pith. sign in

arxiv: 2510.11233 · v3 · submitted 2025-10-13 · 💻 cs.CL

CNSocialDepress: A Chinese Social Media Dataset for Depression Risk Detection and Structured Analysis

Jinyuan Xu , Tian Lan , Xintao Yu , Xue He , Hezhi Zhang , Ying Wang , Pierre Magistry , Mathieu Valette
show 1 more author
Lei Li
This is my paper

Pith reviewed 2026-05-18 08:00 UTC · model grok-4.3

classification 💻 cs.CL
keywords depression detectionChinese social mediadatasetpsychological attributesNLPrisk detectionmental health
0
0 comments X

The pith

CNSocialDepress supplies Chinese social media posts with both binary depression risk labels and structured multidimensional psychological attributes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents CNSocialDepress, a dataset containing 44,178 posts from 233 users with expert annotations on 10,306 depression-related segments. It supplies binary risk labels together with detailed psychological attributes for each segment, moving beyond simple yes-or-no classification. This design supports interpretable, fine-grained examination of depressive signals in Chinese-language content. Experiments confirm the dataset works for tasks such as psychological profiling and adapting large language models to detect depression risk, offering direct value for mental health work aimed at Chinese speakers.

Core claim

CNSocialDepress provides binary risk labels along with structured, multidimensional psychological attributes, enabling interpretable and fine-grained analyses of depressive signals.

What carries the argument

The expert-annotated depression-related segments that carry both binary risk labels and multidimensional psychological attributes for each post.

If this is right

  • The dataset enables structured psychological profiling on Chinese social media content.
  • It supports fine-tuning large language models specifically for depression detection.
  • It supplies concrete material for mental health applications aimed at Chinese-speaking populations.
  • Evaluations demonstrate its effectiveness for both risk identification and detailed psychological analysis.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Pairing CNSocialDepress with English depression datasets could expose language-specific ways depression appears in online text.
  • Testing whether the structured attributes improve model explanations over binary labels alone would show their practical benefit.
  • Extending the annotation scheme to other mental health conditions on the same platform could create comparable resources quickly.

Load-bearing premise

Expert annotations of depression-related segments are accurate and consistent enough to support reliable fine-grained psychological profiling and model training.

What would settle it

Finding low agreement among psychological experts on the assigned multidimensional attributes or showing that binary labels alone match the performance of the full structured dataset on downstream detection tasks would undermine the added value of the annotations.

Figures

Figures reproduced from arXiv: 2510.11233 by Hezhi Zhang, Jinyuan Xu, Lei Li, Mathieu Valette, Pierre Magistry, Tian Lan, Xintao Yu, Xue He, Ying Wang.

Figure 1
Figure 1. Figure 1: Dataset Construction Process: During the data annotation process, we used a subset of the original SWDD [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example entry from the CNSD-Gold dataset [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Module II Prompt [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
read the original abstract

Depression is a pressing global public health issue, yet publicly available Chinese-language resources for depression risk detection remain scarce and largely focus on binary classification. To address this limitation, we release CNSocialDepress, a benchmark dataset for depression risk detection on Chinese social media. The dataset contains 44,178 posts from 233 users; psychological experts annotated 10,306 depression-related segments. CNSocialDepress provides binary risk labels along with structured, multidimensional psychological attributes, enabling interpretable and fine-grained analyses of depressive signals. Experimental results demonstrate the dataset's utility across a range of NLP tasks, including structured psychological profiling and fine-tuning large language models for depression detection. Comprehensive evaluations highlight the dataset's effectiveness and practical value for depression risk identification and psychological analysis, thereby providing insights for mental health applications tailored to Chinese-speaking populations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper introduces CNSocialDepress, a benchmark dataset for depression risk detection on Chinese social media consisting of 44,178 posts from 233 users. Psychological experts annotated 10,306 depression-related segments with binary risk labels and structured, multidimensional psychological attributes. The authors demonstrate the dataset's utility through experiments on structured psychological profiling and fine-tuning large language models for depression detection, claiming it enables interpretable and fine-grained analyses of depressive signals in Chinese-language contexts.

