PROJECT TITLE :

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

ABSTRACT:

Linking the authors of short-text contents has important usages in a wide variety of applications, including Named Entity Recognition (NER) and the detection of human communities. Nevertheless, there are going to be some difficulties. To begin, the contents of the input short text are jumbled, unclear, and do not adhere to the established grammatical rules. Second, conventional text mining techniques are ineffective at gleaning concepts from individual words and phrases. Third, the contents of the text are temporally skewed, which can affect the semantic understanding in a variety of ways depending on the different aspects of time. In conclusion, the use of knowledge bases has the potential to produce results that are biased toward the content of an external database and that depart in meaning from the input short text corpus. We devise a neural network-based temporal-textual framework in order to overcome these challenges. This framework generates the subgraphs with highly correlated authors based on the contents of short-text documents. Our method, on the one hand, computes the relevance score (edge weight) between the authors by taking into consideration a portmanteau of contents and concepts, and, on the other hand, employs a stack-wise graph cutting algorithm in order to extract the communities of the related authors. Both of these aspects are taken into consideration in order to produce accurate results. The results of our experiments indicate that, in comparison to other knowledge-centered competitors, our multi-aspect vector space model is able to achieve a higher level of performance when it comes to linking short-text authors. In addition, taking into consideration the author linking task, the significance of the extracted concepts will increase in direct proportion to the comprehensiveness of the dataset.