Hello! I just made an app really fast. I thought about it a few years ago but I thought it was going to be challenging. Streamlit made it really easy.
It uses Steam Reviews API plus a few link shenanigans. The model is just NMF from scikit-learn , it is not great but it works. I guess Amazon and the like are using LLMs now for this purpose.
The first panel are the top 300 reviews with their topics extracted (and top 4 words shown). The colors are according to if that topic belongs to thumbs up or down reviews.
The second panel forces 100 good reviews and 100 bad reviews, which is good to compare good and bad points from a game.
If you do not know about videogames but are interested in the topic extraction, you could try with “Outer Wilds”, “Skyrim” or “Lies of P: Overture” which is a DLC just released.
There are a few bugs (if you choose a game with few reviews) but I am happy with the result. Any comments will be appreciated. I would love to change the search bar so I only need one input, but the huge list of steam games made selectbox by itself really buggy.
List of requirements:
scikit-learn
nltk
streamlit
wordcloud
pandas
pillow