Aspect Based Semantic Sentiment Analysis Using Sentence and Document Level Online Texts

Abstract

The proliferation of the internet services has led to an exponential increase in newlineonline activities. Easy availability and accessibility has incurred ever-growing newlinemassive volumes of User Generated Content (UGC). This UGC can be explored newlineand analysed to gain deeper insights into user opinions and behaviours. newlineAlthough available in abundance, UGC predominantly comes in the unstructured newlineand semistructured forms. Hence, it becomes more challenging to extract, newlineexplore and analyse UGC, for gaining the required information and knowledge. newlineSentiment analysis has been popularly used to mine unstructured data for newlinediscovering users opinion and emotion quotient collectively. Considerable newlineresearch has been done in this area, however, one big lacuna is that the existing newlinetechniques are mostly domain-specific and problem-specific, and are not-so-easy newlineto extend their applications to other domains. Specifically, there is a dearth of newlinetechniques that employ implicit textual features and word relations to gain newlinedeeper and more reliable insights. Applying techniques that rely on implicit newlinerelations may enable us to consider implicit aspects, ambiguous words, slangs, newlinemisspelled words and other special sentiment bearing words while analyzing newlinethese informal corpora of informal texts. newlineTo overcome some of these lacunas, in this thesis, we explore implicit newlinesemantic relationships, word co-occurrences and contextual information for the newlinetask of aspect-based sentiment analysis. Specifically, we divide this main task by newlinethe type of data, as the task of deriving implicit word relation becomes more newlinechallenging in the case of short texts and sparse representations. The first set newlineincludes short texts present in the form of sentences, acquired from the newlineTwitter.com. The other set of data includes long texts present in the form of newlinedocuments and paragraphs. We employ two real-world datasets comprising of newlineonline consumer reviews acquired from the Amazon.com and TripAdvisor.com. newlineFirst, we showcase the importance of non-textual and non-conventional newlinefeatures in