WebSRC: A Dataset for Web-Based Structural Reading Comprehension

Abstract

Web search is an essential way for humans to obtain information, but it's still a great challenge for machines to understand the contents of web pages. In this paper, we introduce the task of structural reading comprehension (SRC) on web. Given a web page and a question about it, the task is to find the answer from the web page. This task requires a system not only to understand the semantics of texts but also the structure of the web page. Moreover, we proposed Web-SRC, a novel Web-based Structural Reading Comprehension dataset. WebSRC consists of 400K question-answer pairs, which are collected from 6.4K web pages. Along with the QA pairs, corresponding HTML source code, screenshots, and metadata are also provided in our dataset. Each question in WebSRC requires a certain structural understanding of a web page to answer, and the answer is either a text span on the web page or yes/no. We evaluate various baselines on our dataset to show the difficulty of our task. We also investigate the usefulness of structural information and visual features. Our dataset and baselines have been publicly available 1 .

WebFormer: The Web-page Transformer for Structure Information Extraction

“…This section provides discussion that connects WebFormer with previous methods as well as the limitations of our model. If we treat HTML tags as additional text tokens, and combine with the text into a single sequence without the H2H, H2T and T2H attentions, our model architecture degenerates to the sequence modeling approaches [9,51] that serialize the HTML layout. If we further trim the HTML from the sequence, our model is regressed to the sequence model [47] that only uses the text information.…”

“…Recently, there has been an increasing number of works that develop natural language models with sequence modeling [9,20,26,30,34,61] for web information extraction. Zheng et al [59] develop an end-to-end tagging model utilizing BiLSTM, CRF, and attention mechanism without any dictionary.…”

“…More recently, several attribute extraction approaches [47,49,53] have been proposed, which treat each field as an attribute of interest and extract its corresponding value from clean object context such as web title. Chen et al [9] formulate the web information extraction problem as structural reading comprehension and build a BERT [15] based model to extract structured fields from the web documents. It is worth mentioning that there are also methods that work on multimodal information extraction [44,45,48,55], which focus on extracting the field information from the visual layout or the rendered HTML of the web documents.…”

“…Existing sequence modeling methods either directly model the text sequence from web document [26,47] or serialize the HTML with the text in a certain order [9,61] to perform the span based text extraction. In this work, we propose to simultaneously encode the text sequence using the Transformer model and incorporate the HTML layout structure with graph attention.…”

“…SimpDOM [61] treats the problem as DOM tree node tagging task by extracting the features for each text node including XPath, and uses a LSTM to jointly encode with the text features. H-PLM [9] sequentializes the HTML together with the text and builds a sequence model using the pre-training ELECTRA [11] as backbone.…”