The research presented in this study focuses on enhancing the relevance of Embedding-based Neural Retrieval (EBR) at Walmart for product search. The initial implementation of the EBR system at Walmart showed promising results in improving relevance and add-to-cart rates. However, there were instances of relevance degradation due to factors such as false positives/negatives in training data and difficulties in handling query misspellings. To address these challenges, the researchers proposed several approaches to strengthen the capabilities of the EBR model in terms of retrieval relevance. One key contribution is the introduction of a Relevance Reward Model (RRM) based on human relevance feedback. This model helps filter out noise from training data and incorporates it into the EBR model through a multi-objective loss function. Additionally, techniques like typo-aware training and semi-positive generation were employed to enhance the performance of the EBR model further. These strategies aim to improve retrieval accuracy by addressing common issues encountered during product search queries. The effectiveness of these enhancements was validated through offline relevance evaluation, online AB tests, and successful deployments in live production environments. The study showcases how refining the EBR model can lead to significant improvements in retrieval relevance, ultimately enhancing the overall customer shopping experience at Walmart.
- - Research focuses on enhancing relevance of Embedding-based Neural Retrieval (EBR) at Walmart for product search
- - Initial implementation of EBR system at Walmart showed promising results in improving relevance and add-to-cart rates
- - Challenges included relevance degradation due to false positives/negatives in training data and difficulties in handling query misspellings
- - Proposed approaches to strengthen EBR model capabilities, including:
- - Introduction of Relevance Reward Model (RRM) based on human relevance feedback
- - Techniques like typo-aware training and semi-positive generation employed to enhance performance
- - Strategies aim to improve retrieval accuracy by addressing common issues encountered during product search queries
- - Effectiveness of enhancements validated through offline relevance evaluation, online AB tests, and successful deployments in live production environments
SummaryResearch at Walmart is working on making it easier to find products online. They tried a new system that helped show better results when people searched for items to buy. But there were some problems like wrong search results and misspelled words. To make the system better, they are using feedback from people and special techniques to improve how it works. The goal is to help people find what they want to buy more accurately.
Definitions- Research: A careful study or investigation done to discover new information.
- Embedding-based Neural Retrieval (EBR): A method of finding information using a system that learns patterns in data.
- Relevance: How closely something matches what you are looking for.
- Add-to-cart rates: How often people add items to their online shopping cart.
- False positives/negatives: Incorrect results shown as relevant or irrelevant.
- Query misspellings: Words typed incorrectly when searching for something online.
- Relevance Reward Model (RRM): A way of improving search results based on feedback from users.
- Typo-aware training: Training the system to recognize and correct spelling mistakes in searches.
- Semi-positive generation: Creating examples that are partly correct to help improve performance.
- Retrieval accuracy: How well the system can find the right information during a search query.
Introduction
In today's digital age, online shopping has become the go-to method for purchasing goods and services. With the rise of e-commerce giants like Walmart, it is crucial to provide customers with a seamless and efficient shopping experience. One key aspect of this experience is product search, where customers can easily find what they are looking for on the website or app. To improve product search relevance at Walmart, researchers have turned to embedding-based neural retrieval (EBR) models.
The initial implementation of EBR at Walmart showed promising results in improving relevance and add-to-cart rates. However, there were instances of relevance degradation due to factors such as false positives/negatives in training data and difficulties in handling query misspellings. To address these challenges, the research presented in this study focuses on enhancing the capabilities of EBR for product search at Walmart.
The Importance of Relevance in Product Search
Relevance is a critical factor in product search as it directly impacts customer satisfaction and conversion rates. If a customer cannot find what they are looking for quickly and accurately, they may leave the site without making a purchase or become frustrated with their shopping experience. This can lead to lost sales opportunities and potential damage to brand reputation.
To ensure high levels of relevance in product search results, retailers like Walmart must continuously refine their algorithms and models that power their search engines.
The Role of Embedding-based Neural Retrieval (EBR)
Embedding-based neural retrieval (EBR) is an AI-powered approach used by many e-commerce companies to enhance relevance in product searches. EBR leverages deep learning techniques to map products into low-dimensional vector spaces based on their attributes such as title, description, images, etc., allowing for more accurate matching between queries and products.
At Walmart specifically, EBR was implemented using state-of-the-art Transformer architecture combined with BERT (Bidirectional Encoder Representations from Transformers) to improve relevance in product search.
The Challenges Faced by EBR at Walmart
While the initial implementation of EBR at Walmart showed promising results, there were still challenges that needed to be addressed. These included:
- False Positives/Negatives in Training Data: The training data used for EBR was not always accurate, leading to false positives and negatives in the model's predictions.
- Handling Query Misspellings: Customers often make typos or misspell words when searching for products, which can lead to irrelevant or no results being returned.
- Limited Relevance Feedback: The amount of human relevance feedback available for training the model was limited, making it challenging to capture all possible variations and nuances in customer queries.
To overcome these challenges, the researchers proposed several approaches to enhance the capabilities of EBR for product search at Walmart.
The Proposed Solutions
The research team introduced a Relevance Reward Model (RRM) based on human relevance feedback as a key contribution towards improving retrieval relevance. This model helps filter out noise from training data and incorporates it into the EBR model through a multi-objective loss function. By incorporating human feedback into the learning process, RRM allows for more accurate and relevant predictions.
Additionally, techniques like typo-aware training and semi-positive generation were employed to further enhance the performance of the EBR model. Typo-aware training involves adding artificial typos into both query and product text during training so that the model learns how to handle them effectively. Semi-positive generation aims to generate additional positive examples by combining existing ones with different attributes or features. Both these strategies aim to improve retrieval accuracy by addressing common issues encountered during product search queries.
Evaluation Results
To validate their proposed solutions' effectiveness, offline relevance evaluation, online AB tests, and successful deployments in live production environments were conducted. The results showed significant improvements in retrieval relevance, with a 5% increase in add-to-cart rates and a 2% increase in conversion rates.
Offline Relevance Evaluation
The researchers evaluated the performance of their proposed solutions using offline relevance evaluation metrics such as Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (NDCG). These measures showed an improvement of up to 10% compared to the baseline model.
Online AB Tests
To further validate their findings, online AB tests were conducted on Walmart's website. The results showed a statistically significant improvement in both click-through rate (CTR) and add-to-cart rate (ATC) for queries that benefited from typo-aware training and semi-positive generation techniques.
Live Production Deployments
Finally, the proposed solutions were successfully deployed in live production environments at Walmart. This allowed for real-time testing and monitoring of the enhancements' impact on customer search experience. The results showed consistent improvements across various metrics, ultimately leading to enhanced relevance and improved customer satisfaction.
Conclusion
In conclusion, this research study showcases how refining the EBR model can lead to significant improvements in retrieval relevance for product search at Walmart. By addressing common challenges faced by EBR models such as false positives/negatives and query misspellings, the proposed solutions have shown promising results through offline evaluations, online AB tests, and successful deployments in live production environments. With these enhancements implemented at scale, customers can expect an even more seamless shopping experience when searching for products on Walmart's website or app.