Annotated Corpus of Scientific Conference’s Homepages for Information Extraction

Annotated Corpus of Scientific Conference’s Homepages for Information Extraction Piotr Andruszkiewicz, Rafał Hazan Instititute of Computer Science, Warsaw University of Technology Warsaw, Poland P.Andruszkiewicz@ii.pw.edu.pl, R.Hazan@stud.elka.pw.edu.pl Abstract In this paper, we present a new corpus that contains 943 homepages of scientific conferences, 14794 including subpages, with annotations of interesting information: name of a conference, its abbreviation, place, and several important dates; that is, submission, notification, and camera ready dates. The topics of conferences included in the corpus are equally distributed over five areas: artificial intelligence, natural language processing, computer science, telecommunication, and image processing. The corpus is publicly available. Beside the characteristics of the corpus, we present the results of information extraction from the corpus using SVM and CRF models as we would like this corpus to be considered a reference data set for this type of task. Keywords: annotated corpus, scientific conference’s homepages, information extraction 1. Introduction Up-to-date information about conferences is important for scientists who track conferences of their interest and check, e.g., dates of a conference, deadlines, that could change especially during submission period. Thus, a system that gathers such information could ease scientists’ lives. The system should collect data about conferences and keep it up-to-date. Moreover, it should provide data in a structured way to facilitate searching conferences and obtaining infor- mation about any changes. The crucial part of this system are methods for collecting data about conferences automat- ically, e.g., homepages of a conference for the current and previous years, when and where a conference will be held, submission, notification, camera ready dates, etc. In this paper, we present a new corpus that contains home- pages of conferences with annotations of interesting infor- mation, e.g., name of a conference, its abbreviation, several important dates for the conference. The motivation behind this task was that according to our knowledge there is no any publicly available corpus in such a domain. The cor- pus can be used to train a tool for information extraction from unstructured sources containing data describing con- ferences. We chose conference home pages as a source as they contain up-to-date information. Structured services, such as WikiCFP, do not always update information (e.g., when deadline changes) and cannot be used in a real system for gathering up-to-date information about conferences. Beside the characteristics of the corpus, we present results of information extraction as a baseline and a proof that this corpus can be used as a refer- ence data set for information extraction from home- pages of conferences. The corpus is publicly avail- able and can be downloaded from the following website http://ii.pw.edu.pl/∼pandrusz/data/conferences/. The remainder of this paper is organised as follows: In Sec- tion 2. we describe the corpus we created. In Sections 3. the preprocessing and features are described. The ex- perimental results are discussed in Section 4.. Section 5. presents related work. Finally, Section 6. summarizes the conclusions of the study and outlines avenues to explore in the future. 2. The Corpus On the internet one can find corpora for information ex- …

Content truncated to 500 words