A Case-Based Recommendation Approach for Market Basket Data

Abstract

In recent years, Recommender Systems have become an important part of various applications, supporting both customers and providers in their decision making processes.However, these systems still have to overcome limitations that reduce their performance, like recommendations' overspecialization, cold start and difficulties when items with unequal probability distribution appear or recommendations for sets of items are asked. A novel approach, addressing the above issues through a case-based recommendation methodology, is presented in this article. The scope of our approach is to generate meaningful recommendations based on items' co-occurring patterns and to provide more insight into customer buying habits. In contrast to current recommendation techniques that recommend items based on users' ratings or history, and to most case-based item recommenders that evaluate items' similarities, the implemented recommender uses a hierarchical model for the items and searches for similar sets of items, in order to recommend those that are most likely to satisfy a user.

A New Sentiment Case-Based Recommender

“…Additionally, content-based RSs still suffer from the recommendations' limited diversity and overspecialization problems, which limit the items recommended to users only to similar items that were previously rated. Thus, users cannot find something unexpected [3], [7], [9].…”

“…This knowledge may be reused when required without applying the entire procedure from scratch, or when highlighting a procedure that should be eliminated in a similar problem. Therefore, CBR is able to expand the problem-solving performance over time [3].…”

“…Case-based RSs follow the general CBR cycle and depend on the fundamental CBR concepts of similarity and retrieval [3]. Furthermore, there are noticeable matches between the retrieval in CBR and the handling of the user query in RSs; the user query works as a new problem specification, the item descriptions are cases in the case base, and the recommended items are retrieved based on their similarity to the user's request.…”

“…Then a case can be defined as ordered pairs c = (p, s), where p ∈ P, s ∈ S , and C is the case base in a CBR represented by a set of recorded cases C = (P, S ). Then, CBR retrieves cases with similar descriptions and adapts their solutions according to the requirements of the target problem [3], [13]. In product recommendation, the system's task is to find a product description that suits the demand of a customer as much as possible.…”

“…Most of these techniques show inadequate performance when new users or infrequent items appear (coldstart problem) whereas others may tend to overspecialize the resulting items. Furthermore, as they tend to ignore the fundamental structure of users' first choice, and do not evaluate the hidden concepts under which items are selected, their accuracy drops when the item has unequal probability distributions [1]- [3]. A better approach, which addresses the above problems, is through a case-based recommendation methodology.…”