A Trio Neural Model for Dynamic Entity Relatedness Ranking

TL;DR

The paper tackles dynamic entity relatedness ranking by modeling time-varying relevance between entities. It proposes a Trio Neural Ranking framework that jointly learns content, graph, and temporal representations, with a time-weighted CNN to capture recency effects. The model is trained with navigation-derived supervision and evaluated on Wikipedia clickstream data, showing improvements over strong baselines. These findings enable more accurate, time-sensitive entity suggestions in search and recommendation tasks.

Abstract

Measuring entity relatedness is a fundamental task for many natural language processing and information retrieval applications. Prior work often studies entity relatedness in static settings and an unsupervised manner. However, 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 networkbased approach for dynamic entity relatedness, leveraging the collective attention as supervision. 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 competitive baselines.

Figures (6)

• Figure 1: The dynamics of collective attention for related entities of Taylor Lautner in 2016.
• Figure 2: Click (navigation) times distribution and ranking correlation of entities in September 2016.
• Figure 3: The trio neural model for entity ranking.
• Figure 4: Architecture of the temporal convolution module. Local temporal patterns are extracted by a 1-D convolution and then modulated by a monotonic time-decay weighting before projection into the final temporal embedding space.
• Figure 5: Performance results for variation of decay parameter and different entity types.