/search-reranking-model

Primary LanguageJupyter Notebook

Search Reranking Model - POC

This repository contains the source code for a search reranking model that is designed to improve the accuracy and relevance of search results for e-commerce product searches. The model is based on linear regression and uses various features of the search query results and product data to rerank the search results.

It takes in a list of products with their attributes and returns the reordered product IDs and their corresponding predicted scores.

File Structure

The directory structure of the repository is as follows:

search_reranking_model/
├── Readme.md
├── data/
│   ├── bgq_data_search.csv
│   └── cached_dataframe.joblib
├── docs/
│   └── reranking-model-arch.png
├── model/
│   └── search_reranking_model.joblib
├── model_train.py
├── notebooks/
│   ├── Workbook-assumptions_of_LR.ipynb
│   └── search_reranking_model.ipynb
├── out/
│   └── reranked_product_scores.csv
├── requirements.txt
├── rerank_products.py
└── reranking_model.py

  • The data directory contains the input data for the model, including the search query data (bgq_data_search.csv) and a cached version of the preprocessed data (cached_dataframe.joblib). The model directory contains the trained model (search_reranking_model.joblib).

  • The docs directory contains documentation related to the model architecture, including a diagram (reranking-model-arch.png) that illustrates the various components and inputs.

  • The model_train.py file contains the code to train the model on the input data.

  • The notebooks directory contains Jupyter notebooks that were used for exploratory data analysis and model testing. The Workbook-assumptions_of_LR.ipynb notebook explores the assumptions of logistic regression and the search_reranking_model.ipynb notebook demonstrates the use of the trained model on new data.

  • The out directory contains the final output of the model, which is a csv file (reranked_product_scores.csv) containing the reranked scores for each product based on the search query.

  • The requirements.txt file lists the required Python packages and their versions.

  • The rerank_products.py script is the main script for running the model on new data. It preprocesses the data, applies the model, and generates the reranked product scores.

  • The reranking_model.py file contains the code for defining and applying the logistic regression model.

Installation

  • Clone the repository: git clone https://github.com/akhilesh-k/search-reranking-model.git
  • Install the required packages: pip install -r requirements.txt

Usage

To use the product predictor, follow these steps:

  • Setup training data, if you want to play with the model. You can use the bgq query to import the dataframe if you have the access.

  • Prepare your input data as a list of products with their IDs and attributes.

  • Instantiate the ProductPredictor class by passing the path to the trained machine learning model:

    from product_predictor import ProductPredictor
model_path = 'path/to/model.joblib'
predictor = ProductPredictor(model_path)

  • Add each product to the ProductPredictor instance using the add_product method: product_id = 'SKU-0000-12345' attributes = [4.0, 0, 5.0, 1, 0.020829, 0.552162, 0.005076, 0.010152, 1.0, 0.149546, 0.046763, -0.423838, -0.000822, -0.097556, -0.787021] predictor.add_product(product_id, attributes)

  • Call the predict_reranked_product_ids method on the ProductPredictor instance to get the predicted reordered product IDs and their corresponding scores:

    reranked_product_ids, predicted_scores = predictor.predict_reranked_product_ids()

  • Save the predicted scores to a CSV file: results_df = pd.DataFrame({'Product ID': reranked_product_ids, 'Score': [predicted_scores[id] for id in reranked_product_ids]}) results_df.to_csv('path/to/output.csv', index=False)

License

This project is licensed under the MIT License. See the LICENSE file for details.