Adaptation and Evaluation of Recommendations for Short-term Shopping Goals

Abstract

An essential characteristic in many e-commerce settings is that website visitors can have very specific short-term shopping goals when they browse the site. Relying solely on longterm user models that are pre-trained on historical data can therefore be insufficient for a suitable next-basket recommendation. Simple "real-time" recommendation approaches based, e.g., on unpersonalized co-occurrence patterns, on the other hand do not fully exploit the available information about the user's long-term preference profile.In this work, we aim to explore and quantify the effectiveness of using and combining long-term models and shortterm adaptation strategies. We conducted an empirical evaluation based on a novel evaluation design and two real-world datasets. The results indicate that maintaining short-term content-based and recency-based profiles of the visitors can lead to significant accuracy increases. At the same time, the experiments show that the choice of the algorithm for learning the long-term preferences is particularly important at the beginning of new shopping sessions.

Evaluation of session-based recommendation algorithms

“…Generally, BPR and other methods designed for the matrix-completion problems in their original form, i.e., without considering the short-term session context, do not lead to competitive results in session-based recommendation scenarios, as reported, e.g., in [Jannach et al 2015a]. Therefore, we do not consider such algorithms, e.g., traditional matrix factorization techniques, as baselines in our experiments.…”

“…Such techniques are called session-aware according to the terminology of [Quadrana et al 2018]. Examples of such works include [Baeza-Yates et al 2015;Billsus et al 2000;Hariri et al 2012;Jannach et al 2017aJannach et al , 2015aQuadrana et al 2017], and session-aware approaches were applied for various application domains like e-commerce, music, news, or nextapp recommendation. Considering longer-term user preferences in these papers shows to be helpful to improve the recommendations in the current, ongoing session.…”

Using Shortlists to Support Decision Making and Improve Recommender System Performance

“…Ideally, we would like to have a recommender system in use in order to study interface effects in isolation. Since this is not the case, we follow a methodology similar to how recommender system studies are performed over logs of interaction data [10]. In short, we will try to predict a single session's chosen item based on some subset of the user's actions in this session.…”

“…This again shows the need to consider both interface design and feedback elicitation at the same time. On the algorithmic side, several approaches are designed to learn from session-based data [18,25,10]. The work done by Jannach et al [10] also adopts a session-based approach to recommendation, but, in contrast to us, assumes that a long term interest profile is available.…”

Sequence-aware Recommender Systems

“…The corresponding algorithms however typically implement no specific means to take the users' short-term behavior or intents into account in their recommendations; nor are they designed to use the rich information that is contained in the sequentially-ordered user interaction logs that are often available in practical applications.In practice, however, there are many application scenarios where considering short-term user interests and longer-term sequential patterns can be central to the success of a recommender. A typical example problem setting is that of session-based recommendation [45,56], where no longer-term user histories are available. Instead, we have to adapt the recommendations according to the assumed short-term interests of an anonymous user.…”