HN Time-Machine is yet another HackerNews clone, but with a slight twist.
Each story on the current HackerNews frontpage is presented with the
top 3 most similar stories based on their titles between 2006 and
2015. More specifically, the titles are encoded into semantically
meaningful vectors and then ranked using cosine similarity.
If you want to learn more about this, check out our
blog post.
You will probably notice that your milage may vary.
Part of that could be attributed to there just not being a similar
story in the past. But, likely, also part of the reason is that the
model that encodes the sentences was trained on a different domain.
The perception of what is similar can vary greatly between different
domains.
You can help to (maybe) improve the model's understanding of
similarity in the domain of HackerNews stories by making use of the
star rating.
For feedback, questions and suggestions please reach out to
@phileisn.
