Jordi Atserias

Ranking very many typed entities on wikipedia

We discuss the problem of ranking very many entities of different types. In particular we deal with a heterogeneous set of types, some being very generic and some very specific. We discuss two approaches for this problem: i) exploiting the entity containment graph and ii) using a Web search engine to compute entity relevance. We evaluate these approaches on the real task of ranking Wikipedia entities typed with a state-of-the-art named-entity tagger. Results show that both approaches can greatly increase the performance of methods based only on passage retrieval.

Learning to Tag and Tagging to Learn: A Case Study on Wikipedia

The problem of semantically annotating Wikipedia inspires a novel method for dealing with domain and task adaptation of semantic taggers in cases where parallel text and metadata are available.

World knowledge in broad-coverage information filtering

Document retrieval is a well-understood problem, as a consequence search technology has been able to support growth and innovation in scientific and industrial domains. As the Web evolves new types of content emerge: blogs and other types of communities, often based on multimedia content sharing, feeds for information browsing and delivery, vertical domains for shopping and performing other transactions, Web advertising. New types of content are challenging for traditional IR approaches based on the idea that the unit of information is a full “document”, whose content can be reasonably approximated with a bag-of-words representation, and whose ranking or matching can be assessed in isolation, within the vector space model paradigm. News filtering tasks involve monitoring a stream of news to identify stories belonging to a predefined set of categories. In the context of opinion mining we define five categories which capture the “polarity” of the story, a graded positive/negative opinion, with respect to a company. Financial news services are an important component of major Internet content providers (e.g., Google Finance, Yahoo! Finance). Financial news and stock prices tend to be correlated, and opinions and trends of financial stories can be modeled to a certain extent [2, 3, 7]. Tools for automated monitoring of such news could be valuable to users, financial analysts or small investors. Previous work has focused on clues for polarity such as selected keywords (plunge, surge, etc.), instead we consider the task of classifying all news stories that are relevant to a set of companies, coming through a stream. We present an exploratory study on the problem of classifying financial news stories streamed through RSS feeds. In particular, we focus on news stories titles. The reason for this is threefold. First, processing titles is faster than full documents and allows monitoring of larger numbers of sources efficiently. Secondly, and more interestingly, humans seems to be capable of performing such a task effortlessly on the basis of the little information provided by the title, even with no specific domain expertise. Furthermore, processing

El paper de la Lingüística Computacional en la cerca d'informació

Aquest article presenta una visio general de les noves aplicacions de cerca d’informacio que estan emergint fent us de la semantica superficial. Es presenten tres exemples d’aquest tipus d’aplicacions desenvolupades per la fundacio Barcelona media-Yahoo! Research Barcelona, Yahoo! correlator, TimeExplorer i Quest.