Uniting Large Language Models and Knowledge Graphs for Enhanced Knowledge Representation
Вставка
- Опубліковано 31 жов 2023
- 10:40 - 11:10 | GEN AI & DATA SCIENCE THEATRE
THE PERFECT COUPLE: UNITING LARGE LANGUAGE MODELS AND KNOWLEDGE GRAPHS FOR ENHANCED KNOWLEDGE REPRESENTATION
WEDNESDAY 20 SEPTEMBER 2023
SPEAKER: KRISTOF NEYS, NEO4J
Large Language models are amazing but are also black-box models that often fail to capture and accurately represent factual knowledge. Knowledge graphs, by contrast, are structural knowledge models that explicitly represent knowledge and, indeed, allow us to detect implicit relationships. In this talk we will demonstrate how LLMs can be improved by Knowledge Graphs, and how LLM’s can augment Knowledge Graphs. A perfect couple! - Наука та технологія
wow, this does get me excited about graph and LLM!
Wonderful presentation. Thank you.
a tutoriel of how this done would be amazing
A good start would to be to understanding GraphDB. Then, you’ll need to learn some ML algorithms that help you quantify features used for classification and generatlizations. It’s basically a quantification of the entities being featured to arrive at a weighted determination of interest, aka. a mathematically calculated guess based upon accumulated observed/measured “facts”. These facts, of course, observed and measured by humans (always subjective) is supposed to be less biased by virtual of the volume of statistical data considered, and thus makes people feel as though their decisions are more ‘justified’ because they are the more ‘commonly’ made eventualities observed; however, these types of determination, by always arriving at the most common probabilities, exclude the types of decisions that created people like Einstein, Musk, or any other exceptional decisions that made large impacts on human’s existence. ;)
@rmcgraw7943 LMAO at putting an idiot like Musk in the same sentense as Einstein.
Video Summary:
结合大型语言模型和知识图谱，以增强知识表示。
- 00:00 这部分视频介绍了知识图谱的定义以及演示了如何将大型语言模型与知识图谱相结合。
- 04:10 在大型语言模型中，语义概念和图谱是非常重要的，其中有两个关键领域，一个是利用大型语言模型来创建机器可读的语义，另一个是通过图机器学习从网络结构中学习事实信息。
- 08:23 通过将知识图谱与大型语言模型相结合，可以提取子图并进行进一步分析，以获得更好的知识表示。
- 12:35 这一部分介绍了如何将知识图谱与大型语言模型结合起来增强知识表示。
- 16:47 为了解决这个问题，我们可以使用知识图谱来强制大型语言模型使用特定的数据集，这个数据集基于你提供的事实知识，从而减少模型的虚构性。
- 21:00 通过使用知识图谱进行基于大型语言模型的知识表示的优势
- 25:11 可以通过将文章转化为向量并进行语义搜索来增强知识表示。
It would be nice to have a link to the slides....maybe with some way to link comment threads to each. I love3 the concept of NLP to the query language. I hate having to learn varying syntax for differing data structures implemented in competing products....I would love to have international standards for all common commands with knowledge bases which could each have linked (graphed) annotations for ongoing issues.
Combining LLMs with decentralized/distributed knowledge graphs would be even better because it provides for greater optionality and prevents vendor lock in. Projects like Origin Trail are already doing this at scale with clients like the British Standards Institute.
Great recommendation!
Great talk but one of the fundamental problems of neo4j is it's lack of scalability to actually handle large data. When your database grows into the hundreds of millions of nodes and relationships it becomes a nightmare to work in