Most of the traditional recommendation algorithms are based on TF-IDF, a term-based weighting method. This paper proposes a new method for recommending news items based on the weighting of the occurrences of references to concepts, which we call Concept Frequency-Inverse Document Frequency (CF-IDF). In an experimental setup we apply CF-IDF to a set of newswires in which we detect 1,167 instances of a set of 65 concepts from a domain ontology. The proposed method yields significantly better results with respect to accuracy, recall, and F1 than the TF-IDF method we use as a basis for comparison.
|Number of pages||2|
|Publication status||Published - 2 Nov 2011|
|Event||Twenty-Third Benelux Conference on Artificial Intelligence (BNAIC 2011) - Gent, Belgium|
Duration: 2 Nov 2011 → 3 Nov 2011
|Conference||Twenty-Third Benelux Conference on Artificial Intelligence (BNAIC 2011)|
|Period||2/11/11 → 3/11/11|