The Analysis of Rank Fusion Techniques to Improve Query Relevance

Abstract

Rank fusion meta-search engine algorithms can be used to merge web search results of multiple search engines. In this paper we introduce two variants of the IntroductionThere are many proposals for a meta-search engine (MSE). Given a query (a set of keywords), typically an MSE system retrieves web pages that are relevant to the query by exploiting all its underlying search engines. It sends the query to these engines; the results obtained are then merged and ranked. It returns final web documents ranked by relevance. In the Helios architecture [1] the MSE system uses standard merger and ranker modules. To achieve high performance it utilizes async I/O and parallel TCP connections with the remote search engines. In the Tadpole architecture [2,3], the rank fusion algorithms are based on a variety of parameters, such as the rank order and the number of times an URL appears in the results of each of its search engines components, to compute a weight for each collected results [2][3][4]. There is also the concern of user specific needs. For example, an MSE should ideally let the user choose his favourite search engines from an available list, and do query modifications, as well as explore available rank fusion techniques [2].In general, rank fusion algorithms offer improvement of the relevance scores of the returned documents of multiple search engines. Dwork, Kumar, Naor, and Sivakumar propose the use of rank aggregation methods for MSEs viz. the Borda's method, Footrule and Scaled Footrule, and Markov Chain methods [5]. Lam and Leung propose a complete directed graph viz. MST Algorithm [6]. Supervised rank aggregation methods such as Borda-Fuse and supervised Markov Chain based methods are investigated in [7]. KE algorithm [8] and its variants [4] exploit the ranking on the results that an MSE receives from its component engines, by considering the number of documents appearances in the component engines' lists with equal reliability assumption of those engines. Another rank fusion MSE algorithm named Count Function [9] defines web documents ranking as summing ranks as per positions of a URL

Wiki-MetaSemantik: A Wikipedia-derived query expansion approach based on network properties

“…Thirty multi domain queries are used in this comparison. Also, since combining multiple datasets can lead into better accuracy [14], for example as a meta-search engine [18], we propose the use of meta-pseudo relevance feedback (meta-PRF) for automatic judgement purpose as in [17] [18]. Our experiments show benefit of IR performance when using Wiki-MetaSemantik.…”

“…The MSE algorithm viz. Weighted Borda Fuse (or WBF) is as in [18] [6]. As inputs we take only synonyms of user query from online thesauruses (WordNet, WikiSynonyms, Moby Thesaurus), as well as synonyms and related terms from Wikipedia by computing ontology graph properties (degree, closeness, pageRank) as in Step 7 in Section 3.…”