|

楼主 |
发表于 2025-3-29 15:43:18
|
显示全部楼层
+ Z* E- j3 _5 A# c
Can Language Models Solve Graph Problems in Natural Language?
$ [7 ^& {) M: V' H9 C; UHeng Wang*, Shangbin Feng*, Tianxing He, Zhaoxuan Tan, Xiaochuang Han, Yulia Tsvetkov- Y5 B4 K4 n8 X ~# S
Proceedings of NeurIPS, 2023 (spotlight; 3.4% acceptance rate)
6 o4 Y9 N6 c2 X; F6 ^9 q( Ucode / poster
# `) p5 Q* y1 O0 o; rAre language models graph reasoners? We propose the NLGraph benchmark, a test bed for graph-based reasoning designed for language models in natural language. We find that LLMs are preliminary graph thinkers while the most advanced graph reasoning tasks remain an open research question.* T; V+ S' A% B0 X5 B _
1 V' B; D+ q1 N' n7 J1 B9 T$ {1 m
3DSP Detecting Spoilers in Movie Reviews with External Movie Knowledge and User Networks
: X) |: Z5 h" L0 n$ [+ b( uHeng Wang, Wenqian Zhang, Yuyang Bai, Zhaoxuan Tan, Shangbin Feng, Qinghua Zheng, Minnan Luo
( N. A# c& v/ }8 uProceedings of EMNLP, 2023* R& F) n& f: y5 C
code
" `5 ~1 B* T9 i) Z. w5 u& { N' iWe curate a large-scale network-based spoiler detection dataset (LCS), a movie knowledge base (UKM), and propose MVSD, a Multi-View Spoiler Detection framework that takes into account external knowledge and user interaction networks.
" ]( k0 \: H' I4 w
7 V. F, n' F- @- ?3DSP AdaptiveBackdoor: Backdoored Language Model Agents that Detect Human Overseers9 w9 C; v6 ~1 \7 L8 l- L
Heng Wang, Ruiqi Zhong, Jiaxin Wen, Jacob Steinhardt
2 d9 x I6 V+ k, M! lICML 2024 @ NextGenAISafety
' C2 ^+ c9 u2 w6 O/ S9 ~" AWe speculate a new form of cyber attack, where an LM agent is backdoored to detect whether its actions will be overseen by humans and act maliciously when effective oversight is not present, and provide concrete proof-of-concept with AutoGPT.7 H8 W7 }. V1 h4 O4 E7 q3 V
/ g; O/ d2 `7 A
3DSP Can LLM Graph Reasoning Generalize beyond Pattern Memorization?
1 P: m& V- U+ [Yizhuo Zhang*, Heng Wang*, Shangbin Feng*, Zhaoxuan Tan, Xiaochuang Han, Tianxing He, Yulia Tsvetkov , Tianxing He, Yulia Tsvetkov( w! X& t0 C D
EMNLP 2024, findings
+ C& ]3 W; C; r/ Q C3 Ncode/ I. d1 B# h* i5 B
While instruction tuning produces promising graph LLMs, can they generalize beyond patterns in the training data? Mostly no, especially from synthetic to real-world problems, while we explore preliminary solutions. S3 x8 o& q& G: L6 Q" o7 t* \
9 [# s1 G. D; {5 P8 Q% Z( E3DSP Explaining Datasets in Words: Statistical Models with Natural Language Parameters
( s# V* n) m+ K% R3 O- J/ x& l' gRuiqi Zhong, Heng Wang, Dan Klein, Jacob Steinhardt. [9 D2 W( y( C+ X
Proceedings of NeurIPS, 2024
0 \; ~0 }. H& P2 scode
- u( V6 Q' n; o; z! C4 `' LWe build a framework that can use natural language predicates to parameterize a wide range of statistical models, and show that it is versatile, useful, and applicable to both text and vision domains, and explains sophisticated concepts that classical methods struggle to produce.
$ M2 O6 z* s v6 ?: }8 f7 N6 Q) G. Z3 D: ?3 S$ D% ]* w! c" b
3DSP Resolving Knowledge Conflicts in Large Language Models5 Y' V7 n6 \9 M. Y( p
Yike Wang*, Shangbin Feng*, Heng Wang, Weijia Shi, Vidhisha Balachandran , Tianxing He, Yulia Tsvetkov' x4 G6 N) p8 T- D
Proceedings of COLM, 2024
- m$ ~3 {( h _$ J) ecode- F$ [7 S4 Z3 |' t: T+ V. r+ p
We introduce KNOWLEDGE CONFLICT, an evaluation framework for simulating contextual knowledge conflicts and quantitatively evaluating LLMs' abilities to handle knowledge conflicts.
7 Q; _8 E. T( d/ }. u- |1 s; C1 e7 V- M' [$ n% P* i. x
3DSP BotMoE: Twitter Bot Detection with Community-Aware Mixtures of Modal-Specific Experts
* y) f: w% @0 h0 }3 y! X) sYuhan Liu, Zhaoxuan Tan, Heng Wang, Shangbin Feng, Qinghua Zheng, Minnan Luo9 M) V; ~& @, ^6 a
Proceedings of SIGIR 2023./ H9 n% z6 h* z7 Y6 M
code
+ r- I7 F* S, L( |! u6 RWe propose community-aware mixture-of-experts to address two challenges in detecting advanced Twitter bots: manipulated features and diverse communities.
: o0 S& d9 S8 p3 F2 W! j5 ~! y; |1 Q: @2 I
3DSP TwiBot-22: Towards Graph-Based Twitter Bot Detection
9 m" v o% `+ P# jShangbin Feng*, Zhaoxuan Tan*, Herun Wan*, Ningnan Wang*, Zilong Chen*, Binchi Zhang*, Qinghua Zheng, Wenqian Zhang, Zhenyu Lei, Shujie Yang, Xinshun Feng, Qingyue Zhang, Hongrui Wang, Yuhan Liu, Yuyang Bai, Heng Wang, Zijian Cai, Yanbo Wang, Lijing Zheng, Zihan Ma, Jundong Li, Minnan Luo
' r4 \. t) S3 e, f& e( `- n& OProceedings of NeurIPS, Datasets and Benchmarks Track, 2022." k( X/ e+ n8 ?- N" n( ^) F# G H) p
website / GitHub / bibtex / poster
8 @' G: [9 B- C9 X
+ y) | f) t" H" \ C4 b, j西交大四学生的文章,准备去UIUC读PHD |
|