MSc Thesis Proposal of Vinay Kiran Manjunath:"Mining Twitter Sequences of Product opinions with aspect terms "

 

Date:  Friday, April 9th, 2021 

Time:  2:30 pm – 4:30 pm 

Meeting URL:    https://zoom.us/j/93084277345?from=addon 

Passcode: If interested in attending, contact the Graduate Secretary at csgradinfo@uwindsor.ca for the Meeting ID and Passcode

 

 

Social media platforms allow people to express and share their thoughts and opinions on the web in an effortless way. One typical example that illustrates the importance of opinion refers to enterprises that can capture customers' views about their products or their competitors. Twitter (also called Microblog, which restricts the character to 280 per post) provides a rich resource of consumer’s opinions about a product. However, mining opinions and sentiment from Twitter is very challenging due to the vast amount of data being generated. Aspect-based opinion mining (ABOM) allows an enterprise to analyze the data in detail saving time and money automatically. Given a microblog text, the aim is to mine the opinion (love, hate, etc.) made up of significant aspects (referred to as features of the most frequently mentioned products) and their associated opinion polarity (text expressed in the form of positive, negative, or neutral). 

 

Existing systems such as Hate Crime Twitter Sentiment (HCTS) and Microblog Aspect Miner (MAM) have been recently proposed to perform ABOM on Twitter. These systems generally go through the four-step approach of obtaining posts, identifying frequent nouns (candidate aspects), pruning the candidate aspects, and finally obtaining the opinion polarity. However, they differ in techniques for pruning their candidate aspects. HCTS proposed a new embedding feature selection method based on the Association Rule mining (created by analysing the data for frequent ‘if/then’ patterns) and Stanford Dependency Parser (grammatical structure) to extract the candidate aspects but doesnot consider the sequential ordering of the aspects. Microblog Aspect Miner (MAM) generates a summary (positive, negative, or neutral) of the expressed opinions based on the similarity of aspects that appeared in the users’ review. However, MAM doesnot obtain Multi-word aspect along with the sequential ordering. 

 

This thesis proposes a new system Microblog Aspect Sequence Miner (MASM), an extension of MAM by replacing the Apriori algorithm with the modified frequent sequential pattern mining algorithm. MASM takes a five-step approach of firstly classifying the posts into subjective and objective posts using a proposed classification algorithm based on Sentiment VADER scores. Then we extract the Parts of speech tag of reviews from the Twitter based on the keyword (PlayStation, Xbox, Samsung, Sony) to identify frequent nouns. Then to find the frequent noun phrases we apply CM-SPAM algorithm which is based on the frequent sequential patterns. We represent the frequent nouns/noun phrases as vectors to those products that are similar so that we can apply Agglomerative clustering to obtain relevant aspects to the product. To rank the relevant aspects, we finally make use of Latent Dirichlet Allocation which is used to model the relation between topics (candidate aspects) and the observed token (frequent nouns). Experiments show that this improves accuracy of obtaining relevant aspects of products from microblogs. 

 

Keywords: Aspect Based Opinion Mining, Twitter Sentiment Analysis, Sequential Pattern Mining. 

 

MSc Thesis Committee: 

MSc Thesis Proposal Announcement