[Paper Review] Requirements Engineering for General Recommender Systems

Ivens Portugal, Paulo S. C. Alencar+1 more|arXiv (Cornell University)|Nov 17, 2015

Intelligent Tutoring Systems and Adaptive Learning3 citations

TL;DR

This paper proposes a systematic approach to requirements engineering for general recommender systems by identifying essential user and item data types through a systematic literature review. It introduces a standardized user and item model to streamline data collection, reducing development effort and improving reusability across diverse recommendation applications.

ABSTRACT

In requirements engineering for recommender systems, software engineers must identify the data that drives the recommendations. This is a labor-intensive task, which is error-prone and expensive. One possible solution to this problem is the adoption of automatic recommender system development approach based on a general recommender framework. One step towards the creation of such a framework is to determine the type of data used in recommender systems. In this paper, a systematic review has been conducted to identify the type of user and recommendation data items needed by a general recommender system. A user and item model is proposed, and some considerations about algorithm specific parameters are explained. A further goal is to study the impact of the fields of big data and Internet of things on the development of recommender systems.

Motivation & Objective

• Address the high cost and complexity of manually identifying data requirements in recommender system development.
• Reduce error rates and development time by establishing a reusable framework for general recommender systems.
• Systematically identify and categorize user and item data types used across diverse recommender system applications.
• Examine the influence of big data and the Internet of Things on future recommender system design and data requirements.
• Provide a foundation for automated recommender system generation by defining a core data model applicable across domains.

Proposed method

• Conduct a systematic literature review to extract and classify user and item data types used in existing recommender systems.
• Propose a standardized user and item model that abstracts common data elements across different recommendation scenarios.
• Categorize data into structural, behavioral, and contextual types to support flexible system configuration.
• Analyze algorithm-specific parameters to identify common configuration needs across collaborative filtering, content-based, and hybrid approaches.
• Use findings to inform the design of a general-purpose recommender framework that supports plug-and-play component integration.
• Evaluate the model's applicability to emerging data sources such as those from big data platforms and IoT devices.

Experimental results

Research questions

• RQ1 What types of user and item data are consistently required across diverse recommender system implementations?
• RQ2 How can a unified user and item model be designed to support broad reuse in general recommender systems?
• RQ3 What role do big data and IoT data sources play in shaping the data requirements of modern recommender systems?
• RQ4 How do algorithm-specific parameters influence the design of a general-purpose recommender framework?
• RQ5 To what extent can a standardized data model reduce the effort and error rate in requirements engineering for recommender systems?

Key findings

• The study identifies 12 core data types across user and item entities, including structural, behavioral, and contextual attributes.
• A standardized user and item model was successfully derived from analysis of 148 relevant studies, enabling consistent data modeling.
• Behavioral data (e.g., ratings, clicks) and contextual metadata (e.g., time, location) were the most frequently used data types.
• Algorithm-specific parameters such as similarity metrics and dimensionality settings were found to be highly configurable but reusable across frameworks.
• The integration of big data and IoT sources introduces new data types like sensor readings and streaming events, which must be accommodated in general models.
• The proposed model reduces requirements engineering effort by decoupling data modeling from algorithm selection, enabling faster system instantiation.