Recommender System Based on Purely Probabilistic Model from Pooled Sequence Statistics

Discovery Challenge Workshop 2011.

工业技术;计算机科学

Javier A. Kreiner ; Eitan Abraham

Others  :  http://ceur-ws.org/Vol-770/paper2.pdf
PID  :  42380

来源: CEUR

【 摘 要 】

In this paper we present a method to obtain a recommendation ranking for items in a collection using a marginalization technique to estimate conditional probabilities. The method uses no content-related information and rests on a probabilistic model based on implicitly collected data from past user behaviour. Given a query triplet of items for which a list of recommended results is required, the technique uses estimates for the conditional probabilities of items appearing after the three doublets defined by the triplet. The technique leads to the evaluation of a score function which takes the simple form of a sum of these conditional probabilities. Results show that the approach has good performance with respect to other methods.