[論文レビュー] Deep Learning for Sequential Recommendation: Algorithms, Influential Factors, and Evaluations

Hui Fang, Danning Zhang+2 more|arXiv (Cornell University)|Apr 30, 2019

Recommender Systems and Techniques参考文献 146被引用数 78

ひとこと要約

DLベースのシーケンシャルリコメンダーシステムの総合的な調査で、3タイプの行動シーケンス分類法、影響要因、および設計と実践を導く実証的評価を導入する。

ABSTRACT

In the field of sequential recommendation, deep learning (DL)-based methods have received a lot of attention in the past few years and surpassed traditional models such as Markov chain-based and factorization-based ones. However, there is little systematic study on DL-based methods, especially regarding to how to design an effective DL model for sequential recommendation. In this view, this survey focuses on DL-based sequential recommender systems by taking the aforementioned issues into consideration. Specifically,we illustrate the concept of sequential recommendation, propose a categorization of existing algorithms in terms of three types of behavioral sequence, summarize the key factors affecting the performance of DL-based models, and conduct corresponding evaluations to demonstrate the effects of these factors. We conclude this survey by systematically outlining future directions and challenges in this field.

研究の動機と目的

• Sequential recommendation に用いられる DL 技術の包括的概要を提供する。
• 三つの行動シーケンスタイプからなる sequential recommendation タスクの分類フレームワークを提案する。
• DL ベースのシーケンショナルリコメンダが影響を受ける要因を特定・分析し、実験でその効果を検証する。
• DL ベースのアプローチと従来の手法を比較し、実践的な指針と未解決課題を概説する。

提案手法

• 経験ベース、取引ベース、相互作用ベースの推奨タスクの3カテゴリにわたる DL ベースのシーケンシャルリコメンダーシステムの調査と分類法。
• DL 手法（RNN、CNN、MLP、注意機構、GNN）と、それらがシーケンスデータにどのように適用されるかのレビュー。
• 実データセット上での実証的評価を通じて、影響要因が性能に与える影響を示す。
• 文献と実験的知見に基づく未解決の課題と今後の研究方向の総合化。

実験結果

リサーチクエスチョン

• RQ1 What DL techniques are used for sequential recommendation and how can tasks be categorized by behavior sequences?
• RQ2 What factors influence the performance of DL-based sequential recommenders and how do they affect accuracy?
• RQ3 How do DL-based approaches compare to traditional methods for sequential recommendation in practice?
• RQ4 What are the key open issues and promising directions for future research in DL-based sequential recommendation.

主な発見

• DL-based sequential recommender models achieve state-of-the-art performance compared to traditional approaches like Markov chains and MF-based methods.
• A long sequence can be leveraged to learn the overall theme of user behavior, improving robustness on sparse data and handling varied input lengths.
• Different DL techniques (RNNs, CNNs, MLPs, attention, GNNs) offer complementary strengths for capturing intra- and inter-session dynamics.
• Multi-behavior and multi-task learning can enhance modeling of user intents expressed via different behavior types.
• There are important challenges in explainability and training efficiency, motivating future work on interpretable and scalable models.