Inspiration
This app leverages the multilingual semantic model from COhere to revolutionize media and news industry for multilingual markets like Africa
What it does
Enable any person to track news in real time without translating or understanding other regional languages.
How we built it
We used Coherer's multilingual embeddings to build an index for 6 African languages. Using Annoy, we performed a nearest-neighbor search. The user interface was built and hosted using Streamlit.
Challenges we ran into
African languages have difficulty understanding eg. Tonation sometimes gives different meanings. Cohere Multilingual Model still has more resource language to expand due to a large number of people speaking it.
What's next for AFri News Multilingual Embedding
- Train or finetune a custom large language model on more African languages to improve the vectorization of them
- Interpret the attention mechanism or the internals of the transformer to identify salient sentences in the news articles to be used as new concepts
- Updating the pipeline to recommend the latest news article
Built With
- cohere
- python
- streamlit
Log in or sign up for Devpost to join the conversation.