Constructs multi-video summarization benchmark and evaluates nine MLLMs showing positional bias is domain- and model-dependent with middle positions often weaker and budgets not uniformly fixing it.
Where is the answer? An empirical study of positional bias for parametric knowledge extraction in language model
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
A Systematic Evaluation of Positional Bias in Multi-Video Summarization with MLLMs
Constructs multi-video summarization benchmark and evaluates nine MLLMs showing positional bias is domain- and model-dependent with middle positions often weaker and budgets not uniformly fixing it.