional neural networks do, but operates with graphs (or tree structures - special variations of graphs). To assess the accuracy of the proposed methodology, a privately collected and labeled data set was created. Although the data set used for the experiment is relatively limited, results on this limited data set suggest, that this methodology may be a promising path for automatic content generation.

Anglický abstrakt

Websites are a mixture of structured HTML tags, unstructured natural language and styling, which gives a wide range of possibilities how a website can look like. The paper introduces a website node detector based on the so-called graph neural networks - a new kind of neural networks, which are not working just with tensors like traditional neural networks do, but operates with graphs (or tree structures - special variations of graphs). To assess the accuracy of the proposed methodology, a privately collected and labeled data set was created. Although the data set used for the experiment is relatively limited, results on this limited data set suggest, that this methodology may be a promising path for automatic content generation.

BibTex


@inproceedings{BUT157768,
  author="Vojtěch {Myška} and Radim {Burget} and Peter {Brezany}",
  title="Graph neural network for website element detection",
  annote="Websites are a mixture of structured HTML tags, unstructured natural language and styling, which gives a wide range of possibilities how a website can look like. The paper introduces a website node detector based on the so-called graph neural networks - a new kind of neural networks, which are not working just with tensors like traditional neural networks do, but operates with graphs (or tree structures - special variations of graphs). To assess the accuracy of the proposed methodology, a privately collected and labeled data set was created. Although the data set used for the experiment is relatively limited, results on this limited data set suggest, that this methodology may be a promising path for automatic content generation.",
  address="IEEE",
  booktitle="42nd International Conference on Telecommunications and Signal Processing (TSP)",
  chapter="157768",
  doi="10.1109/TSP.2019.8769036",
  howpublished="online",
  institution="IEEE",
  year="2019",
  month="july",
  pages="216--219",
  publisher="IEEE",
  type="conference paper"
}