Ganesh Jagadeesan’s Post

Wondering how the most popular metrics for evaluating Retrieval Augmented Generation systems work? Wonder no longer! This new blog explores precision@K, recall@K, MAP@K, MRR@K, and NDCG@K as evaluation methods for information retrieval systems - how they work, and which are the most popular 🚀 Read more: https://lnkd.in/eYURjwgK Leonie Monigatti is a pro at explaining complicated concepts simply 🫶 Super awesome read

When embarking on a content clustering project, selecting the appropriate algorithm is crucial for achieving desired outcomes. Different algorithms excel in various aspects, such as managing vast datasets or yielding meaningful groupings. Understanding these distinctions is the first step in the decision-making process. For a comprehensive overview, visit https://buff.ly/3YEpqrQ . @[Penfriend](urn:li:organization:100267386)

Unveiling Algorithm Efficiency, Key Insights from Algorithm Analysis

Insight cannot be achieved when comparing and/or managing two discrete ideas, things, or concepts. It is only by understanding the connections between ideas, things, concepts, people, places, organizations, products, governments, agencies, partners, consumers, etc., that we can truly uncover really insightful things about our world, an organization or enterprise. Connecting, linking, relating, and labeling such connections between different things (e.g. a connection between a person and her employer or her spouse or partner, her parents, her grocery store, etc.,) is the raison d'etre of graph databases. 

How many chunks is the right number of chunks? Testing out the contextual retrieval process with The Wiz. https://lnkd.in/gx5EttHH

🔍 New Blog Alert! Dive into the world of online recommendation systems with our latest post on "Dynamic Online Recommendation for Two-Sided Market with Bayesian Incentive Compatibility." Discover how we formalize the challenges of balancing exploration and exploitation, and address users' self-interested behaviors. Our two-stage algorithm promises to revolutionize the field! Read the full post here: https://bit.ly/4eho6R0

Using hybrid search and rerankers with LanceDB is as simple as it gets. In my latest blog, I compare retrieval performance using different search and ranking techniques. I found the results pretty impressive. TLDR- "Reranking over-fetched vector search results with Cohere reranker 𝐠𝐞𝐭𝐬 𝐚𝐫𝐨𝐮𝐧𝐝 𝟖% 𝐚𝐧𝐝 𝟏𝟏% 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐢𝐧𝐜𝐫𝐞𝐚𝐬𝐞 in case of 𝐁𝐆𝐄 𝐚𝐧𝐝 𝐂𝐨𝐥𝐁𝐄𝐑𝐓 embedding models. Hybrid search with Cohere reranker again stands out in both cases as it achieves more than 90% hit-rate. 𝐓𝐡𝐚𝐭 𝐢𝐬 𝐚𝐩𝐩𝐫𝐨𝐱𝐢𝐦𝐚𝐭𝐞𝐥𝐲 𝟐𝟑% 𝐚𝐧𝐝 𝟒𝟗% 𝐢𝐦𝐩𝐫𝐨𝐯𝐞𝐦𝐞𝐧𝐭 𝐨𝐧 𝐭𝐡𝐞 𝐰.𝐫.𝐭 𝐭𝐡𝐞 𝟐 𝐞𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠 𝐟𝐮𝐧𝐜𝐭𝐢𝐨𝐧𝐬." Read the blog to learn more👇 https://lnkd.in/gAh9RrMw

Hiii I wrote a blog post on why we need the next generation of intelligent scrapers, check this out 🤗🤗🤗 https://lnkd.in/dEhukCH2

Low Rank Adaptation is a fundamental tool in being able to fine tune LLMs and serve them at scale. This article describes the theory, and the practice. https://buff.ly/3wzlRrt Author: https://buff.ly/3JVGHEn

In this article, we’ll explore how LLMs are changing recommendation systems as we speak. Let's first start with a brief overview.