Making the Most of Preference Feedback by Modeling Feature Dependencies

Exploiting the Interplay among Products for Efficient Recommendations

“…We evaluate our methods with the standard evaluation procedure followed in CCBR-RSs literature [8]. It is widely used in CCBR-RS literature ( [8], [9], [13], [1], [14]). We take three real-world datasets namely Camera, PC and Used Cars , each of which has 210, 120 and 956 cases respectively.…”

“…The feature weights model the preferences of the user and thus helps with the personalisation of the recommendations. The work by authors in [13] points out the drawback with learning feature weights independent of the other features in the domain. For the example considered above, one might have chosen 'Air Cooled' engine because the motorcycle is the cheapest of the lot.…”

“…Assigning higher weights to 'Air Cooled' engine without considering the cost would not reflect the preferences of the user in an adequate manner. In [13], the authors make an assumption that the utility of PBF is greater than or equal to the utility of each of the rejected products. They use a weighted MAUT based similarity model as given in the equation below to model the utility of a product, where w i is the weight given to feature i.…”

“…The idea of profiling the products even before the recommendation could start is proposed by authors in [1]. They employ a method similar to the one used by authors in [13] to characterize each product in the domain. Their work argues that each product in the domain has prospective buyers.…”

“…Similar to the work by the authors in [13], inequalities are formed and solved for the feature weights. The solution to the inequalities is a region in the feature weights space which the authors term as dominance region of the product.…”