Article Preview
TopPersonal Knowledge Background For Article Selection
A classical approach is to use the cosine similarity by frequency of keywords for article selection. However, this is often insufficient, as it may be too rough in decision. One improvement is the Graph Model (Wu et al., 2009) that considers both the frequency and position property of keywords in articles to obtain a more accurate result by calculating the similarity of two articles. Thus, this method can also used for detecting fraudulent or plagiarized articles, which can be hardly done by only using keyword frequency. There are also other methods proposed to improve the quality of article selection, such as the use of ontology in text classification (Yang et al., 2009), interest mode (Zhao et al., 2011; Teevan et al., 2005) and semantic-based methods (Lv & Liu, 2005).
However, those methods are not dealing with text documents with personal preferences. A word may often have different meanings to different people and thus it is reasonable to obtain different results with the same keywords given when personalization is considered. In this sense, a keyword is not purely a symbol but a symbolized representation of a concept that can be personal in terms of its meaning.
As a research topic, personalized article selection has attracted a lot of attention for its possible applications in Search Engine Optimization (SEO), such as Brin et al. (1999) and Leubner and Kießling (2002). However, most of them capture personal preferences only in a limited space by a rather inflexible set of features represented by keywords, and unable to represent a broad range of user long-term interest and knowledge with incremental learning.