← Back to projects

Data Mining · NLP · 2026

Food Delivery Pattern Analysis

Association rules, PageRank and BERT sentiment turned into smart meal recommendations.

PythonAprioriFP-GrowthNetworkXBERTPandasMatplotlib

5

Techniques

PageRank

Graph ranking

BERT

NLP model

Recommender

Output

The problem

Delivery platforms such as Talabat, Uber Eats and Deliveroo sit on huge order logs and review text, but rarely connect the two into a single view of what customers actually want.

The solution

A combined data mining and NLP pipeline that discovers which meals travel together, ranks the most influential dishes in an order graph, reads the sentiment of reviews, and merges all three signals into recommendations.

Association rule mining

Apriori and FP-Growth surface frequent itemsets and the rules behind them.

  • Burger ➜ Fries, Pizza ➜ Cola, Pasta ➜ Garlic Bread, Shawarma ➜ Pepsi
  • Support, confidence and lift used to keep only meaningful rules
  • Rules feed bundle offers and cross-sell suggestions

Graph analysis & PageRank

  • Orders modelled as a graph of meals connected by co-purchase
  • PageRank ranks the most central and popular dishes
  • Link analysis exposes clusters of meals ordered by the same segments

BERT sentiment & recommendations

  • Customer reviews classified with a BERT transformer
  • Sentiment score blended with rule confidence and PageRank weight
  • Final recommender proposes meals customers are likely to order and to enjoy

Results & impact

  • Clear, explainable ordering patterns instead of black-box suggestions
  • Popularity ranking that reflects order structure, not just raw counts
  • Recommendations that account for how customers felt, not only what they bought

More projects