Face Validation of Recommendations

Sample:

• Recommendation lists for $100$ information products were randomly selected (from the generated $1827$ lists)
• For these products $59750$ co-occurrences where found
• The lists contain $1259$ outliers (recommendations)
• Each of the $1259$ recommendations was inspected for plausibility

Percentage of plausible recommendations:

• $31$ lists, significant LSD-model, $87.71$ %
• $25$ lists, LSD model not significant, $89.45$ %
• $44$ lists, no $\chi^{2}$ test, $75.74$ %

-> Even if the LSD-model is not significant, it still seems useful to identify non-random outliers

Index: Contents