Quantifying textual terms of items for similarity measurement

It is well known that recommender systems rely on the similarity between items to be recommended. Most current research projects in this area utilize traditional similarity measurement algorithms, such as cosine distance or derivatives of these. However, the most challenging problem facing these approaches is to quantify the non-numerical attributes of items. This is quite intractable and cannot be solved with regular similarity measurement algorithms. This paper proposes two novel methods, the Taxonomic Trees Similarity Measurement (TTSM) and the Decomposed Structures Similarity Measurement (SDSM), so that the similarities between the textual attributes can be measured using numeric values after they have been quantified. Also, the quantifying process is completely based on the semantic meanings of the textual terms. Furthermore, a maximized term matching (MTM) mechanism is induced and applied to the group-based textual attributes of items in recommender systems. Finally, we evaluate our methods by implementing a recipe recommender system which achieves a 74.4% overall satisfaction rate as evaluated by real users.