Lecture-E8

Lecture-E8 - Lecture E A Study of Heterogeneity in...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon
Lecture E A Study of Heterogeneity in Recommendations for a Social Music Service A. Bellogin, I. Cantador, and P. Castells Proc. of HetRec, pp. 1-8, 2010
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Research Questions Which sources of information available in Web 2.0 are more valuable for recommendation? Exploiting different sources of information, i.e., ratings, tags, social contacts/communication, for recommenders by studying several performance metrics Do recommenders exploiting different resources in Web 2.0 really offer heterogeneous item suggestions from which hybrid strategies could benefit? Employing non-performance metrics , such as novelty , that measure item recommendation characteristics to conduct the study 2
Background image of page 2
The Study Study several approaches for recommendation using a heterogeneous dataset extracted from Last.FM Content-based Collaborative filtering Social recommenders Compare the performance of the recommenders with precision , recall , and ranking metrics, along with other novel metrics that measure Coverage Novelty Overlap Diversity 3
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Evaluated Recommenders Given U = { u 1 , …, u M } be a set of users I = { i 1 , …, i N } be a set of items g : U × I → R, where R is a totally ordered set such that g ( u m , i n ) measures the gain of usefulness of item i n to user u m The goal is to identify items i max, u ( I ) unknown to the user that maximize g 4
Background image of page 4
Content-Based Recommenders (CB) CB recommenders – create, tag, and share items Are based on user and item profiles defined in terms of vectors of weighted tags Compute similarities between vectors to yield personal recommendations Given a folksonomy F = { T, U, I, S }, where T = { t 1 , …, t L } is the set of tags that comprise the vocabulary expressed by the folksonomy U and I are the set of users and items , respectively that annotate and are annotated with tags in T S = { ( u m , t i , i n ) U × T × I } is the set of annontations of each tag t i to an item i n by user u m 5
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
CB Recommenders Profile of user u m as a vector u m = { u m, 1 , …, u m,L } u m , l (1 l L ), a weight that measures the informativeness of tag t l to characterize contents annotated by u m Profile of item i n as a vector i n = { i n, 1 , …, i n,L } i n , l (1 l L ), is a weight that measures the “ relevance ” of tag t l to describe i n Several alternatives exist to weight the components of tag-based user and item profiles based on information available in an individual profile or folksonomy 6
Background image of page 6
Image of page 7
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 22

Lecture-E8 - Lecture E A Study of Heterogeneity in...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online