Table 1 Clustering consistencies among various configurations of different LLMs

From: Leveraging large language models for academic conference organization

 

text-embedding-ada-002

AoE

Ti

TiAb

TiAbKw

Ti

TiAb

TiAbKw

text-embedding-ada-002

Ti

-

0.2465

0.1803

0.0558

0.0292

0.0659

TiAb

 

-

0.3189

0.0235

0.0637

0.1300

TiAbKw

  

-

0.0271

0.0558

0.1086

AoE

Ti

   

-

0.0062

0.0177

TiAb

    

-

0.1393

TiAbKw

     

-

  1. We experimented with two LLM embedding models (text-embedding-ada-002 based on GPT architecture and angle-optimized text embeddings AoE11 based on LlaMa 2) with different combinations of title (Ti), abstract (Ab) and keywords (Kw) for presentations to be clustered. We calculated the adjusted mutual information (AMI) between the clustering results, score between 0 and 1 reflecting their agreement with each other (higher score indicates more agreement).
Back to article page