Learning-to-Rank: A New Web Ranking Algorithm using Artificial Neural Network

DOI:10.21742/ijhit.2653-309X.2021.1.1.02

AUTHORS

Falah Al-akashi,Faculty of Engineering, University of Kufa, Najaf, Iraq

ABSTRACT

The objective of this manuscript is to present a neural network algorithm for boosting the relevancy rank of 50000 documents (1000 results for each query) retrieved by our traditional algorithm that ranks a top on the list of algorithms participated in the National Institute of Standards and Technology (NIST). The assessors in NIST explore and evaluate web retrieval technological approaches over a large collection of Web data. Considering the deficiency of current learning to rank approaches lacks the continual learning ability, we introduce a new lifelong learning model that combines web search items with artificial neural networks. Working with the metaphor of our neural network algorithm, each node represents a search item while items potentially learned through observing other items towards optimized communications in the learning environment. Once all items built the relevancy ranks at the end of the iteration, top items would make their decision to declare the relevancy paths for moving to the next iteration. At that end, items would discover some ranking decisions in the available paths as relevant or not. The algorithm integrates unbiased relevance ranks while provides an explicit controller that balances the selection of documents to maximize the marginal relevance in top k results. Theoretical and empirical analysis showed that, with small compromises on long list ranking, our proposed method achieves superior relevancy and efficiency compared with the state-of-the-art algorithms meanwhile the rate of ranking is coupled to the size of the learning environment.