ViTOR: Learning to Rank Webpages Based on Visual Features

Staff - Faculty of Informatics

Date: 15 November 2019 / 13:30 - 14:30

USI Lugano Campus, room SI-007, Informatics building (Via G. Buffi 13)

Ilya Markov, University of Amsterdam, The Netherlands

Standard learning to rank (LTR) approaches combine multiple types of features, namely content-based, link-based, and user-based. Recently, visual-based features have been introduced in LTR. These features are extracted from visual snapshots of entire webpages using neural networks. In this talk, I will introduce the Visual learning TO Rank (ViTOR) model that integrates state-of-the-art visual features extraction methods: (i) transfer learning from a pre-trained image classification model, and (ii) synthetic saliency heatmaps generated from webpage snapshots. I will also discuss the ViTOR dataset that was built to evaluate visual-based features for LTR and show that our proposed features improve LTR performance. Our paper paper on this topic won the best short paper award at WWW 2019 and can be found here:

Ilya Markov is an assistant professor at the University of Amsterdam. His research builds around heterogeneous information access and algorithms that learn from users. The first line of Ilya's research focuses on models of user behavior with an emphasis on click models. He co-authored a book on Click Models for Web Search, created a corresponding tutorial, and developed PyClick - an open source Python library of commonly used click models. The second line of Ilya's research is focused on algorithms that learn from users, where he co-authored a number of papers on online LTR and bandit algorithms. Ilya did his PhD at the University of Lugano on uncertainty in distributed information retrieval.

Host: Prof. Fabio Crestani