Cheng-Jye Luh, Sheng-An Yang and Ting-Li Dean Huang
– The purpose of this paper is to estimate Google search engine’s ranking function from a search engine optimization (SEO) perspective.
Abstract
Purpose
The purpose of this paper is to estimate Google search engine’s ranking function from a search engine optimization (SEO) perspective.
Design/methodology/approach
The paper proposed an estimation function that defines the query match score of a search result as the weighted sum of scores from a limited set of factors. The search results for a query are re-ranked according to the query match scores. The effectiveness was measured by comparing the new ranks with the original ranks of search results.
Findings
The proposed method achieved the best SEO effectiveness when using the top 20 search results for a query. The empirical results reveal that PageRank (PR) is the dominant factor in Google ranking function. The title follows as the second most important, and the snippet and the URL have roughly equal importance with variations among queries.
Research limitations/implications
This study considered a limited set of ranking factors. The empirical results reveal that SEO effectiveness can be assessed by a simple estimation of ranking function even when the ranks of the new and original result sets are quite dissimilar.
Practical implications
The findings indicate that web marketers should pay particular attention to a webpage’s PR, and then place the keyword in URL, the page title, and snippet.
Originality/value
There have been ongoing concerns about how to formulate a simple strategy that can help a website get ranked higher in search engines. This study provides web marketers much needed empirical evidence about a simple way to foresee the ranking success of an SEO effort.
Details
Keywords
Lin‐Chih Chen and Cheng‐Jye Luh
This study aims to present a new web page recommendation system that can help users to reduce navigational time on the internet.
Abstract
Purpose
This study aims to present a new web page recommendation system that can help users to reduce navigational time on the internet.
Design/methodology/approach
The proposed design is based on the primacy effect of browsing behavior, that users prefer top ranking items in search results. This approach is intuitive and requires no training data at all.
Findings
A user study showed that users are more satisfied with the proposed search methods than with general search engines using hot keywords. Moreover, two performance measures confirmed that the proposed search methods out‐perform other metasearch and search engines.
Research limitations/implications
The research has limitations and future work is planned along several directions. First, the search methods implemented are primarily based on the keyword match between the contents of web pages and the user query items. Using the semantic web to recommend concepts and items relevant to the user query might be very helpful in finding the exact contents that users want, particularly when the users do not have enough knowledge about the domains in which they are searching. Second, offering a mechanism that groups search results to improve the way search results are segmented and displayed also assists users to locate the contents they need. Finally, more user feedback is needed to fine‐tune the search parameters including α and β to improve the performance.
Practical implications
The proposed model can be used to improve the search performance of any search engine.
Originality/value
First, compared with the democratic voting procedure used by metasearch engines, search engine vector voting (SVV) enables a specific combination of search parameters, denoted as α and β, to be applied to a voted search engine, so that users can either narrow or expand their search results to meet their search preferences. Second, unlike page quality analysis, the hyperlink prediction (HLP) determines qualified pages by simply measuring their user behavior function (UBF) values, and thus takes less computing power. Finally, the advantages of HLP over statistical analysis are that it does not need training data, and it can target both multi‐site and site‐specific analysis.