Semantics-based news recommendation with SF-IDF+

摘要: Content-based news recommendations are usually made by employing the cosine similarity and TF-IDF weighting scheme for terms occurring in messages user profiles. Recent developments, such as SF-IDF, have elevated recommendation to a new level of abstraction additionally taking into account term meaning through exploitation synsets from semantic lexicons similarity. Other state-of-the-art recommenders, like SS, make use lexicon-driven similarities. A shortcoming current recommenders is that they do not take various relationships between synsets, providing only limited understanding semantics. Therefore, we extend SF-IDF technique considering synset lexicon. The proposed method, SF-IDF+, well several methods been implemented Ceryx, an extension Hermes personalization service. An evaluation on data set containing financial shows overall (by accounting all considered cut-off values) SF-IDF+ outperforms TF-IDF, F1-scores.