Significance. If the annotations are reliable, the dataset addresses a clear scarcity of Chinese-language resources that go beyond binary classification, providing multidimensional attributes that could support more nuanced mental health NLP research and applications. The scale (over 10k annotated segments) and experimental demonstrations of utility for profiling and LLM fine-tuning represent concrete strengths for a dataset release paper.

major comments (1)
  1. [Annotation and Data Construction] The central claim that the dataset 'enables interpretable and fine-grained analyses of depressive signals' (Abstract) rests on the reliability of the expert annotations for multidimensional psychological attributes. However, the manuscript provides no inter-annotator agreement statistics, no detailed coding scheme or guidelines for the specific psychological dimensions, and no validation against external criteria. This directly weakens support for reproducible fine-grained profiling and model training.
minor comments (2)
  1. [Abstract] The abstract refers to 'comprehensive evaluations' and 'practical value' without specifying the exact tasks, baselines, or quantitative metrics used in the experiments; adding these details would improve clarity.
  2. [Dataset Description] Consider adding a table summarizing the exact psychological dimensions/attributes and their distributions to make the structured aspect of the dataset more immediately accessible to readers.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback, which helps us strengthen the presentation of our annotation methodology. We address the major comment point by point below and will revise the manuscript to improve clarity and reproducibility.

read point-by-point responses

  1. Referee: [Annotation and Data Construction] The central claim that the dataset 'enables interpretable and fine-grained analyses of depressive signals' (Abstract) rests on the reliability of the expert annotations for multidimensional psychological attributes. However, the manuscript provides no inter-annotator agreement statistics, no detailed coding scheme or guidelines for the specific psychological dimensions, and no validation against external criteria. This directly weakens support for reproducible fine-grained profiling and model training.

    Authors: We agree that inter-annotator agreement statistics, a detailed coding scheme, and discussion of validation are important for supporting claims of interpretability and reproducibility. The annotations were performed by licensed psychological experts using a structured coding manual that defines each multidimensional attribute (e.g., symptom indicators, cognitive patterns, and risk factors) with explicit criteria and examples. In the revised manuscript, we will add a new subsection on the annotation protocol that includes (1) inter-annotator agreement metrics computed on a double-annotated sample, (2) key excerpts from the coding guidelines, and (3) an explanation of how the binary risk labels were derived from the structured attributes. Regarding external validation, ethical and privacy constraints on social media data precluded access to clinical records; the labels instead reflect expert consensus aligned with established psychological frameworks. These additions will directly address the concern and better support use of the dataset for fine-grained profiling and model training. revision: yes

Circularity Check

0 steps flagged

Dataset release paper with no derivations or predictions exhibits no circularity

full rationale

The paper is a dataset release paper whose central contribution is the CNSocialDepress resource itself (44,178 posts, 10,306 expert-annotated segments, binary risk labels plus multidimensional attributes). No mathematical derivations, equations, fitted parameters, or predictions are claimed. The load-bearing steps are the data collection and annotation process, which are presented as direct contributions rather than derived quantities that reduce to inputs by construction. No self-citations, uniqueness theorems, or ansatzes are invoked in a load-bearing way. The paper is self-contained against external benchmarks as a resource paper; annotation quality concerns (e.g., missing IAA metrics) are validity issues, not circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are invoked; the work rests on standard practices of data collection and expert annotation whose details are not provided in the abstract.

pith-pipeline@v0.9.0 · 5692 in / 996 out tokens · 43291 ms · 2026-05-18T08:00:25.885929+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

72 extracted references · 72 canonical work pages · 2 internal anchors

  1. [1]

    Qwen Technical Report

    Machine learning algorithms for depression: diagnosis, insights, and research directions.Electron- ics, 11(7):1111. Falwah Alhamed, Julia Ive, and Lucia Specia. 2024. Classifying social media users before and after de- pression diagnosis via their language usage: A dataset and study. InProceedings of the 2024 Joint International Conference on Computationa...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  2. [2]

    InProceedings of the First ACL Workshop on Ethics in Natural Language Process- ing, pages 94–102, Valencia, Spain

    Ethical research protocols for social media health research. InProceedings of the First ACL Workshop on Ethics in Natural Language Process- ing, pages 94–102, Valencia, Spain. Association for Computational Linguistics. André Bittar, Sumithra Velupillai, Angus Roberts, and Rina Dutta. 2019. Text classification to inform sui- cide risk assessment in electro...

    work page arXiv 2019

  3. [3]

    Zhihua Guo, Nengneng Ding, Minyu Zhai, Zhenwen Zhang, and Zepeng Li

    Identifying suicide ideation and suicidal at- tempts in a psychiatric clinical research database us- ing natural language processing.Scientific reports, 8(1):7426. Zhihua Guo, Nengneng Ding, Minyu Zhai, Zhenwen Zhang, and Zepeng Li. 2023. Leveraging domain knowledge to improve depression detection on chi- nese social media.IEEE Transactions on Computa- ti...

    work page 2023

  4. [4]

    InProceedings of the Seventh Workshop on Computational Linguistics and Clinical Psychology: Improving Access, pages 15–24, Online

    On the state of social media data for men- tal health research. InProceedings of the Seventh Workshop on Computational Linguistics and Clinical Psychology: Improving Access, pages 15–24, Online. Association for Computational Linguistics. Khan Md Hasib, Md Rafiqul Islam, Shadman Sakib, Md Ali Akbar, Imran Razzak, and Moham- mad Shafiul Alam. 2023. Depressi...

