E-reputation evaluation system

• 

  system demo explaining the steps of the functioning: extracting social media data, storage, sentiment analysis then updating dashboards.

• 

  system interface showing the track of the online reputation of the djezzy telecom company on facebook including: reactions, interactions ...

Inspiration

With the rapid increase in the global population of Internet users connected to the Internet every day, and especially to social networks, digital awareness today is not taken lightly. Indeed, in a few minutes, a bad buzz can go viral and spoil the communication work of companies and brands, causing their image to change. Taking care of your e-reputation on the web is therefore more than a good practice: it is now a necessity. A company's digital image is made up of all the content it produces and distributes on the web. But more than that, it is mainly formed by the content that is posted by users themselves, on blogs, forums, community platforms, social networks, etc. In fact, it is commonly considered that 80% of what is said about a brand does not come from the brand itself, but is of great importance.

What it does

It consists of extracting comments from social media then analysing them using machine learning (treats Arabic, French, English and Algerian dialect), and finally updating dashboards (diagrames, numbers...). it detects the trending keywords.

How I built it

• Sentiment analysis using SVM • Creating an Algerian dialect dataset (Facebook + Twitter comments) • Adapting the Sentistrength solution to the Algerian dialect • Creating an Algerian dialect dictionary • Using web scraping to fetch the comments from Facebook (could be expended to instagram and twitter easily)

Challenges I ran into

the creation of the dataset was delicate to balace the positive and negative comments (not too political, not too comercial..), also some social media platform doesn't allow the free use of extracting APIs like facebook so i had to use web scraping. also the this solution treats the dialects not the academic langage, so it is adapted to social media dialect. and of course algerian dialect doesn't have a dataset or any kind of natural langage processing studies made on it because of it's complection and variety, as a result i had to create a new one and do many researchs regarding other arabian dialects like egyptian, syrian...

Accomplishments that I'm proud of

this is the the first effective solution for the algerian dialect, so enfluencers, companies and any other organism with existance on social media can keep track of thier online reputation in real time and i had a precision of 89,44%, it has been tested in OTA djezzy company and it worked just fine. and of course any other organism in the world can benefit from it anytime anywere since it could be accessed from any divice and it treats french and english and arabic as well.

What I learned

i learned a lot about the marketing strategies during my meetings in the companies to see thier needs. also, i learned a lot of technical things and gained experience. and i learned from the researchs about the social media dialects and the way people experess themselves and how i could find solutions to that in a deep level.

What's next for E-reputation evaluation system

We can improve the dictionary and dataset Hybridize both approaches and apply unsupervised Apply more machine learning methods on the Algerian language take advantage of more data collected from the web

Built With

• machine-learning
• online-reputation
• python
• sentiment-analysis