Hybrid online-offline learning to rank using simulated annealing strategy based on dependent click model

Learning to rank (LTR) is the process of constructing a model for ranking documents or objects. It is useful for many applications such as Information retrieval (IR) and recommendation systems. This paper introduces a comparison between Offline and Online (LTR) for IR. It also proposes a novel Offline (1 + 1)-Simulated Annealing Strategy (SAS-Rank) and introduces the first Hybrid Online-Offline LTR techniques using SAS-Rank and ES-Rank with Online Dependent Click Model (DCM). SAS-Rank is a combination of Simulated Annealing method and Evolutionary Strategy. From the obtained experimental results, we can conclude that the Offline LTR techniques outperformed the well-known Online Dependent Click Model (DCM) technique. Moreover, the Hybrid Online-Offline SAS-Click outperformed the predictive ranking results on unseen data in most evaluation fitness metrics using LETOR 4 dataset compared to other approaches. On the other hand, Hybrid ES-Click is a competitive approach with SAS-Click in evolving ranking models for training and validation data. Regarding Offline LTR, the SAS-Rank outperformed the well-known ES-Rank which has been compared in previous studies with fourteen machine learning techniques. This research uses the best available Linear LTR approaches existing in the literature which are offline ES-Rank with Online DCM. The linear LTR approach output is a linear ranking model which can be represented as a vector of feature importance weights. This paper demonstrated the results and findings obtained using the LETOR 4 dataset, and Java Archive Package is provided for facilitating reproducible research.

Automated summary

This paper compares the application of offline and online LTR in information retrieval and proposes a new offline ranking strategy. It also introduces a hybrid approach that combines online and offline LTR techniques and demonstrates through experiments that offline LTR outperforms online LTR.