    work page 2023

  5. [5]

    Md Rafiqul Islam, Muhammad Ashad Kabir, Ashir Ahmed, Abu Raihan M Kamal, Hua Wang, and An- waar Ulhaq

    Psycollm: Enhancing llm for psycho- logical understanding and evaluation.Preprint, arXiv:2407.05721. Md Rafiqul Islam, Muhammad Ashad Kabir, Ashir Ahmed, Abu Raihan M Kamal, Hua Wang, and An- waar Ulhaq. 2018. Depression detection from so- cial network data using machine learning techniques. Health information science and systems, 6:1–12. Yoon Kim. 2014. ...

    work page arXiv 2018

  6. [6]

    Tin Lai, Yukun Shi, Zicong Du, Jiajie Wu, Ken Fu, Yichao Dou, and Ziqi Wang

    The phq-9: validity of a brief depression sever- ity measure.Journal of general internal medicine, 16(9):606–613. Tin Lai, Yukun Shi, Zicong Du, Jiajie Wu, Ken Fu, Yichao Dou, and Ziqi Wang. 2023. Psy-llm: Scal- ing up global mental health psychological services with ai-based large language models.arXiv preprint arXiv:2307.11991. Xiaochong Lan, Yiming Che...

    work page arXiv 2023

  7. [7]

    Long Short-Term Memory Based Recurrent Neural Network Architectures for Large Vocabulary Speech Recognition

    Long short-term memory based recurrent neu- ral network architectures for large vocabulary speech recognition.arXiv preprint arXiv:1402.1128. Wesley Ramos dos Santos, Rafael Lage de Oliveira, and Ivandré Paraboni. 2024. Setembrobr: a social media corpus for depression and anxiety disorder prediction. Language Resources and Evaluation, 58(1):273–300. Jürge...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    InIJCAI, pages 3838–3844

    Depression detection via harvesting social media: A multimodal dictionary learning solution. InIJCAI, pages 3838–3844. Tiancheng Shen, Jia Jia, Guangyao Shen, Fuli Feng, Xiangnan He, Huanbo Luan, Jie Tang, Thanassis Tiropanis, Tat Seng Chua, and Wendy Hall. 2018. Cross-domain depression detection via harvesting so- cial media. InProceedings of the Interna...

    work page arXiv 2018

  9. [9]

    InProceedings of the Thirteenth Language Resources and Evalua- tion Conference, pages 2682–2692, Marseille, France

    MentSum: A resource for exploring summa- rization of mental health online posts. InProceedings of the Thirteenth Language Resources and Evalua- tion Conference, pages 2682–2692, Marseille, France. European Language Resources Association. Matthew Squires, Xiaohui Tao, Soman Elangovan, Raj Gururajan, Xujuan Zhou, U Rajendra Acharya, and Yuefeng Li. 2023. De...

    work page arXiv 2023

  10. [10]

    Xingwei Yang, Rhonda McEwen, Liza Robee Ong, and Morteza Zihayat

    Fine-grained depression analysis based on chinese micro-blog reviews.Information Processing & Management, 58(6):102681. Xingwei Yang, Rhonda McEwen, Liza Robee Ong, and Morteza Zihayat. 2020. A big data analytics frame- work for detecting user-level depression from social networks.International Journal of Information Man- agement, 54:102141. Andrew Yates,...

    work page 2020

  11. [11]

    InProceedings of the 62nd Annual Meeting of the Association for Compu- tational Linguistics (Volume 3: System Demonstra- tions), Bangkok, Thailand

    Llamafactory: Unified efficient fine-tuning of 100+ language models. InProceedings of the 62nd Annual Meeting of the Association for Compu- tational Linguistics (Volume 3: System Demonstra- tions), Bangkok, Thailand. Association for Computa- tional Linguistics. Han-yu Zhou, Wen-qi Zhu, Wen-yi Xiao, Ya-ting Huang, Kang Ju, Hong Zheng, and Chao Yan. 2023. F...

    work page 2023

  12. [12]

    Also pay attention to the user ’s depressive state, ensuring that each dimension’s expla- nation is consistent with this state

    Carefully read the complete user Weibo text below and extract key information across six specified dimensions according to the content. Also pay attention to the user ’s depressive state, ensuring that each dimension’s expla- nation is consistent with this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For ea...

    work page

  13. [13]

    Also con- sider the user’s depressive status, ensuring that each dimension’s description matches this state

    Read the entire user Weibo text below and, based on its content, extract core information under the following six dimensions. Also con- sider the user’s depressive status, ensuring that each dimension’s description matches this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if relevant inf...

    work page

  14. [14]

    Pay close at- tention to the user’s depressive state, ensur- ing that the explanations are consistent with it

