Suggest me a movie for tonight: Leveraging Knowledge Graphs for Conversational Recommendation

Proceedings of the 28th International Conference on Computational Linguistics, pages 407–413 Barcelona, Spain (Online), December 8-13, 2020 407 Leveraging HTML in Free Text Web Named Entity Recognition Colin Ashby and David Weir TAG Lab, Department of Informatics, University of Sussex, Brighton, UK {c.ashby, d.j.weir}@sussex.ac.uk Abstract HTML tags are typically discarded in free text Named Entity Recognition from Web pages. We investigate whether these discarded tags might be used to improve NER performance. We compare Text+Tags sentences with their Text-Only equivalents, over five datasets, two free text segmentation granularities and two NER models. We find an increased F1 performance for Text+Tags of between 0.9% and 13.2% over all datasets, variants and models. This performance increase, over datasets of varying entity types, HTML density and construction quality, indicates our method is flexible and adaptable. These findings imply that a similar technique might be of use in other Web-aware NLP tasks, including the enrichment of deep language models. 1 Introduction Named Entity Recognition (NER) is the identification of the proper names of objects. NER can be infor- mational in its own right, but also serves as pre-processing for other Information Extraction (IE) tasks. The Web as a data source for NER offers opportunities as a massive corpus of structured information and free text, but also presents challenges of a far from uniform layout and variable writing style; the Web is formatted for ease of reading, not for ease of extraction. Web page text elements can be divided into two categories: structured and unstructured free text. Structured text is contained within records, lists and tables. NER for these elements is usually per- formed with Wrapper methods that identify text areas using precise contextual patterns. HTML is an essential delimiter of records in Wrapper methods. Techniques for handling HTML tags range from string matching to CNN based approaches. Wrappers are insufficient for dealing directly with free text. Unstructured free text natural language elements, such as <p>, make up the majority of Web content. Most recent free text NER approaches make use of sentence based NLP neural network techniques (Goyal et al., 2018; Yadav and Bethard, 2019; Li et al., 2020), such as BERT (Devlin et al., 2018) or LSTM+CRF (Ma and Hovy, 2016), with pre-trained language models. HTML contained in these free text sentences is generally discarded. The reasons for this seem to be expectation, convenience and heritage. HTML in natural language is not expected to be of any benefit: “Free text exhibits no implicit structure at all” (Goebel and Ceresna, 2009). Any information to be extracted from free text is assumed to be held entirely in natural language grammar and semantics. Many NLP code libraries, such as Beautiful Soup (Richardson, 2007), strip HTML with a single function call: “Since so much text on the Web is in HTML format, we will also see how to dispense with markup.” (Bird et al., 2009). A plain text approach to NER is used in various genres of Web site, from Web newspapers (Ekbal et al., 2012; Wibawa …

Content truncated to 500 words