Great companies strive to provide outstanding customer satisfaction like it's an additional product or service. As a business leader, understanding the market's opinion toward your product or brand is extremely valuable, but getting that information can be expensive and hard. This full-stack web app leverages the opinions of millions of Twitter users to deliver an easy way for leaders to guage consumer sentiment toward their product or brand based on what people are tweeting about it.
The user searches for a business and city name located in Colorado. If a matching business is found in our database, the search term entered by the user is used to retrieve tweets containing that term from Twitter's API. A sentiment score between 1-5 is returned to the user along with one positive and negative tweet used during the analysis.
For optimal results:
- Narrow search to only keywords.
- Avoid apostrophes and other puncutaion.
- Example: Instead of searching for "The Cheesecake Factory", try "Cheesecake Factory".
Sentiment is determined using Machine Learning and Natural Language Processing. One hundred of the latest tweets containing the keyword search term are retrieved from Twitter's API before our algorithm assigns a classification label (positive or negative) to each one.
A "sentiment score" between 1-5 is determined based on the number of positive vs. negative tweets. The model was built using Python and the Natural Language ToolKit library. You can explore our algorithm's source code and a detailed use case in this repository.
Access to larger Twitter datasets upon API requests require a paid Twitter developer account. This would greatly increase the accuracy of our sentiment score as well as allow us to conduct more in-depth analysis. We have opted to continue without signing up for a premium account until the app is production ready.
Raw data is collected from Twitter's API at the time of user interaction. Real-time access to Twitter's database allows us to calculate consumer sentiment toward any given keyword(s) and deliver the latest information. All business data is housed in our MongoDB database. This information was collected using various data mining techinques and API's.
- Analyze each tweet for the user's intent.
- This can provide actionable insights and help business leaders understand how their customers are interacting with their products and services.
- Analyze larger amounts of tweets.
- Twiter restricts the number of returned tweets to 100 per API request when using a standard developer account.
- Upgrading to premium or enterprise account is required to access more tweets per request.
- Lemmatize words like 'hi' and 'hiiii'.
- Detect sarcasm in tweets.
- Training the NaiveBayesClassifier - https://www.digitalocean.com/community/tutorials/how-to-perform-sentiment-analysis-in-python-3-using-the-natural-language-toolkit-nltk#step-1-%E2%80%94-installing-nltk-and-downloading-the-data
