Large language model (LLM) research and applications have advanced remarkably in recent times. These generative models have enthralled the substitute intelligence community, and lots of models trained on various tasks and modalities have been made available recently. A rising agreement has resulted from these developments, stating that LLMs are a crucial step towards artificial general intelligence (AGI). Still, several things might be improved in how LLMs are actually designed and implemented despite all of their advantages. Their dependence on unstructured text is one among their most evident drawbacks since it might probably occasionally lead the models to overlook clear, logical inferences or imagine false conclusions. 

One other is that LLMs have inherent limitations based on the time period they were taught, and it could be difficult to integrate “latest” knowledge about how the world has evolved. One of the adaptable forms of data representation is graph-structured data, which offers a possible treatment for each problems. It’s interesting to notice that only somewhat research has been done on the intersection of graphs and LLMs despite this potential. For example, though graph databases and LLMs have received much attention, more research must be done on the broader applications of graph-structured data. Wang et al. have recently made an effort to unravel this by making a graph benchmarking challenge specifically for language models. 

The removal of varied natural graph challenges and the dearth of variation within the varieties of graph structures addressed make for a lot of unanswered concerns, though their work marks an interesting starting effort into assessing LLM’s graph reasoning skills. Other recent work goals to make use of LLMs as an alternative of graph-structured data, however it ignores a few of the core problems with LLMs. Researchers from Google Research conducted the primary thorough investigation on reasoning over graph-structured data as text that LLMs may read on this paper. They break down the problem into graph prompt engineering and graph encoding to look at graph reasoning in further detail. 

We will use LLM’s acquired representations in graph problems by experimenting with different graph encoding techniques. While researching prompt engineering methods, one may select the perfect approach to ask an LLM the query that they wish to be answered. Their test findings aim to discover the scenarios during which various prompt heuristics perform optimally. To accomplish that, they supply a brand-new set of benchmarks called GraphQA for evaluating the performance of LLM reasoning on graph data. Using graphs with a way more diverse and realistic graph structure than those previously investigated using LLMs sets GraphQA apart. 

Specifically, their work has contributed to the next: 

1. An intensive examination of graph-structure prompting approaches to be used in LLMs. 

2. Best practices and insights for encoding graphs as text for LLM usage. 

3. A brand-new graph benchmark called GraphQA to let the community higher explore how graph structure affects LLM prompting.

Take a look at the Paper. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to affix our 32k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the newest AI research news, cool AI projects, and more.

In case you like our work, you’ll love our newsletter..

We’re also on Telegram and WhatsApp.