The previous article did a deep dive into the prompting-based pointwise, pairwise, and listwise techniques that directly use LLMs to perform reranking. In this article, we will take a closer look at some of the shortcomings of the prompting methods and explore the latest efforts to train ranking-aware LLMs. The article also describes several strategies to build effective and efficient LLM-based rerankers.

Large Language Models (LLMs) have demonstrated impressive zero-shot performance on a wide variety of NLP tasks. Recently, there has been a growing interest in applying LLMs to zero-shot text ranking. This article describes a recent paradigm that uses prompting-based approaches to directly utilize LLMs as rerankers in a multi-stage ranking pipeline.

Auto-regressive search engines emerge as a promising paradigm for next-gen information retrieval systems. This article introduces this generative retrieval and the various latest techniques that have been proposed to improve its effectiveness.

Large Language Models (LLMs) have emerged as viable tools for various recommendation tasks. This article highlights various methods for incorporating users, items, and behavior data into the instructions for LLMs.

In previous articles, I gave an introduction to graph representation learning and highlighted several shallow methods for learning homogeneous graph embeddings. This article focuses on shallow representation learning methods for heterogeneous graphs.

While homogeneous networks have only one type of nodes and edges, heterogeneous networks contain different types of nodes or edges. So, a homogeneous network can also be considered as a special case of a heterogeneous network. Heterogeneous networks, also called heterogeneous information networks (HIN), are ubiquitous in real-world scenarios. For example, social media websites, like Facebook, contain a set of node types, such as users, posts, groups and, tags. By learning heterogeneous graph embeddings, we learn low-dimensional representations of the graph while preserving the heterogeneous structures and semantics for the downstream tasks (such as node classification, link prediction, etc.).

Instruction-tuning methods enable open-source Large Language Models (LLMs) usage for building highly effective recommender systems on private data. This article highlights the latest research work on this paradigm of using LLMs for recommendation tasks.