SoulMate: Short-Text Author Linking Through Multi-Aspect Temporal-Textual Embedding

Linking authors of short-text contents has important usages in many applications, including Named Entity Recognition (NER) and human community detection. However, certain challenges lie ahead. First, the input are noisy, ambiguous, do not follow grammatical rules. Second, traditional text mining methods fail to effectively extract concepts through words phrases. Third, textual temporally skewed, which can affect semantic understanding by multiple time facets. Finally, using knowledge-bases make results biased content external database deviate meaning from short corpus. To overcome these challenges, we devise a neural network-based temporal-textual framework that generates subgraphs with highly correlated contents. Our approach, on one hand, computes relevance score (edge weight) between considering portmanteau concepts, other employs stack-wise graph cutting algorithm communities related authors. Experimental show compared knowledge-centered competitors, our multi-aspect vector space model achieve higher performance linking In addition, given author task, more comprehensive dataset is, significance extracted will be.