Context enhanced joint word Embedding for detecting changes in Evolving text

Abstract

Semantics is the study of meaning in language. Semantic similarity newlinegives sense level as well as context level similarity between two terms or newlineconcepts. Semantic similarity models represent the implicit meaning of text newlineby specifying the concepts and relationships within the text. In this thesis, a newlinereview is performed to study the various aspects of semantic similarity of newlinelinguistic units and the features of semantic similarity models which provide newlinethe choice of models for the study of semantic change and topic drift. newlineSemantic change refers to the change in the meaning of a word that is newlinecompletely different from its previous usage. But topic drift denotes the newlinechange in the distribution of a group of words that describe a topic. newlineSemantic change happens due to the changes in the word usage and newlinerequires to be processed with different perceptions such as frequency, newlinesyntactic and semantic variations. Semantic change is generally related to the newlinevariation in the usage of n-gram contexts (especially noun or verb) of a word. newlineHowever, distant contexts of other word types also have an impact on newlinesemantic change. In this thesis, a knowledge-enhanced temporal word newlineembedding model has been built which jointly learns contexts from both newlinedependency and lexical relation vectors over time, for detecting the semantic newlinechange in the diachronic text corpus as well as type of change. newlineTopic drift occurs due to the changes in the groupwise distribution newlineof words and other tags co-occurring with words in the documents. Topic drift newlinein scholarly documents is normally linked to author context-based topical newlinechanges without discriminating the citation context from author context and newlinethe older citations from the recent one. In this thesis, a context-enhanced joint newlinetopical word embedding model has been designed which jointly learns newline