Reviewer 4 (chair) Expertise Expert The Meta-Review Based on all three reviewers overall positive evaluation, I recommend the paper for publication. Final Recommendation after Meta-Review Probably accept: I would argue for accepting this paper. Reviewer 1 (PC) Review Rating Probably accept: I would argue for accepting this paper. Expertise Passing Knowledge Contribution The authors present a model called TransFm, which combines translation and metric based approaches for sequential recommendations with Factorization Machines(FM). So they adapt the main ideas from FM into translation-based sequential recommenders. In each featured dimension the inner product of the learned embedding and translation space is replaced with the squared euclidean distance. Furthermore the authors expose that the model equation can be computed in linear time and can be optimized with tradition FM features. In order to expose the advantages of their system, the authors started comprehensive experiments and evaluated TransFm on several Data-sets compared to other related baseline models taking account of additional features or not. Furthermore they present a generalization of the approach and compare them to other related models by merging FMs with similar baseline models. Relevance to RecSys Since the authors consider sequence based recommendations of items to users over time the topics of 'User modelling', 'Personalisation' and 'Evaluation metrics and studies' are hit. Your Review Positives: The motivation and the intoruction of this contribution are well verbalized and presented in an appealing manner. I see a golden thread through the whole paper and especially the experiments are performed in a conscientious way. Negatives: This approach is predestinated to include more vivid examples. So parts of the introduction are a bit complex to understand. Suggested Revisions: I think it is necessary to evaluate the quality of the proposed model with a user study for future work(!). Missing Related Work --- Typos/Formatting --- Reviewer 2 (PC) Review Rating Probably accept: I would argue for accepting this paper. Expertise Knowledgeable Contribution This paper describes a new sequential recommendation algorithm, called TransFM, which combines translation and metric-based approaches for sequential recommendation with factorization machines. The authors' idea is to incorporate in FM model the distance and translation components of the TransRec model. In this way it is possible to exploit the FMs ability to incorporate with simplicity different types of features and models the interactions between all the them. Then the translation component allows to represents the user interaction sequence. Finally the Euclidean distance, used to replace the inner product term in FM, is able to improve the generalization performance. The results show that the authors' propose wins against all the baselines on a variety of datasets. Besides the benefit of new proposals, I find this paper very useful for learning about sequential RSs, FMs and other state of the art baselines Relevance to RecSys I really think this work is in the scope of RecSys topics, proposing a new sequential recommendation algorithm Your Review * Positives This paper contributes with a new and very sound sequential recommender algorithm and a comprehensive and sound experimental analysis. I find it well-written, clear and very formative, even for people with no strong knowledge in RS, factorization machines and sequential RS. * Negatives The only think that concerns me is the use of only one evaluation metric (AUC) and that's why I propose to "probably accept" the paper. Can authors justify this choice? Can't other metrics be used for sequential RS evaluation? * Suggested Revisions I find noteworthy that HRMavg wins over most of the baselines on the Google Local datasets, since it is one of the simpler methods. Even though TransFMcontent always beats it, I wonder if those improvements (sometimes small) pay off the complexity of TransFMcontent. Maybe authors could try to answer that question in further researches, in not in this paper. * Typos/Formatting issues in 4.6 "To analyze the sensitivity of TransFM the parameter dimensionality., We adjust" Reviewer 3 (PC) Review Rating Probably accept: I would argue for accepting this paper. Expertise Knowledgeable Contribution This work proposes an approach to sequential recommendation using factorisation machines. The approach builds on past work and combines the distance and translation components of the earlier TransRec model with the ability to readily add content-based features to enhance performance. Relevance to RecSys The work is clearly very relevant to RecSys. Your Review This is a very well written and structured paper and builds nicely on past work. The proposed model is technically sound. The general nature of the proposed framework, which can incorporate arbitrary features (e.g. temporal, spatial, demographic etc.) without necessitating changes to the model, is novel and has merit from both a research and applied perspective. The literature review is thorough and informative. A particular strength of the paper is the comprehensive nature of the evaluation. In particular, a number of publicly available datasets from diverse domains are considered. Moreover, comparisons to relevant baselines are provided, and the results show that the proposed approach outperforms other techniques on the datasets evaluated. The paper also considers the combination of factorisation machines with other baselines - these approaches are also found to provide good performance and thereby further validate the intuition behind the proposed approach. Overall, this work is very interesting, it addresses an important problem, makes a significant contribution to the state of the art, and is very relevant to the conference.