Automatic titling of Articles Using Position and Statistical Information

Proceedings of Recent Advances in Natural Language Processing, pages 727–732, Hissar, Bulgaria, 12-14 September 2011. Automatic titling of Articles Using Position and Statistical Information C´edric Lopez, Violaine Prince, and Mathieu Roche LIRMM - CNRS - University of Montpellier 2, France {lopez, prince, mroche}@lirmm.fr Abstract This paper describes a system facilitating information retrieval in a set of textual documents by tackling the automatic ti- tling and subtitling issue. Automatic ti- tling here consists in extracting relevant noun phrases from texts as candidate titles. An original approach combining statisti- cal criteria and noun phrases positions in the text helps collecting relevant titles and subtitles. So, the user may benefit from an outline of all the subjects evoked in a mass of documents, and easily find the informa- tion he/she is looking for. An evaluation on real data shows that the solutions given by this automatic titling approach are rele- vant. 1 Introduction Web pages contain a multitude of information concerning many domains. Very often, the user has to supply heavy cognitive efforts to find the information he/she is looking for. For handi- capped persons, while the access to Internet is a tremendous vector of integration in society, the localization of information remains complex. One of the key domains of web pages accessibility, such as defined by a standard proposed by hand- icap associations (W3C standard), concerns the titling (and subtitling) of web pages. The main goal is to increase the legibility of pages obtained from a search engine, where the relevance of results is often weak, disheartening readers, or to improve pages indexing, in order to obtain a better search. Besides, automatic titling can be integrated into diverse applications. For instance, it might help the editorial staff, proposing to the author of a given text, a segmented version, according to the issue tackled by (Akrifed, 2000; Prince and Labadi´e, 2007) and automatically titled. So, a new industrial application, based on automatic titling, would include the automatic generation of contents, saving time. One of the major benefits of the system de- scribed in this paper, is to help the user in assimi- lating the semantic contents of a set of textual doc- ument. Another is to allow him/her to quickly find the relevant information. Applied to textual re- sources, the proposed approach consists in provid- ing texts subjects by using the automatically gen- erated titles, and so to facilitate information com- munication and localization. Titles determination requires to know titles morphosyntactic structure, as well as their associated subtitles. From some statistical studies, performed on data described in section 3, concerning morphosyntactic character- istics, we propose a two-stages process. The main idea is to extract, from a given text, the most rele- vant noun phrase and use it as title. The first stage consists in extracting all noun phrases existing in the text (section 4.1). The second stage determines the most relevant phrase among those previously extracted (section 4.2). An evaluation, performed by human judgment on real data, is presented (sec- tion 5) and …

Content truncated to 500 words