Recommender systems are designed to recommend items to users based on their interests. Considering that in real-world scenarios user feedback is generated continuously at unpredictable rate, it becomes desirable to design models that learn from data streams at least as fast as data arrives, and are also able to recommend items based on up-to-date information. However, successful recommendation algorithms are not designed to adapt to continuous flow of data, raising scalability issues. In this work, we propose IGSIπ^t\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\text {IGSI}_{{\hat{\pi }}^t}$$\end{document}, a model designed to operate in data streams that continuously incorporates user feedback in an incremental item-graph of sequential user interactions using implicit feedback from a data stream, with the assumption that user behavior can be extracted from such sequence of interactions as time passes. IGSIπ^t\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\text {IGSI}_{{\hat{\pi }}^t}$$\end{document} recommends items based on simulations of short random walks in order to allow the generation of scalable recommendations in data stream settings. We evaluated the proposed model by recommending items with different strategies, compared the results with several incremental algorithms using a prequential approach, and demonstrate that our method obtains superior accuracy with competitive update and recommendation times.
