Not All or None: Dynamic Construction of Target-aware Memory Graph for Conversational Stance Detection

Yifan Xiang; Bin Liang; Yuqi Huang; Ruifeng Xu; Kam-Fai Wong

cs.CLAug 29, 2026

Not All or None: Dynamic Construction of Target-aware Memory Graph for Conversational Stance Detection

Authors: Yifan Xiang, Bin Liang, Yuqi Huang, Ruifeng Xu, Kam-Fai Wong

Organizations: The Chinese University of Hong Kong, Hong Kong, China · MoE Key Laboratory of High Confidence Software Technologies, China · The Hong Kong Polytechnic University, Hong Kong, China · Harbin Institute of Technology, Shenzhen, China

Abstract

Stance detection is crucial for understanding the underlying attitude of an expression towards a target. Conversational stance detection is a more challenging stance detection task in real-world social media scenarios, as it involves detecting the user's stance by leveraging the target-related historical statements across conversational sessions. In this paper, we propose target-aware Memory Graph TamGraph, a novel method that dynamically leverages target-related statements for conversational stance detection. Instead of considering all preceding historical conversations or using no prior conversation information for stance detection, our TamGraph employs a stepwise, entropy-guided backtracking mechanism to selectively activate memory from historical conversations and dynamically constructs a target-aware graph to model the stance relations among utterances. This allows the exploitation of target-related information from the conversation history for stance detection while preventing the introduction of noise. Experimental results on both English and Chinese benchmarks demonstrate that our TamGraph substantially improves LLM performance on conversational stance detection.

Figures & tables

Figure 1: The comparison between our method and other prompting baselines.

Model	MT-CSD (acc. / F1 in %)						ZS-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall	ZS-CSD (acc. / F1 in %)
Qwen2.5-3B (single)	47.03/42.96	47.22/43.78	46.11/46.77	31.39/30.02	48.84 /34.43	44.66/42.88	33.55/25.53
Qwen2.5-3B (multi)	48.69 /43.80	37.41/35.26	41.74/40.98	34.31/ 34.69	38.15/28.49	40.19/39.13	31.11/20.77
Qwen2.5-3B (CoT)	49.44 / 49.29	49.63 / 44.15	49.22 / 48.32	42.58 / 43.29	46.70 / 37.49	47.62 / 45.59	35.41 / 33.91
Qwen2.5-3B (Ours)	46.09/ 45.88	62.22 / 54.62	64.80 / 61.00	45.50 /33.14	56.51 / 43.16	54.66 / 50.74	39.51 / 34.68
Qwen2.5-7B (single)	44.98/42.32	70.91 / 58.24	64.17 /57.37	45.74/31.76	66.67/51.15	58.58/51.32	39.43/28.53

Table 1: Comparison of single-turn, multi-turn, CoT, and our method on MT-CSD and ZS-CSD. On ZS-CSD, we exclude Llama-3.1-8B-Instruct due to its limited native support for Chinese. For each model block and each column, the best (second-best) value for accuracy and macro F1 is shown in bold ( underline ).

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B (BM25, q=target)	45.35 /40.26	44.26/38.63	47.98/49.16	36.01/32.37	49.55/35.58	44.83/41.51
Qwen2.5-3B (BM25, q=utterance)	44.98/ 40.42	49.81 / 42.94	52.96 / 52.91	39.90 / 34.00	51.34 / 37.06	47.79 / 44.17
Qwen2.5-3B (BM25, q=u+t)	43.87/39.08	48.15/40.27	51.40/51.80	37.71/33.70	49.38/35.56	46.10/42.65
Qwen2.5-3B (Ours)	46.09 / 45.88	62.22 / 54.62	64.80 / 61.00	45.50 / 33.14	56.51 / 43.16	54.66 / 50.74
Qwen2.5-7B (BM25, q=target)	46.65 / 45.95	63.89/56.26	63.86 / 60.85	49.15 / 41.28	66.67/47.93	58.08/54.36

Table 2: Detailed results of BM25 retrieval baselines with different query modes on MT-CSD for five models, compared with our method. For each model block, we report three query variants and our method.

