Academic journal article Journal of Electronic Commerce Research

A Conditional Feature Utilization Approach to Itemset Retrieval in Online Shopping Services

Academic journal article Journal of Electronic Commerce Research

A Conditional Feature Utilization Approach to Itemset Retrieval in Online Shopping Services

Article excerpt


Due to the increasing number of items with a variety of descriptions for a product type, itemset retrieval is considered as an essential function for enhancing shopping experiences of customers in online malls. This paper considers an itemset retrieval problem to construct an itemset consisting of items belonging to the same product type against a query item in which a customer is interested. In contrast to the previous approaches that require additional prior information such as itemset memberships and the known number of itemsets, we propose a semi-supervised itemset retrieval model that can automatically find a target itemset for a query item based on two item features, namely textual description and price. Specifically, in order to precisely identify itemsets, the proposed model conditionally utilizes price feature of an item only when its textual description feature is relevant to that of a query item. Experiment results based on two real-world datasets show that the proposed model outperformed the other alternatives.

Keywords: Itemset retrieval; Semi-supervised approach; Conditional feature utilization; Finite mixture model; e- Commerce

(ProQuest: ... denotes formulae omitted.)

1. Introduction

Owing to the recent proliferation of various types of online shopping services such as open markets, Internet auctions, and social commerce where sellers are allowed to vend their items with their own pricing strategies, there exist many items with diverse prices and various descriptions for a single product type across many online shopping malls [Ramachandran et al. 2011; Wu et al. 2011]. In Google Shopping, for instance, a product type named "Sony Micro Vault USB 16G" is sold in the forms of 65 distinct items with different prices and descriptions. While the availability of multiple items for a single product type offers a wide variety of purchase choices to a customer, it makes difficult for a customer to identify the items belonging to the same product type of interest.

To enhance the customer's shopping experience, two types of search services, namely item search and itemset search, are often provided. The item search aims to retrieve relevant items to a user query that usually consists of words or phrases while the itemset search seeks to find the set of items belonging to the same product type as that of an item found to be interesting to a customer for the purpose of price comparison. This paper focuses on the itemset retrieval problem, and we refer a given item as a query item and the set of items to be suggested as a target itemset against the query item. In Figure 1, the relationship between a product type and items as well as the relationship between a query item and its target itemset are illustrated.

Through facilitating automatic retrieval of the target itemset for a query item, customers are provided with comparison results for the items from the same product type, resulting in reduced item search costs [Tan et al. 2010]. In the meantime, the automatic itemset retrieval method is also beneficial to service providers as they are no longer required to prepare all the possible target itemsets in advance. Moreover, the target itemsets can be further utilized to improve the performances of item ranking [Kim et al. 2012b], item recoimnendation [Linden et al. 2003], and item bundling [Garfinkel et al. 2008] by allowing exploration of the relationships among items.

There have been many research results pertaining to the itemset retrieval. Yet, they have limited applicability due to the requirement of information such as the known number of itemsets [Kannan et al. 2011b; Kim et al. 2012a], the predefined hierarchical structure of itemsets [Abbott et al. 2011; Benjelloun et al. 2009], and the prior knowledge for adjusting model parameters [Kopcke et al. 2010; Wong et al. 2008].

Supervised approaches presented in [Abbott et al. 2011; Geng et al. 2012; Kim et al. 2013b] that require training data on item membership against an itemset may not be viable options for small and medium sized shopping services due to the significant amount of time and cost involved for obtaining the training data [Kannan et al. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.