leiluk1 / matching-items

Matching items and searching for the most similar products

Matching and finding the most similar items

Matching is one of the basic machine learning tasks found in information retrieval, computer vision, recommender systems, etc.

Project tasks

🔸 Develop an algorithm that for all items from given validation dataset will suggest several options of the most similar products from base;

🔸 Evaluate the performance using Accuracy@5 metric.

Results

Overall, a search for the nearest 5 items on base for items from the validation dataset using Annoy Index showed an accuracy@5 metric of $71$%.

To check the results, you may see val_5_targets.csv in the results directory. This dataset contains the following features: id for the validation query (id), 5 nearest neighbours obtained for each query (target_0-target_4), and the expected most similar result for a query (expected) derived from the given validation_answers.