An empirical study of the structure of relevant keywords in a search engine using the minimum spanning tree

Abstract

This paper provides a comprehensive study of the structure of relevant keywords in a search engine using the minimum spanning tree (MST) approach. In the process of constructing MST’s, we introduce a novel metric to measure a distance between keywords by applying an integration of the Pearson correlation and the query-based cosine similarity. From this work, we made several meaningful observations about the networks of relevant keywords. First, keyword networks in a search engine exhibit the small-world effect and the scale-free property. Second, only a few among relevant keywords in the order of popularity are positively correlated and there is no significantly positive or negative relationship for the rest of relevant keywords. Third, the degree of searching activity for relevant keywords varies depending on whether they are branded keywords or non-branded keywords as well as the characteristics of product categories. Fourth, the mean correlation coefficient for keyword impressions during slow season increases. Finally, both k_max and the betweenness centrality for high-involvement products are higher than those for low-involvement products.

Introduction

Online advertising is a form of promotion that is a key component of marketing strategies to increase sales in online marketplaces as well as search engines. It delivers marketing messages to attract consumers. A successful form of online advertising is the sponsored search (Wu et al., 2009). It is a very effective medium for advertising as it allows precise targeting of advertisements to users (Yang, Chen, & Chen, 2007). The sponsored search is the “pay-for-placement” service that displays text-only ads alongside search engine results. Billions of dollars are spent each year on sponsored search. According to a study by eMarketer, Internet advertising spend is expected to grow from $16.4 billion in 2006 to $36.5 billion in 2011 and 40% of this ad spend occurs on sponsored search (eMarketer). Sponsored search or Search Engine Marketing (SEM) is a form of advertising on the Internet where advertisers pay to appear alongside organic search results (Abhishek & Hosanagar, 2007). In sponsored search advertising, advertisers have to decide how to spread their budget across target keywords. Since advertisers typically have a fixed daily budget that should not exceeded, an advertiser must allocate the budget as productively as possible by selecting which keywords to use and then deciding how much to allocate for each keyword (Ozluk & Cholette, 2007). There have been many studies on the allocation of an advertising budget to maximize product sales (Borgs et al., 2007, Freuchter and Dou, 2005, Holthausen and Assmus, 1982). They have mainly focused on optimal budget allocation among a large set of heterogeneous keywords. Meanwhile, it is often difficult to even select a set of keywords from the set of all possible keywords appearing in the list of keywords someone typed in order to find a product. So, advertisers usually select keywords that are relevant to their products and then purchase those keywords from paid listing providers (Faber, Lee, & Nan, 2004). In this respect, top relevant keywords in a product category can be used for the purpose of online advertising. In fact, major search engines including Google automatically display a list of relevant keywords based on the same characters being typed when a user types a keyword query into the search engine. Furthermore, Abhishek and Hosanagar (2007) suggested a method to generate keywords for search engine advertising using semantic similarity between terms. The semantic similarity of the proposed model is obtained by computing the TFIDF (Term Frequency – Inverse Document Frequency) from the retrieved documents associated with a query word. However, advertising search keywords generated from a seed keyword do not guarantee the profitability of advertising which was the original object of their study because users may have different preferences on keyword search behavior so that they would choose a different seed keyword.

In this paper, we apply network theory to the analysis of relevant keywords in sponsored search because a network topology provides efficient ways of understanding its structural properties. Several attempts have been made to apply network theory for finding relevant keywords. For instance, Liu, Weichselbraun, Scharl, and Chang (2005) adopted the augmented semantic network for identifying the network’s most relevant keywords. However, there has been no research on the construction of relevant keywords by linking search keywords based on the minimum spanning tree (MST). We examine the network properties of the search keywords and interpret the network topology in marketing terms. The remainder of the paper is organized as follows. In Section 2, we discuss the previous works about constructing keyword correlation graphs. Section 3 introduces methodologies that are able to construct a network of search keywords based on their daily impressions and web logs identifying interactions with the rest of keywords. Section 4 presents some statistics and empirical results extracted from the MSTs of relevant keywords. Finally, we describe conclusion and future work of the study in Section 5.