Figure 2: The comparison between the performance of our method on Qwen2.5-14B-Instruct and Qwen3-4B-Instruct-2507 and GPT-4o-mini. The detailed results are presented in Table 12 in the Appendix.

Figure 3: Ablation studies on the entropy threshold vs. performance across five models. The horizontal axis denotes the entropy threshold, and the vertical axis shows the overall weighted accuracy and macro F1.

Figure 4: Ablation studies on the entropy threshold vs. token usage across five models. The horizontal axis denotes the entropy threshold, and the vertical axis shows the average number of token usage per sample.

Figure 5: The comparison between the performance of our method and the “w/o entropy threshold”.

Figure 6: Qualitative example of our method and CoT baseline.

Appendix figures & tables20 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Label	Proportion	Precision	Recall	F1
MT-CSD	favor	19.65%	52.37	33.69	41.00
MT-CSD	against	30.54%	53.04	60.58	56.56
MT-CSD	none	49.81%	62.84	66.20	64.47
ZS-CSD	favor	34.87%	56.60	11.54	19.17
ZS-CSD	against	29.06%	35.68	43.81	39.33
ZS-CSD	none	36.07%	45.70	72.48	56.05

Appendix

Table 3: Pooled per-label performance of TamGraph across the evaluated model backbones.

Dataset	# Models	CoT	TamGraph	Speedup
MT-CSD	5	$\sim$ 33.0	$\sim$ 27.0	1.22 $\times$
ZS-CSD	4	$\sim$ 29.0	$\sim$ 24.0	1.21 $\times$
Overall	–	$\sim$ 31.0	$\sim$ 25.5	1.22 $\times$

Appendix

Table 4: Dataset-level end-to-end latency in seconds per example. The overall row averages the two dataset-level results.

Dataset / Target	# Instances	Avg. Turns	Irrelevant Utterances	Irrelevant History Turns
MT-CSD / Biden	411	4.79	60.75%	54.86%
MT-CSD / Bitcoin	538	4.65	27.64%	25.69%
MT-CSD / SpaceX	321	4.52	17.03%	15.15%
MT-CSD / Tesla	540	4.86	22.18%	18.80%
MT-CSD / Trump	561	4.91	24.69%	22.44%
MT-CSD Overall	2,371	4.77	30.04%	27.02%

Appendix

Table 5: Dataset-level analysis of target-irrelevant conversational content. History turns exclude the final utterance whose stance is evaluated.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B (w/o entropy threshold)	45.35/45.48	52.59/48.09	58.26/56.14	45.26/ 39.97	49.55/39.29	49.73/47.97
Qwen2.5-3B (Ours)	46.09 / 45.88	62.22 / 54.62	64.80 / 61.00	45.50 /33.14	56.51 / 43.16	54.66 / 50.74
Qwen2.5-7B (w/o entropy threshold)	48.33 / 48.52	60.93/55.26	63.86/60.13	47.93/ 40.45	64.35/50.94	57.02/55.23
Qwen2.5-7B (Ours)	43.68/41.06	72.22 / 70.75	66.98 / 66.79	48.18 /37.92	70.05 / 70.14	60.35 / 57.68
Qwen2.5-14B (w/o entropy threshold)	58.18 / 57.16	58.33/54.69	61.37/59.53	44.28/ 43.58	55.26/42.24	55.55/55.32

Appendix

Table 6: Detailed results of ablation study on the “w/o entropy threshold” on MT-CSD for five models and the comparison with our method. For each group, the best value for accuracy and macro F1 is shown in bold.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B (random turn selection)	46.84 / 46.82	55.56/50.64	57.01/54.90	43.55/ 36.29	49.73/39.10	50.32/48.15
Qwen2.5-3B (Ours)	46.09/45.88	62.22 / 54.62	64.80 / 61.00	45.50 /33.14	56.51 / 43.16	54.66 / 50.74
Qwen2.5-7B (random turn selection)	45.17 / 45.41	65.00/58.54	64.80/60.70	47.45/ 38.44	63.46/49.35	57.06/54.53
Qwen2.5-7B (Ours)	43.68/41.06	72.22 / 70.75	66.98 / 66.79	48.18 /37.92	70.05 / 70.14	60.35 / 57.68
Qwen2.5-14B (random turn selection)	52.79 / 52.26	60.00/56.62	60.44/58.33	46.23/ 42.39	57.58/44.08	55.46/55.02

