A Trio Neural Model for Dynamic Entity Relatedness Ranking

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 31–41 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 31 A Trio Neural Model for Dynamic Entity Relatedness Ranking Tu Ngoc Nguyen L3S Research Center tunguyen@L3S.de Tuan Tran Robert Bosch GmbH anhtuan.tran2@de.bosch.com Wolfgang Nejdl L3S Research Center nejdl@L3S.de Abstract Measuring entity relatedness is a fundamen- tal task for many natural language processing and information retrieval applications. Prior work often studies entity relatedness in static settings and an unsupervised manner. How- ever, entities in real-world are often involved in many different relationships, consequently entity-relations are very dynamic over time. In this work, we propose a neural network- based approach for dynamic entity relatedness, leveraging the collective attention as supervi- sion. Our model is capable of learning rich and different entity representations in a joint framework. Through extensive experiments on large-scale datasets, we demonstrate that our method achieves better results than com- petitive baselines. 1 Introduction Measuring semantic relatedness between entities is an inherent component in many text mining ap- plications. In search and recommendation, the ability to suggest most related entities to the entity-bearing query has become a standard fea- ture of popular Web search engines (Blanco et al., 2013). In natural language processing, entity re- latedness is an important factor for various tasks, such as entity linking (Hoffart et al., 2012) or word sense disambiguation (Moro et al., 2014). However, prior work on semantic relatedness often neglects the time dimension and consider entities and their relationships as static. In prac- tice, many entities are highly ephemeral (Jiang et al., 2016), and users seeking information re- lated to those entities would like to see fresh infor- mation. For example, users looking up the entity Taylor Lautner during 2008–2012 might want to be recommended with entities such as The Twi- light Saga, due to Lautner’s well-known perfor- mance in the film series; however the same query in August 2016 should be served with entities re- lated to his appearances in more recent films such as “Scream Queens”, “Run the Tide”. In addition, much of previous work resorts to deriving seman- tic relatedness from co-occurrence -based compu- tations or heuristic functions without direct opti- mization to the final goal. We believe that desir- able framework should see entity semantic relat- edness as not separate but an integral part of the process, for instance in a supervised manner. In this work, we address the problem of en- tity relatedness ranking, that is, designing the se- mantic relatedness models that are optimized for ranking systems such as top-k entity retrieval or recommendation. In this setting, the goal is not to quantify the semantic relatedness between two entities based on their occurrences in the data, but to optimize the partial order of the related entities in the top positions. This problem dif- fers from traditional entity ranking (Kang et al., 2015) in that the entity rankings are driven by user queries and are optimized to their (ad-hoc) in- …

Content truncated to 500 words