Inductive Building of Search Results Ranking Models to Enhance the Relevance of Text Information Retrieval

Zosimov, Viacheslav; Stepashko, Volodymyr; Bulgakova, Oleksandra

doi:10.1109/dexa.2015.70

Cited by 9 publications

(5 citation statements)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…At first glance, this task seems very laborious, considering the large number of results, which in most cases reaches hundreds of thousands, or even several million. However, upon closer analysis of the algorithms used by search engines, it turns out that the computational complexity of the task is several orders of magnitude lower [14]. The actual number of search results that a user has access to be much smaller than what the system claims when performing a search query.…”

Section: Resultsmentioning

confidence: 99%

Approach to Search Result Rankings Based on User Ratings

Zosimov,

Bulgakova

2023

Computer Modeling and Intelligent Systems

View full text Add to dashboard Cite

This paper proposed an approach to rank search results based on user ratings, with a subjective approach to the ranking process. Expert groups unique to each user are created to achieve this effect. When no common ratings are available, the expert group is formed based on a model built on the user's social profile, utilizing inductive algorithms such as a polynomial neural network with active neurons. The approach of ranking search results provides a unique order of elements for each user's list of web resources. This effect is achieved by using evaluations from an expert group that is unique to each user, as well as by incorporating each evaluation into the model for calculating final rankings of web resources with its unique weight, calculated based on their previous activity in the system. The resulting ranking model is much more difficult to falsify because it is based on subjective factors. The ranking model will be unique for each user. Falsifying it would require reproducing the preferences of each user, which is much more difficult than acquiring links to one's web resource from authoritative sources.

show abstract

Section: Resultsmentioning

confidence: 99%

Approach to Search Result Rankings Based on User Ratings

Zosimov,

Bulgakova

2023

Computer Modeling and Intelligent Systems

View full text Add to dashboard Cite

show abstract

“…MR has three potential values: 1 -classical multilayered, 2 -relaxative, 3 -combined algorithms [23]. In the case where DM is set to 1, we encounter three standard variations of iterative algorithms: MIA=GIA(1,1,1), RIA=GIA(1,1,2), CIA=GIA(1,1,3), as well as three iterative-combinatorial variants: MICA=GIA(1,2,1); RICA=GIA(1,2,2), CICA=GIA (1,2,3).…”

Section: Methodsmentioning

confidence: 99%

“…Search engine ranking algorithms take into account a large number of factors, but the main weight is given to the page rating, which is calculated based on the analysis of the number and quality of external links to the page [3]. Such assessment methods are objective, but they are easy to falsify in the presence of a certain advertising budget, due to the purchase of the necessary number of quality links from external sources [4].…”

Section: Abbreviationsmentioning

confidence: 99%

See 1 more Smart Citation

User Evaluation-Driven Ranking Concept

Zosimov,

Bulgakova,

Perederyi

2023

RIC

View full text Add to dashboard Cite

Context. The problem of personalizing search engine results, empowering users with search result management tools and developing new ranking models based on user’s subjective information needs. The object of the study was to modeling information search results in the Internet based on user ratings. Objective. The goal of the work is to form unique expert groups for each user, based on calculating the measure of agreement between the current user’s opinions and potential experts. Method. Introducing a novel method for ranking search results based on user ratings, which takes a subjective approach to the ranking process. This approach involves the formation of distinct expert groups tailored to individual users. Experts are selected based on the level of agreement between their opinions and the current user, determined by shared ratings on a specific set of web resources. User selection for the expert group is based on their weight relative to the current user, serving as a measure of agreement. The proposed methodology offers a fresh approach to forming unique expert groups for each user, utilizing three different strategies depending on the presence of shared ratings on a particular set of web resources between the user and potential experts. The developed ranking method ensures that each user receives a personalized list of web resources with a distinct order. This is accomplished by incorporating unique ratings from the expert group members associated with each user. Furthermore, each rating contributes to the ranking model of web resources with an individual weight, calculated based on an analysis of their past system activity. Results. The developed methods have been implemented in software and investigated for complex web data operation in real time. Conclusions. The conducted experiments have confirmed the effectiveness of the proposed software and recommend its practical use for solving complex web data operation in real time. Prospects for further research may include optimizing software implementations and conducting experimental investigations of the proposed methods on more complex practical tasks of various nature and dimensions

show abstract

“…This algorithm from the previous is different in that each pair of pairs are formed only from intermediate arguments and from intermediate and initial [22].…”

Section: Combined Iterative Algorithmmentioning

confidence: 99%