News item recommendation is commonly performed using the TF-IDF weighting technique in combination with the cosine similarity measure. However, this technique does not take into account the actual meaning of words. Therefore, we propose two new methods based on concepts and their semantic similarities, from which we derive the similarities between news items. Our first method, Synset Frequency - Inverse Document Frequency (SF-IDF), is similar to TF-IDF, yet it does not use terms, but WordNet synonym sets. Additionally, our second method, Semantic Similarity (SS), makes use of five semantic similarity measures to compute the similarity between news items for news recommendation. Test results show that SF-IDF and SS outperform the TF-IDF method on the F 1-measure. Copyright 2012 ACM.

Additional Metadata
Keywords Content-based recommender, News personalization, Recommender systems, Semantic similarity, Semantic Web, User profiling
Persistent URL dx.doi.org/10.1145/2254129.2254163, hdl.handle.net/1765/90471
Conference 2nd International Conference on Web Intelligence, Mining and Semantics, WIMS 2012
Citation
Capelle, M, Moerland, M, Frasincar, F, & Hogenboom, F.P. (2012). Semantics-based news recommendation. Presented at the 2nd International Conference on Web Intelligence, Mining and Semantics, WIMS 2012. doi:10.1145/2254129.2254163