SentiMetrics

Inspiration

As a team we wanted to prioritise user experience and kept a problem-solving business model at the core of our ideation process.
We were very clear that our intention was to solve a problem to make mundane and tedious tasks less cumbersome for our users.
Brand image is usually analysed by a medium sized team of public relations officers and is usually very difficult to keep updating the data on a regular basis, given the volatility that social media has brought about in public opinion.

Uses the BlueSky and X APIs to retrieve public opinion reviews filtered by hashtags specified by our users.
Performs a sentiment analysis on the data to generate a Sentiment Score for that particular user.
Automates the brand image analysis process essential to many personalities whose brand value is heavily reliant on public opinion, most widely expressed in today’s world through social media platforms.

Threads API: we were unable to fully understand the documentation and found it difficult to extract and access the token for the API.
Reddit API: there was a problem running it in the local host and the documentation was outdated. We solved these issues by exploring other social media platforms such as BlueSky and LinkedIn
Difference of opinion while deciding on a track and idea. To combat this, we listened attentively and contributed meaningfully to the "debate", keeping the interests of the team at the centre of our decision making process.

Chatbot embedded into the website.
Use of colour-coded sentiment analysis made for user-friendly interface.
Our determination and efficiency as a team throughout the course of the hackathon - we supported each other when needed and made sure we met our team goals to avoid rushing to complete before the deadline.
Using the various APIs together with the sentimentAnalyzer python module.

All team members significantly improved their knowledge of API usage, post requests to OpenAI and integration of the different frameworks and languages with each other.
We also significantly furthered our understanding of managing version control with Git, specifically tackling merge requests and failed pushes.

With more time and resources we would aim to train our own LLM to produce a more advanced sentimentAnalyzer module to further refine its classification of certain keywords as positive, negative or neutral.
We would improve the selection of tweets and posts to avoid tweets with hyperlinks and excessive emojis.
We would also expand the dataset by adding more social media APIs to our backend, e.g. Threads and Quora APIs.