Assessment of the Premise's Influence Period (Research Report) [1]: Models are required to infer the duration of an event's impact on a company's operations.
Labels: < 1 month, 1 to 3 months, 4 to 6 months, 7 to 12 months, and > 1 year
Example:
(1) Given Argument:
Catcher may have a short-term technical rebound especially when July could be a strong month due to ramp for Sony Xperia casings
Label: 1 to 3 months
(2) Given Argument:
We think the high-end smartphone market is nearly saturated, and growth for the next 1-2 years will only be 10%. With intensifying competition from other tier-one players and entrance of lower-tier players, we think a long-term margin downtrend is inevitable for the smartphone market.
Label: > 1 year
Detection of Argument Temporal References (Earnings Conference Call) [2]: Models must identify the temporal tags associated with an argument.
Labels:
0: No time reference
1: Long past (more than half a year)
2: Short past (less than half a year): during this quarter or up to 2 quarters.
Example:
claim_text：
claim: Now the biggest trend that we see in consumer behavior is definitely video.
premise_texts：
premises: ["And there's a strong technological underpinning for that, which is that if you go back five years and you tried to watch a video on your phone, it would probably have to buffer for a minute or so before you'd actually get to watch it, which wasn't a good experience.", "And if you wanted to upload a video, whether it was a longer video like what you'd post to News Feed or a 10-second story like what you'd post in any of the apps, even that might take 30 seconds to upload, so it wasn't a good experience.", "So now as the technology on the network level improves to support that, what we're seeing is the ability to serve what is a large amount of demand for what's a very engaging type of content.", "And that demand flows across social content like we're seeing in Stories and Feed to clearly a huge amount of public content.", 'Pages are engaging in this, and some of the trends that Dave just talked about for just a lot of video behavior across the platform.']
year: 2017
quarter: Q2
label: 0
Assessment of the Claim's Validity Period (Social Media) [3]: Models should predict the validity period of a claim. For instance, the opinion "I think $AAPL will reach $200 today" should not be considered valid after three days from its publication time.
Labels:
"Longer than 1 week", "Within 1 week", "Unsure"
Example:
'text': '看來中華電還是比定存好\n \n 等低點來買一些好了',
'Label_duration': 'Longer than 1 week',
'Label_sentiment': 'positive'
[1] Chin-Yi Lin, Chung-Chi Chen, Hen-Hsen Huang, and Hsin-Hsi Chen. 2024. Argument-Based Sentiment Analysis on Forward-Looking Statements. In Findings of the Association for Computational Linguistics: ACL 2024
[2] Alaa Alhamzeh. Financial Argument Quality Assessment in Earnings Conference Calls. In International Conference on Database and Expert Systems Applications, pp. 65-81. 2023.
[3] Chr-Jr Chiu, Chung-Chi Chen, Hen-Hsen Huang, Hsin-Hsi Chen. 2025. Pre-Finetuning with Impact Duration Awareness for Stock Movement Prediction. In Companion Proceedings of the Web Conference 2025 (WWW'25).
The annotated dataset is licensed under the Creative Commons Attribution-Non-Commercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.