Athena: Sentiment Analysis
This project consisted of three parts:
(I) The first piece is the sentiment analysis of electronic products from www.TechCrunch.com and www.CNET.com.
(II) The second part is focused on analyzing sentiment of amazon reviews for "George Foreman GRP99" product.
(III) Finally, the last part is on analyzing Macys reviews posted for "George Foreman GRP99" product.
Part I: Analyzing TechCrunch and CNET Reviews
Input
Users' online reviews from two technological websites: www.techcrunch.com and www.cnet.com for iPhone 5 which was release in Sep. 2012.
Output
People opinions sentiments for new iPhone 5
Dependencies
For running this code you need to have Python 2.7 and install PycURL & BeautifulSoup packages on your machine. Check link below for more details:
www.crummy.com/software/BeautifulSoup/
pycurl.sourceforge.net/
Technical Description
The goal of this project was to mine/analyze people opinions about the new iPhone 5 which has been released on Spetember 12, 2012 (www.cnet.com/iphone-5/) by mining relevent websites.
The code consists of two steps
A- Crawling online comments from web, cleaning up data and extracting important information from html content:
We developed a Python web crawler to pull out people's comments from techcrunch.com and cnet.com websites where they discussed th new released iPhone. This crawler downloads all reviews posted by people. Next, we used BeautifulSoup Python library to extract important data from HTML contents: www.crummy.com/software/BeautifulSoup/.
B- Running sentiment analysis using PycURL and text-processing API:
For the second step, we used a machine learning algortihm for sentiment analysis to find out what people think about the iPhone 5. We found an available API for the sentiment analysis implemented by Python NLTK (text-processing.com/demo/sentiment/). Thus, we used this website API in order to analyze the polarities of posted comments. For sending HTTP requests to text-processing website, we used a Python library called PycURL. PycURL is a Python interface to libcurl (www.pycurl.sourceforge.net/).
Results
For techcrunch.com, we collected 79 comments in total where 10 of them were labeled neutral, 13 positive, and 56 of them were labelled negative respect to the topic (iPhone 5). Below is the details:
pos=16.4%
neutral=12.6%
neg=71%
For cnet.com, we collected 37 comments in total where 5 of them were labeled neutral, 7 positive, and 25 of them were labelled negative. Below is the details:
pos=19%
neutral=13.5
neg=67%
Even if our dataset is limited, our results are relatively consistent across both websites. The average results considering the data from both websites are:
pos=17.7%
neutral=13.05%
neg=69%
Observations
Above results show how people thought about the new iPhone release after its demo on September 12, 2012 and it does not reflect the developer's opinion.
Part II: Analyzing Amazon Reviews:
In Nov 2014, I was contacted by Eliza Suzuki when we worked on the problem of extracting reviews for Amazon product and compute their sentiments. Eliza's goal was to see if there is any difference between Amazon stars and real people reviews. We focused on reviews about "George Foreman GRP99" product. For sentiment computation we used vaderSentiment library (https://pypi.python.org/pypi/vaderSentiment/0.5).
For more technical details please visit the article we published here: www.aioptify.com/crawling.php.
Part III: Analyzing Macys Reviews:
In Nov 2014, we also got interested in extracting Macys reviews for "George Foreman GRP99" product. The challenge with Macys website was that we couldn't use "urllib" module anymore due to content nature of the Macys website. So, we ended up using PhantomJS for crawling the website (http://phantomjs.org/).
For more technical details please visit the article we published here: www.aioptify.com/crawling.php.
If you have any question regard to this project, please drop me a note @ "k DOT jahanbakhsh AT gmail DOT com".