Appendix

Table 7: Detailed results of ablation study on the “random turn selection” on MT-CSD for five models and the comparison with our method. For each group, the best value for accuracy and macro F1 is shown in bold.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B	44.80/44.73	60.37/54.30	63.86/61.16	46.72/34.94	55.44/42.93	53.77/50.55
Qwen2.5-7B	44.61/42.98	72.78/62.50	69.16/62.97	47.20/33.60	70.05/55.33	60.82/54.94
Qwen2.5-14B	50.56/50.14	71.67/66.47	64.49/62.28	48.18/39.29	62.21/46.86	59.60/57.25
Qwen3-4B-2507	46.65/46.73	68.89/61.26	65.11/60.62	48.91/34.88	69.16/52.76	59.93/56.33
Llama-3.1-8B	50.93/49.26	62.96/58.93	53.58/52.78	45.01/38.87	59.36/46.85	55.00/53.96

Appendix

Table 8: Detailed results of ablation study on the entropy threshold (

\text{entropy}=0.0

) on MT-CSD for five models.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B	44.42/44.30	60.74/53.97	62.62/59.28	46.47/34.18	56.51/43.51	53.82/50.23
Qwen2.5-7B	43.31/41.05	72.41/60.80	67.91/60.09	48.18/32.39	70.94/52.37	60.65/53.35
Qwen2.5-14B	46.10/45.87	73.52/64.53	68.22/64.64	45.74/33.53	67.74/53.73	60.40/56.66
Qwen3-4B-2507	45.91/45.77	70.56/59.83	66.98/60.30	46.23/31.10	69.70/50.93	60.06/54.89
Llama-3.1-8B	50.74/49.09	63.15/59.49	53.58/52.61	43.80/37.65	59.71/47.32	54.87/53.71

Appendix

Table 9: Detailed results of ablation study on the entropy threshold (

\text{entropy}=0.2

) on MT-CSD for five models.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B	43.87/43.74	61.11/54.23	63.24/59.23	46.23/33.93	55.97/42.95	53.69/49.97
Qwen2.5-7B	43.31/40.79	71.67/59.64	67.91/59.72	47.93/31.71	71.30/52.73	60.52/52.85
Qwen2.5-14B	46.65/46.35	73.33/63.90	68.85/65.17	46.47/33.75	67.74/51.42	60.69/56.69
Qwen3-4B-2507	45.54/45.46	70.37/59.82	66.98/60.12	46.47/31.24	69.34/50.67	59.89/54.72
Llama-3.1-8B	51.12/49.55	62.41/58.28	53.27/52.36	43.07/37.02	59.54/47.66	54.58/53.47

Appendix

Table 10: Detailed results of ablation study on the entropy threshold (

\text{entropy}=0.4

) on MT-CSD for five models.

Model	MT-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	Overall
Qwen2.5-3B	45.17/45.05	62.78/55.08	64.49/60.19	44.53/33.08	58.11/44.64	54.74/50.96
Qwen2.5-7B	43.31/40.33	70.93/58.52	66.36/57.45	47.69/30.73	71.30/52.68	60.10/51.59
Qwen2.5-14B	47.58/47.02	73.15/63.44	71.34/66.95	46.96/33.67	67.38/51.63	61.20/56.92
Qwen3-4B-2507	46.28/46.04	68.89/58.71	67.60/60.52	46.23/31.58	68.63/50.19	59.60/54.43
Llama-3.1-8B	51.30/49.53	60.00/55.65	54.83/53.92	41.12/35.81	58.29/44.75	53.65/52.35

Appendix

Table 11: Detailed results of ablation study on the entropy threshold (

\text{entropy}=0.8

) on MT-CSD for five models.

