Supervised learning for image search result diversification

Göynük, Burak
Due to ambiguity of user queries and growing size of data living on the internet, methods for diversifying search results have gained more importance lately. While earlier works mostly focus on text search, a similar need also exists for image data, which grows rapidly as people produce and share image data via their smartphones and social media applications such as Instagram, Snapchat, and Facebook. Therefore, in this thesis, we focus on the result diversification problem for image search. To this end, as our first contribution, we adopt R-LTR, a supervised learning approach that has been proposed for textual data and modify it to allow tuning the weights of visual and textual features separately, as would be required for better diversification. As a second contribution, we extend R-LTR by applying an alternative paradigm that takes into account an upperbound for the future diversity contribution that can be provided by the result being scored. We implement R-LTR and its variants using PyTorch’s neural network framework, which enables us to go beyond the original linear formulation. Finally, we create an ensemble of the most promising approaches for the image diversification problem. Our experiments using a benchmark dataset with 153 queries and 45K images reveal that the adopted supervised algorithm, RLTR, significantly outperforms various ad hoc diversification approaches in terms of thesub-topicrecallmetric. Furthermore,certainvariantsofR-LTRproposedhereare superior to the original method and provide additional (relative) gains of up to 2.2%


Advanced methods for diversification of results in general-purpose and specialized search engines
Yiğit Sert, Sevgi; Altıngövde, İsmail Sengör; Ulusoy, Özgür; Department of Computer Engineering (2020-12-28)
Diversifying search results is a common mechanism in information retrieval to satisfy more users by surfacing documents that address different possible intentions of users. It aims to generate a result list that is both relevant and diverse when ambiguous and/or broad queries appear. Such queries have different underlying subtopics (a.k.a., aspects or interpretations) that search result diversification algorithms should consider. In this thesis, we first address search result diversification as a useful met...
Limitations and improvement opportunities for implicit result diversification in search engines
Ulu, Yaşar Barış; Altıngövde, İsmail Sengör; Department of Computer Engineering (2019)
Search engine users essentially expect to find the relevant results for their query. Additionally, the results of the query should contain different possible query intents, which leads to the well-known problem of search result diversification. Our work first investigates the limitations of implicit search result diversification, and in particular, reveals that typical optimization tricks (such as clustering) may not necessarily improve the diversification effectiveness. Then, as our second contribution, we...
Explicit Search Result Diversification Using Score and Rank Aggregation Methods
Ozdemiray, Ahmet Murat; Altıngövde, İsmail Sengör (2015-06-01)
Search result diversification is one of the key techniques to cope with the ambiguous and underspecified information needs of web users. In the last few years, strategies that are based on the explicit knowledge of query aspects emerged as highly effective ways of diversifying search results. Our contributions in this article are two-fold. First, we extensively evaluate the performance of a state-of-the-art explicit diversification strategy and pin-point its potential weaknesses. We propose basic yet novel ...
Characterizing web search queries that match very few or no results
Altıngövde, İsmail Sengör; Cambazoglu, Berkant Barla; Ozcan, Rifat; Sarigil, Erdem; Ulusoy, Özgür (2012-12-19)
Despite the continuous efforts to improve the web search quality, a non-negligible fraction of user queries end up with very few or even no matching results in leading web search engines. In this work, we provide a detailed characterization of such queries based on an analysis of a real-life query log. Our experimental setup allows us to characterize the queries with few/no results and compare the mechanisms employed by the major search engines in handling them.
Supervised approaches for explicit search result diversification
Yigit-Sert, Sevgi; Altıngövde, İsmail Sengör; Macdonald, Craig; Ounis, Iadh; Ulusoy, Özgür (Elsevier BV, 2020-11-01)
Diversification of web search results aims to promote documents with diverse content (i.e., covering different aspects of a query) to the top-ranked positions, to satisfy more users, enhance fairness and reduce bias. In this work, we focus on the explicit diversification methods, which assume that the query aspects are known at the diversification time, and leverage supervised learning methods to improve their performance in three different frameworks with different features and goals. First, in the LTRDiv ...
Citation Formats
B. Göynük, “Supervised learning for image search result diversification,” Thesis (M.S.) -- Graduate School of Natural and Applied Sciences. Computer Engineering., Middle East Technical University, 2019.