notesum.ai

Published at December 10

SST framework for Document Matching

cs.IR

cs.CL

Released Date: December 10, 2024

Authors: Youchao Zhou¹, Heyan Huang¹, Zhijing Wu¹, Yuhang Liu¹, Xinglin Wang¹

Aff.: ¹School of Computer Science and Technology, Beijing Institute of Technology

Arxiv: http://arxiv.org/pdf/2412.07573v1

Refer to caption

Models		CNSE		CNSS
Models		Acc	F-1	Acc	F-1
I	BM25	69.63	66.62	67.66	70.41
I	BERTopic	67.43	65.85	69.28	72.45
II	BERT	83.97	81.67	89.63	90.09
	XLNeT	83.21	81.09	90.22	90.28
	CoLDE	84.47	82.18	89.97	90.42
III	BERT-PLI	77.90	75.73	85.95	86.22
IV	CIG	84.64	82.75	89.77	90.07
	CIG- $Sim_{g}$	84.21	82.46	90.03	90.29
	OPT-Match	83.80	81.88	90.81	90.97
	Match-Ignition	86.32	84.55	91.28	91.39
V	Baichuan- $7B^{z}$	43.75	60.85	50.82	67.37
	GPT- $3.5^{z}$	66.67	59.29	62.66	70.94
	GPT- $4^{z}$	77.14	77.62	80.12	80.72
	Baichuan- $7B^{f}$	84.67	82.56	91.82	92.22
VI	$BERT+SST^{dh}$	86.25	85.10	92.52	92.71
	$BERT+SST^{ds}$	86.91	85.69	92.76	92.90
	$BERT+SST^{ah}$	87.04	85.84	93.01	93.24
	$BERT+SST^{as}$	87.65	86.45	93.45	93.58