-
HeadlineHound home page
HeadlineHound
HeadlineHound is a Typescript-based project that uses natural language processing to summarize news articles for you. With HeadlineHound, you can quickly get the key points of an article without having to read the entire thing. It's perfect for anyone who wants to stay up-to-date with the news but doesn't have the time to read every article in full. Whether you're a busy professional, a student, or just someone who wants to be informed, HeadlineHound is a must-have tool in your arsenal.
How it Works
HeadlineHound uses natural language processing (NLP) via a fine-tuned ChatGPT to analyze and summarize news articles. It extracts the most relevant sentences and phrases from the article to create a concise summary. The user simply inputs the URL of the article they want summarized, and HeadlineHound does the rest.
How we built it
We first fine-tuned a ChatGPT model using a dataset containing news articles and their summaries. This involved a lot of trying different datasets, testing various different data cleaning techniques to make our data easier to interpret, and configuring OpenAI LLM in about every way possible :D. Then, after settling on a dataset and model, we fine-tuned the general model with our dataset. Finally, we built this model into our webapp so that we can utilize it to summarize any news article that we pass in. We first take in the news article URL, pass it into an external web scraping API to extract all the article content, and finally feed that into our LLM to summarize the content into a few sentences.
Challenges we ran into
Our biggest challenge with this project was trying to determine which dataset to use and how much data to train our model on. We ran into a lot of memory issues when trying to train it on very large datasets and this resulted in us having to use less data than we wanted to train it, resulting in summaries that could definitely be improved. Another big challenge that we ran into was determining the best OpenAI model to use for our purposes, and the best method of fine-tuning to apply.
Accomplishments that we're proud of
We are very proud of the fact that we were able to so quickly learn how to utilize the OpenAI APIs to apply and fine-tune their generalized models to meet our needs. We quickly read through the documentation, played around with the software, and were able to apply it in a way that benefits people. Furthermore, we also developed an entire frontend application to be able to interact with this LLM in an easy way. Finally, we learned how to work together as a team and divide up the work based on our strengths to maximize our efficiency and utilize our time in the best way possible.
What we learned
We learned a lot about the power of NLP. Natural language processing (NLP) is a fascinating and powerful field, and HeadlineHound is a great example of how we can use LLMs can be used to solve real-world problems. By leveraging these generalized AI models and then fine-tuning them for our purposes, we were able to create a tool that can quickly and accurately summarize news articles. Additionally, we learned that in order to create a useful tool, it's important to understand the needs of the user. With HeadlineHound, we recognized that people are increasingly time-poor and want to be able to stay informed about the news without having to spend their precious time reading articles. By creating a tool that meets this need, we were able to create something that people saw value in.
What's next for Headline Hound
Here are a few potential next steps for the HeadlineHound project:
Improve the summarization algorithm: While HeadlineHound's summarization algorithm is effective, there is always room for improvement. One potential area of focus could be to improve the algorithm's ability to identify the most important sentences in an article, or to better understand the context of the article in order to generate a more accurate summary.
Add support for more news sources: HeadlineHound currently supports summarizing articles from a wide range of news sources, but there are always more sources to add. Adding support for more sources would make HeadlineHound even more useful to a wider audience.
Add more features: While the current version of HeadlineHound is simple and effective, there are always more features that could be added. For example, adding the ability to search for articles by keyword could make the tool even more useful to users.
Log in or sign up for Devpost to join the conversation.