User opinion learning

The theme of my dissertation is users' opinion learning. We propose three different studies to learn users' opinion using various approaches and to address several important research questions. Firstly, in order to discover the significant factors that induce the rating differences from user-generated reviews, we first extract possible specific influences from the review, known as aspects, and then we propose an unsupervised aspect-based sentiment learning system that assigns sentiment scores to potential aspects. Based on the sentiment scores, we adopt linear regression models to identify the aspects that lead to the rating differences. Food quality, service, dessert and drink quality, location, value, and general opinion toward the restaurants are recognized as the main influential factors that cause the Yelp rating differences among chain restaurants. Secondly, to understand the impact of time reminder designs such as counting down clock, progressing bar indicator, and remaining number of advertisements reminder embedded in specific long and short advertisement videos, we propose a 4 by 2 between-subject experimental study with follow-up survey questions to collect user's opinions toward different temporal designs in the video. Thirdly, our study analyzes the advertisement video designs from the content level. We design the advertisement video with high and low content relevance levels with the desired video. A 2 by 2 betweensubject experimental study with follow-up survey questions is proposed. Results point out that advertisement videos with high content relevance levels can lead to shorter video iv duration perception and less negative attitudes toward the video, but can also diminish the effectiveness of the advertisement with users recalling fewer products and brands promoted in both longer and shorter advertisement videos.