Model	MT-CSD (acc. / F1 in %)					ZS-CSD (acc. / F1 in %)
Model	Bitcoin	Tesla	SpaceX	Biden	Trump	ZS-CSD (acc. / F1 in %)
GPT-4o-mini (single)	49.07/48.33	71.48/67.73	61.99/59.38	44.28/34.61	66.49/53.72	43.65/38.94
GPT-4o-mini (multi)	53.53/53.48	69.26/62.61	66.67/64.86	48.18/39.96	68.81/51.96	45.99/42.70

Appendix

Table 12: The detailed results of GPT-4o-mini under single-turn and multi-turn inputs on MT-CSD and ZS-CSD.

Figure 7: Additional qualitative example of our method and CoT baseline. The CoT baseline provides garbled text in some cases.

Figure 8: Additional qualitative example of our method and CoT baseline.

Panel of Table 13 — Table 13: Prompt of single-turn input stance detection evaluation.

Panel of Table 14 — Table 14: Prompt of multi-turn input stance detection evaluation.

Panel of Table 15 — Table 15: Prompt of CoT baseline stance detection evaluation.

Panel of Table 16 — Table 16: Prompt of Relevance Detection.

Panel of Table 17 — Table 17: Prompt of Target-related Proposition Extraction.

Panel of Table 18 — Table 18: Prompt of Relation Induction.

Panel of Table 19 — Table 19: Prompt of Natural Language Relation Rephraser.

Panel of Table 20 — Table 20: Prompt of Stance Detection of TamGraph.

Stance Detection

Stance detection identifies whether people support, oppose, or remain neutral toward a target, claim, or position in text and conversation.

See all 36 papers

Contextualized Prompting For Stance Detection On Social Media
Jun 4, 2026Tilman Beck, Shakib Yazdani, Simon Kruschinski +2Social Media Analysis Stance Detection
Institute of Intensive Care, University Hospital of Zurich and University of Zurich · Institute of Computer Science, University of Goettingen · GESIS Leibniz Institute for the Social Sciences +2
StanceEval 2026: The Second Stance Detection Shared Task
Oct 2, 2026Rasha Albalawi, Nuha Albadi, Hamzah Luqman +6Stance Detection Cross-Domain Generalization
Information and Computer Science Department, KFUPM, Saudi Arabia · SDAIA–KFUPM Joint Research Center for Artificial Intelligence, Saudi Arabia · HUMAIN, Saudi Arabia
StanceMoE: Mixture-of-Experts Architecture for Stance Detection
Apr 1, 2026Abdullah Al Shafi, Md. Milon Islam, Sk. Imran Hossain +1Mixture-of-Experts Models
Department of Computer Science and Engineering, Khulna University of Engineering & Technology

Jun 4, 2026J/K move · Enter open · S save

Contextualized Prompting For Stance Detection On Social Media

Tilman Beck, Shakib Yazdani, Simon Kruschinski, Marcus Maurer+1

Institute of Intensive Care, University Hospital of Zurich and University of Zurich · Institute of Computer Science, University of Goettingen · GESIS Leibniz Institute for the Social Sciences+2

Social Media Analysis Stance Detection Zero-Shot Text Classification LLM Prompting

Open paper AskPDF arXiv

Stance detection on social media is challenging due to short, noisy, and context-dependent language. While large language models (LLMs) show zero-shot generalization, they are typically prompted without contextual information, which limits their ability to interpret ambiguous posts. In this work, we systematically investigate the impact of incorporating real-world (e.g., user biographies), derived (e.g., political party), and LLM-generated (e.g., target descriptions) contextual features into zero-shot prompting for stance detection on Twitter. Our evaluation spans four benchmark datasets, including a new high-quality German Twitter stance dataset. Across multiple LLMs, we find that integrating contextual information improves performance, but only under specific conditions. LLM-generated target descriptions consistently enhance accuracy, while other user metadata has mixed or even detrimental effects. Notably, we show that the inclusion of other tweets by the same user, often beneficial in supervised learning, can impair performance due to input noise. Our qualitative analysis reveals that LLMs struggle to distinguish task-specific useful information from irrelevant context. Our findings highlight both the promise and challenges of prompting with context information in noisy real-world settings. We publish code and data at this page.

Abstract

Stance Detection

Explore similar work

Contextualized Prompting For Stance Detection On Social Media

StanceEval 2026: The Second Stance Detection Shared Task

StanceMoE: Mixture-of-Experts Architecture for Stance Detection