Brett Drury

An Error Correction Methodology for Time Dependent Ontologies

An increasing number of applications have become dependent upon information described in ontologies. Information may be correct for a limited period of time, for example, the assertion: ”Barack Obama is the current president of the USA” will be incorrect in 2017. A presidential lifespan can be measured in years, however in a more dynamic domain, assertions may have lifespans of: months, weeks or days. In addition, erroneous relations may be introduced into an Ontology through mistakes in the information source or construction methodology. Ontologies which contain a large number of errors may impair the effectiveness of applications which depend on it. This paper describes an error correction methodology for ontologies automatically generated from news stories. The information contained in news stories can have a very limited lifespan, consequently constructing an Ontology by an addition of assertions will overtime accumulate errors. The proposed method avoids this problem through an assignment of a lifespan to each relation. A relation’s lifespan is dependent upon: frequency of assertion, relation volatility and domain volatility. Once a relation’s lifespan has elapsed the relation is either deleted or archived as a temporal ”snapshot” of the domain. Individuals with 0 relations are also removed or archived. An evaluation of an Ontology constructed with the proposed scheme revealed a gain in the total number of relations overtime without an increase in the number of the errors. A comparison with an Ontology constructed with an accumulative addition of relations over an eight week period revealed that the proposed method reduced the error count by 81%.

Predicting Market Direction from Direct Speech by Business Leaders

Direct quotations from business leaders can communicate to the wider public the latent state of their organization as well as the beliefs of the organizations leaders. Candid quotes from business leaders can have dramatic effects upon the share price of their organization. For example, Gerald Ratner in 1991 stated that his companys products were crap and consequently his company (Ratners) lost in excess of 500 million pounds in market value. Information in quotes from business leaders can be used to make an estimation of the organizations immediate future financial prospects and therefore can form part of a trading strategy. This paper describes a contextual classification strategy to label direct quotes from business leaders contained in news stories. The quotes are labelled as either: 1. positive, 2. negative or 3. neutral. A trading strategy aggregates the quote classifications to issue a buy, sell or hold instruction. The quote based trading strategy is evaluated against trading strategies which are based upon whole news story classification on the NASDAQ market index. The evaluation shows a clear advantage for the quote classification strategy over the competing strategies.

Construction and maintenance of a fuzzy temporal ontology from news stories

Construction of business ontologies from news sources can produce a detailed representation of the chosen area, but the ontology may over time gather errors because information can become rapidly outdated. The two stage strategy described in this paper attempts to identify outdated relations in an ontology by affixing a confidence score to each relation and decaying the relation until a preset value where it is deleted or archived. The relation score is refreshed if the information is repeated in a news story. An evaluation demonstrates that over time erroneous information is removed and new information is added to the ontology.