In the world of connectivity, a companies reputation is perhaps its most vital asset. How reputation changes with time and in different socioeconomic groups and forums can give insight on how different markets respond to branding and corporate actions. It can also give insight on what markets to target for different products and campaigns making branding more effective.

What it does

This proof-of-concept application pulled 1 month of public reddit comment data. It shows how sentiment in the forum changes over time with respect to companies. It can be compared to journalistic data (CNN Money, previously demo'd) to show how they correlate and possibly feed off one another.

How I built it

Coding was built on PHP (symphony) with a data analysis done with azure and stored on mysql.

Challenges I ran into

Passing data from different frameworks created complexities and time lags that need to be optimized.

Accomplishments that I'm proud of

Optimized code to increase run rate. Parsed 30GB of JSON. Optimized database to allow for real-time fuzzy-searching of millions of comments. Started a web app for publicly querying the dataset.

What I learned

Using Azure Data Base Using Blue Mix Utilizing different languages and code fragments Passing with database

What's next for INCedin

Next steps include (1) finding more interesting correlations between comment metadata and sentiment. We spent more time on the ability to efficiently query than the front end, so next is playing around with relationships. Also, we would like to see how different reddit subgroups respond differently to corporations. Also interesting would be to do personality analysis of posters based on linked list of all posts to compare how different personality profiles effect brand response. This could display a lack of brand connection with certain types of users. This data could be used to either target the market for products to the users best connected to the company or to add branding and marketing to connect with users currently unconnected.

Share this project: