----------------------- REVIEW 1 ---------------------
PAPER: 149
TITLE: Vista: A Visually, Socially, and Temporally-aware Model for Artistic Recommendation
AUTHORS: Ruining He, Chen Fang, Zhaowen Wang and Julian McAuley

OVERALL EVALUATION: 2 (accept)
REVIEWER'S CONFIDENCE: 4 (high)
Relevance for RecSys: 5 (excellent)
Novelty: 3 (fair)
Technical quality: 4 (good)
Significance: 3 (fair)
Presentation and readability: 4 (good)

----------- Review -----------
This paper presents ideas in recommendation of art images based on users' historical interactions with art images uploaded by other users. The main contribution of the paper seems to be the proposal of a recommendation model based on Factorized Personalized Markov Chains, which incorporates rich content, temporal dynamics and social dynamics into recommendation context, and Bayesian Personalized Ranking to "capture individual users' preferences towards particular visual art styles".

Strong points of the paper are: 1) a proposal of an interesting yet seemingly complex problem of modeling artistic preference; 2) an appropriate modification and combination of state-of-the-art preference modeling method and state-of-the-art sequence optimization method; 3) good presentation of experimental results that show superiority of the proposed method (especially the huge improvement in cold item recommendation).

However the paper also suffers from notable problems: 1) the paper is well presented and written overall, however, the exact theoretical contribution compared to previous works cannot be identified easily; 2) the authors imply that the consideration of latent affinity between a user and other item creators is sufficient for incorporating social dynamics into the model; however, 'social dynamics' is more broadly understood as dynamics in actual social relationships between users. It would help to differentiate between the different aspects of social context/dynamics and clarify the authors' intentions.

Overall, the paper is well written, proposes an interesting problem, and provides an appropriate solution to the problem.


----------------------- REVIEW 2 ---------------------
PAPER: 149
TITLE: Vista: A Visually, Socially, and Temporally-aware Model for Artistic Recommendation
AUTHORS: Ruining He, Chen Fang, Zhaowen Wang and Julian McAuley

OVERALL EVALUATION: 1 (weak accept)
REVIEWER'S CONFIDENCE: 2 (low)
Relevance for RecSys: 3 (fair)
Novelty: 3 (fair)
Technical quality: 3 (fair)
Significance: 3 (fair)
Presentation and readability: 3 (fair)

----------- Review -----------
The paper describes a RS that addresses one of the challenges of this class of system (dealing with large, sparse, and long-tailed datasets) and takes into account the role of "social dynamics", i.e., the influence of social factors in shaping users' preferences. These problems are tacked in domain (artistic content) and a specific case - a social art web site - where an additional element of complexity is the need of modelling content
in terms of its visual appearance.

The paper states clearly its goals and the open issues in the current state of the arts. The adopted techniques are sound as well as the evaluation process. It is hard to judge the degree at which the specific solutions can be generalized to other domain characterized by a strong influence of the "social" factor and by the complexity of content modeling. In that respect, the paper would benefit from a more self-critical analysis of results


----------------------- REVIEW 3 ---------------------
PAPER: 149
TITLE: Vista: A Visually, Socially, and Temporally-aware Model for Artistic Recommendation
AUTHORS: Ruining He, Chen Fang, Zhaowen Wang and Julian McAuley

OVERALL EVALUATION: 1 (weak accept)
REVIEWER'S CONFIDENCE: 4 (high)
Relevance for RecSys: 5 (excellent)
Novelty: 4 (good)
Technical quality: 5 (excellent)
Significance: 4 (good)
Presentation and readability: 5 (excellent)

----------- Review -----------
In this paper the authors use visually and socially-aware
Markov chains to model artistic preferences based on visual appearance and social dynamics.
In particular, the authors present a new formulation of the Markov chain factorization (EQ 1) which tries to capture both
long-term and short-term interest of the user. In Eq 2 the authors extend the first-order Markov model to a higher-order
by using a personalized decaying scheme. The evaluation of the new method is performed on a real social art website on
two common sequential prediction tasks. The results show that the new method outperforms the state-of-the-art baselines.


The paper is well-written and easy to follow. Nevertheless, the concept of "social dynamics" is still not totally clear to me. The authors should explain the actual intention when they refer to "social dynamics" - does it contain new interactions between users. In addition the authors should explain how the ideas presented in this paper can be used/generalized to other recsys domains.

All in all, the paper presents interesting challenges and a new solution. The system was also evaluated on a real dataset.