A Survey on Session-based Recommender Systems (1902.04864v3)

Abstract: Recommender systems (RSs) have been playing an increasingly important role for informed consumption, services, and decision-making in the overloaded information era and digitized economy. In recent years, session-based recommender systems (SBRSs) have emerged as a new paradigm of RSs. Different from other RSs such as content-based RSs and collaborative filtering-based RSs which usually model long-term yet static user preferences, SBRSs aim to capture short-term but dynamic user preferences to provide more timely and accurate recommendations sensitive to the evolution of their session contexts. Although SBRSs have been intensively studied, neither unified problem statements for SBRSs nor in-depth elaboration of SBRS characteristics and challenges are available. It is also unclear to what extent SBRS challenges have been addressed and what the overall research landscape of SBRSs is. This comprehensive review of SBRSs addresses the above aspects by exploring in depth the SBRS entities (e.g., sessions), behaviours (e.g., users' clicks on items) and their properties (e.g., session length). We propose a general problem statement of SBRSs, summarize the diversified data characteristics and challenges of SBRSs, and define a taxonomy to categorize the representative SBRS research. Finally, we discuss new research opportunities in this exciting and vibrant area.

• The paper proposes a unified framework categorizing session-based recommendation scenarios and formalizes the session modeling problem.
• It analyzes session characteristics and challenges such as data sparsity and dependency capture in various recommender approaches.
• The survey compares conventional, latent, and deep learning methods while outlining promising directions for agile, context-aware recommendations.

A Comprehensive Review of Session-based Recommender Systems

The paper presents a detailed survey of Session-based Recommender Systems (SBRSs), which have gained traction as an innovative form of recommender systems (RSs) focusing on capturing short-term, dynamic user preferences. The distinction of SBRSs lies in their objective to provide more timely recommendations by leveraging session contexts rather than relying on static and long-term user preferences as traditional recommender systems do.

Core Contributions

1. Unified Framework and Problem Statement: The authors propose a comprehensive framework for SBRSs, categorizing existing research into three main scenarios - next interaction recommendation, next partial-session recommendation, and next session recommendation. This classification effort aims to clarify the often ambiguous distinction between session-based and sequential recommender systems (SRSs). Additionally, they present a definitive problem statement that formally encapsulates the diverse structures and characteristics of session data.
2. Characteristics and Challenges of SBRSs: The paper explores various session characteristics such as session length, internal order, user information, action types, and session-data structure. It offers an in-depth analysis of the challenges these characteristics pose in terms of session modeling, capturing dependencies, and handling data sparsity.
3. Taxonomy of SBRS Approaches: The authors classify SBRS approaches into three main categories: conventional approaches, latent representation approaches, and deep neural network approaches. They discuss in detail methodologies ranging from pattern and rule mining to neural network architectures, elaborating on the types of dependencies these methods aim to capture. Notably, the paper contrasts the strengths and deployment scenarios of each approach, particularly focusing on their learned dependencies, which are essential for effective recommendation.
4. Comprehensive Comparison and Analysis: Highlighted are comparisons that delineate the applicability of each approach, offering insights into their strengths and limitations. The paper critically examines the suitability of various modeling techniques in SBRSs for different session characteristics, offering empirical evidence and theoretical reasoning for their placement.
5. Prospective Research Directions: In acknowledging the dynamic nature of user behavior and session characteristics, the paper identifies future challenges such as incorporating heterogeneous contextual factors, leveraging cross-domain information, and addressing system constraints. These open issues suggest avenues where SBRS research could further diversify and improve, particularly as these systems become increasingly pivotal in complex, real-world applications.

Implications and Future Directions

The paper's thorough exploration of SBRSs contributes significantly to understanding how session-based approaches can be leveraged across different domains, such as e-commerce and media. Considering both theoretical implications and practical applications, SBRSs arguably address some limitations of conventional RSs by offering more agile and context-sensitive recommendation engines.

Future research might delve into enhancing SBRSs' capacity for dynamic contextual and cross-domain factor incorporation, potentially by employing advanced machine learning techniques such as reinforcement learning and generative modeling. With the rapid evolution in AI methodologies and systems architectures, the pathway outlined by the authors underscores a vibrant future for SBRSs in generating meaningful, user-centric recommendations.

In conclusion, this survey presents a foundational understanding of SBRSs. It not only consolidates current knowledge and methodologies but also sets the stage for future advancements, guiding researchers to address persisting challenges and explore new possibilities in session-based recommendation.