Trustworthy Location Review

Inspiration

For many location-based platforms such as Google Maps, there are often many irrelevant, misleading, or low-quality reviews and even scams. We wanted to create a tool that can help people quickly identify trustworthy and contextually relevant feedback, so decision-making becomes easier and more reliable.

What it does

Trustworthy Location Review uses AI to filter and classify location-based reviews to get rid of these 'bad' reviews. It detects spam, irrelevant rants, and advertisements, while surfacing genuine, context-aware insights that help users make better choices.

How we built it

Development Tools: Google Colab Frameworks/Libraries: Hugging Face Transformers, PyTorch, pandas Datasets: Google Local Reviews dataset + manually labeled samples from Google Maps We clean up the data provided and use a prompt based on the three policies for improper review format to use the Qwen3-8B model to filter and classify the reviews.

Challenges we ran into

Handling missing data: we discussed how we shall deal with missing data in the dataset and in the end we decide to remove data that lacks user id and restaurant names. Figuring out how to use the trained model: we got conflict that we want to use the attribute text-generation, but Qwen3-8B does not have that attribute, so we changed to 'conversational'; but because we are using InferenceClient, it has no attribute 'conversational'. Time management: this hackathon happes in week 3 where all of us are very busy due to tutorials, labs and CCAs so it's hard for us to find time working together while coping with school stuff. Because of this, we were only able to dedicate ourselves to filtering out "advertisement" reviews as we feel that this is a growing problem with the rise of bot and spam on reviews.

Accomplishments that we're proud of

Figured out ways to solve the challenges that we met.

What we learned

How to use Colab, how to import hugging face model, how to fine-tune it,

What's next for Trustworthy Location Review

We want to do critique fine-tuning and scale it to multiple forms, and also add a user-interface.

Built With

• colab
• github
• google-maps
• huggingface
• python