#### Presentation Title

Effective Sentiment Analysis Models

#### Faculty Mentor

Abhishek Verma

#### Start Date

18-11-2017 10:00 AM

#### End Date

18-11-2017 11:00 AM

#### Location

BSC-Ursa Minor 75

#### Session

Poster 1

#### Type of Presentation

Poster

#### Subject Area

engineering_computer_science

#### Abstract

Sentiment Analysis takes large amounts of data and derive an opinion from that data using a variety of tools. The purpose of this analysis is to take key words and possible thoughts from a movie review and compare it to the person’s opinion who is writing the review to provide an accurate rating data cluster. The analysis would also need to account for satirical tools, such as sarcasm, negation, and ambiguity also to differentiate a good review and a sarcastic review with similar keywords.

In this case, the analysis done takes mass datasets IMDB reviews in order to decide if a movie is generally good or bad. These are common review systems for people when looking at possible movies, that being said, using mass data mining systems would be exceedingly useful to understanding positive, negative, neutral, and satirical reviews. This particular dataset is to be labeled into five sectors, negative, somewhat negative, neutral, somewhat positive, and positive. Stanford University analyzed this and formed a source code with mostly accurate information using deep learning, vector modeling as well as recursive modeling (Socher, 2). This system also examines the entirety of a sentence and ordering of the words. This gives us a clear, decisive, and most importantly, an accurate system to understand the true entertainment value of a given movie based on the reviews, without actually seeing each and every review. This kind of system would help review system optimization as well as well as for people looking at these mass reviews (Maas, 1).

#### Summary of research results to be presented

Expected Results

The expected results are fairly simple, it is expected to find the logistic regression win/loss method to be more effective in accuracy versus the baseline model, Naive Bayes. The binary analysis used by logistic regression is expected to serve a much greater use than the basic probability baseline model that Naive Bayes provides. It is also expected that Naive Bayes scores around a 66% accuracy due to previous sentiment analysis tests appeared that Naive Bayes tends to score around that area. Whereas, logistic regression tends to score around an 80% or so. These are the results we expect to find, but not guaranteed.

Effective Sentiment Analysis Models

BSC-Ursa Minor 75

Sentiment Analysis takes large amounts of data and derive an opinion from that data using a variety of tools. The purpose of this analysis is to take key words and possible thoughts from a movie review and compare it to the person’s opinion who is writing the review to provide an accurate rating data cluster. The analysis would also need to account for satirical tools, such as sarcasm, negation, and ambiguity also to differentiate a good review and a sarcastic review with similar keywords.

In this case, the analysis done takes mass datasets IMDB reviews in order to decide if a movie is generally good or bad. These are common review systems for people when looking at possible movies, that being said, using mass data mining systems would be exceedingly useful to understanding positive, negative, neutral, and satirical reviews. This particular dataset is to be labeled into five sectors, negative, somewhat negative, neutral, somewhat positive, and positive. Stanford University analyzed this and formed a source code with mostly accurate information using deep learning, vector modeling as well as recursive modeling (Socher, 2). This system also examines the entirety of a sentence and ordering of the words. This gives us a clear, decisive, and most importantly, an accurate system to understand the true entertainment value of a given movie based on the reviews, without actually seeing each and every review. This kind of system would help review system optimization as well as well as for people looking at these mass reviews (Maas, 1).