Sentiment mining based on the human reading model

Abstract

Sentiment mining is a recent field that looks at finding methods for automating the extraction and interpretation of people sentiment in text. Past research in the field has primarily focused on finding trechniques that rely on data mining approaches with primary emphasis on efficient statistical and mathematical algorithms to classify subjectivity in the text. While these methods have provided significant advances in the field of sentiment mining, several challenges remain in achieving high accuracy across a variety of topics, and in extracting the semantics associated with the topic. In this thesis, we propose to address these challenges by taking a new and fundamentally different approach to the problem of sentiment mining. The approach first examines how it may ideally be performed by looking at how humans’ brains process the task of sentiment interpretation from text. We examine how humans read, interpret and then extract sentiment when reading.We conclude based on studies in psychology, that humans use two phases of low level and high level mental processing and then use pre-concieved notions of the topic to derive the sentiment in text. We propose to automate the defined human reading process towards the interpretation of sentiment in text with primary focus on modeling pre-conceievd notions. The automation of the notion-centric of the human process involves the use of natural language processing (NLP) techniques to extract text features and the development of notions corpora. The remaining step in the automation process is the use of machine learning algorithms to classify sentiments based on the extracted features and the developed notions corpora. In summary, the key contributions of this thesis are: 1. A model representing human’s pre-conceived notions, 2. The development of notions corpora for sample topics, and 3. A mapping of notion-centrichuman reading process to derive a method for sentiment classification. Experiments show the effectiveness of the method and its ability to provi