Incorporating Deep Visual Features into Multiobjective based Multi-view Search Results Clustering

Proceedings of the 27th International Conference on Computational Linguistics, pages 3793–3805 Santa Fe, New Mexico, USA, August 20-26, 2018. 3793 Incorporating Deep Visual Features into Multiobjective based Multi-view Search Results Clustering Sayantan Mitra Department of CSE, IIT Patna, Bihta, India sayantaniem@gmail.com Mohammed Hasanuzzaman ADAPT Centre, School of Computing, Dublin City University, Dublin, Ireland Sriparna Saha Department of CSE, IIT Patna, Bihta, India Andy Way ADAPT Centre, School of Computing, Dublin City University, Dublin, Ireland Abstract Current paper explores the use of multi-view learning for search result clustering. A web-snippet can be represented using multiple views. Apart from textual view cued by both the semantic and syntactic information, a complementary view extracted from images contained in the web- snippets is also utilized in the current framework. A single consensus partitioning is finally ob- tained after consulting these two individual views by the deployment of a multi-objective based clustering technique. Several objective functions including the values of a cluster quality mea- sure evaluating the goodness of partitionings obtained using different views and an agreement- disagreement index, quantifying the amount of oneness among multiple views in generating par- titionings are optimized simultaneously using AMOSA. In order to detect the number of clusters automatically, concepts of variable length solutions and a vast range of permutation operators are introduced in the clustering process. Finally a set of alternative partitionings are obtained on the final Pareto front by the proposed multi-view based multi-objective technique. Experimental results by the proposed approach on several bench-mark test datasets with respect to different performance metrics evidently establish the power of visual and text based views in achieving better search result clustering. 1 Introduction Web search results clustering (SRC), also known as ephemeral clustering or post-retrieval clustering has garnered much attention in the past few decades for making web browsing easier for users. The key objective of SRC systems is: for a given query it can return some meaningful labeled clusters from a set of web documents (or web snippets) retrieved from a search engine. Recent years have witnessed a large number of attempts in solving this SRC problem (Di Marco and Navigli, 2013; Scaiella et al., 2012). Most of them have developed some clustering algorithms optimizing a single objective criteria (Osinski and Weiss, 2005; Zamir and Etzioni, 1998; Moreno et al., 2013). But a complex data set like set of web-snippets can be clustered into several alternative partitionings. Therefore to detect all possible partitionings containing clusters of different shapes automatically, application of multiobjective optimization (MOO) for solving the problem of clustering becomes prevalent (Maulik et al., 2011). In this context, Acharya et al. (Acharya et al., 2014) have proposed a multi objective optimization based approach for solving SRC problem by extracting both semantic and syntactic information present in web-snippets. Results attained by this approach outperformed the other existing single objective based approaches. Moreover a web-snippet can be represented using different views, for example semantic view, syntactic view. Recently, Wahid et al. (Wahid et al., 2014) have developed a multi-view based …

Content truncated to 500 words