Building Scalable Retrieval System with Two-Tower Models | Query-Item Retrieval Recsys Model | ML AI

#datascience #machinelearning #artificialintelligence #recsys #recommendations #amazon

Explore the power of Two-Tower Model Architecture in this YouTube video! Discover how this deep learning approach is revolutionizing retrieval tasks for search and recommendation systems. Whether one is in e-commerce, content search, or social networking, Two-Tower Models offer incredible flexibility and scalability.

In this video, we will take a deep dive into the realm of Two-Tower Models, which comprise two distinct towers—one for queries and one for items. These models have the potential to revolutionize various recommendation and search tasks, whether it's finding relevance between users and products, queries and documents, people and people, or users and images/videos, among others. The versatility of the two-tower architecture extends to any pair of entities we wish to analyze.

We will also demonstrate the process of training these neural networks using Amazon's open-source datasets to address query-item retrieval challenges, making them highly adaptable to the specific requirements. Once trained, these models generate embeddings for the entities involved, facilitating the efficient retrieval of relevant items at scale.

In the context of e-commerce, we will discuss the limitations of relying solely on word matching and why it's imperative to take into account other factors and actual purchase behavior to enhance the retrieval of more relevant items. We will explore the concept of fine-tuning embeddings by incorporating additional signals such as category, price, ratings, reviews, and personalized data. This approach significantly boosts the scalability and accuracy of your retrieval system.

In this video, we will go through the process of building a Two-Tower Search-Item Retrieval model from scratch using an open-source Amazon dataset in details, allowing one to witness firsthand how this architecture can create a highly effective and scalable retrieval system. The codes and all the relevant links are provided.

#datascience #machinelearning #statistics #deeplearning #programming #python #datatrek #youtube #interview #interviewpreparation #interviewquestions #datascientist #dataanalytics #machinelearningengineer #datasciencejobs #datasciencetraining #datasciencecourse #datascienceenthusiast #career #careeropportunities #careergrowth #careerdevelopment #datascienceenthusiast #interviewing #ml #ai

#datatrek #datascience #machinelearning #statistics #deeplearning #ai

About DataTrek Series

[kunalpatel]

Learning and applying data science for the last 5 years but never found someone to explain industry-oriented concepts and projects. Appreciate Abhishek for your efforts.

[saankhyamondal]

This is one of the most underrated channels in YT. Such videos help us get an idea about industry use cases. Thank you for sharing.

[neelmishra]

This is the most awesome content, I have every watched. Straight up in league of stanford ai lectures.

[OmkarKulkarni-db]

Not sure if the process to generate the raw embeddings is correct for products. You are creating separate list and then concatenating them as a pandas dataframe. Wouldn't you just do `df.values.tolist()` and just loop over the list of tuples in batches?

[pauloalexandremello]

do you think is adequate adapt Two Towers to a tower for jobs descriptions and another for applicants jobs experiences and create an Applicant Tracking system?

[vyaslkv]

thank you sir for sharing your knowledge :)

[gblazerify]

Great video, I got "ImportError: cannot import name 'is_nltk_available' from error when I ran model.encode part in Kaggle. Ideas? I guess it only works with sentence_transformers 2.2.2, but breaks with 2.5.1