    Read through the complete user Weibo text below and extract key information across six dimensions based on its content. Pay close at- tention to the user’s depressive state, ensur- ing that the explanations are consistent with it. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If relevant evidence is found in a dime...

    work page

  15. [15]

    At the same time, take into account the user’s depressive state, en- suring that explanations for each dimension align with this state

    Carefully read the full Weibo text below and extract key evidence across six dimensions according to the text. At the same time, take into account the user’s depressive state, en- suring that explanations for each dimension align with this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if ...

    work page

  16. [16]

    At the same time, pay attention to the user’s de- pressive state, ensuring that your explanations match this state

    Please read the full Weibo text below and ex- tract key information from six dimensions. At the same time, pay attention to the user’s de- pressive state, ensuring that your explanations match this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if evidence is detected, output the text inde...

    work page

  17. [17]

    Pay attention to the user’s depressive state and ensure that your descriptions are consistent with it

    Read the complete Weibo text below carefully and extract key information across six dimen- sions based on the text. Pay attention to the user’s depressive state and ensure that your descriptions are consistent with it. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If relevant evi- dence is found, provide the text i...

    work page

  18. [18]

    Pay atten- tion to the user’s depressive state, ensuring that each explanation is aligned with this state

    Please read through the complete user Weibo text below and extract key information across six dimensions based on its content. Pay atten- tion to the user’s depressive state, ensuring that each explanation is aligned with this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimen- sion, if evidence ex...

    work page

  19. [19]

    [Input]User depressive state: {label}, Full Weibo text: {text}

    Read the Weibo text below and extract key evidence across six dimensions according to the content, while considering the user’s de- pressive state to ensure explanations are con- sistent. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If a dimension con- tains relevant evidence, provide the text index (e.g., text_xx...

    work page

  20. [20]

    Pay attention to the user’s depressive state, ensuring that explanations match this state

    Read the complete Weibo text carefully and extract key evidence across six dimensions based on the text. Pay attention to the user’s depressive state, ensuring that explanations match this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if relevant evidence is found, output the text index (...

    work page

  21. [21]

    Pay attention to the user’s depressive state and ensure that the descriptions are consistent with this state

    Read through the entire Weibo text and ex- tract core information across six dimensions according to the content. Pay attention to the user’s depressive state and ensure that the descriptions are consistent with this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if relevant evidence exist...

    work page

  22. [22]

    Ensure that the ex- planations for each dimension align with the depressive state given

    Please read the user Weibo text below com- pletely and extract essential information across six dimensions. Ensure that the ex- planations for each dimension align with the depressive state given. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If relevant evi- dence is detected, list the text index (e.g., text_xx, 文...

    work page

  23. [23]

    Pay attention to the user’s depressive state, ensuring that outputs remain consistent with this state

    Read carefully the full Weibo text provided and extract key evidence from six dimensions. Pay attention to the user’s depressive state, ensuring that outputs remain consistent with this state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If a dimension con- tains evidence, output the text index (e.g., text_xx, 文本_...

    work page

  24. [24]

    Ensure the notes correspond with the user’s depressive condition

    Please go through the complete Weibo text below and identify critical information across six dimensions. Ensure the notes correspond with the user’s depressive condition. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- Where evidence is found, list the text index (e.g., text_xx, 文本 _xx) with a short explanation (e.g....

    work page

  25. [25]

    Make sure the explanation for each aligns with the depressive state

    Read carefully the following Weibo text and extract evidence across six preset dimensions. Make sure the explanation for each aligns with the depressive state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If evidence exists for a dimension, output the text index (e.g., text_xx, 文本_xx) and a short remark (e.g., tex...

    work page

  26. [26]

    Ensure the output is consistent with the depressive state specified

    Read through the following Weibo text and extract major evidence across six given dimen- sions. Ensure the output is consistent with the depressive state specified. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If evidence is found, provide the text index (e.g., text_xx,文本_xx) with a brief note (e.g., text_xx[hopel...

    work page

  27. [27]

    All notes must align with the de- pressive condition of the user

    Please carefully review the Weibo text below and extract key information from six preset dimensions. All notes must align with the de- pressive condition of the user. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If there is evidence, output the corresponding text index (e.g., text_xx, 文本_xx) and a short remark (e....

    work page

  28. [28]

    Make sure your notes are consistent with the depressive state provided

    Please go through the complete Weibo text and extract the critical information across six given dimensions. Make sure your notes are consistent with the depressive state provided. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- For each dimension, if evidence is found, provide the text index (e.g., text_xx, 文本_xx) wi...

    work page

  29. [29]

    Keep the explanations aligned with the depressive state

    Carefully read the Weibo text provided and ex- tract important evidence across six specified dimensions. Keep the explanations aligned with the depressive state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If a dimension contains evidence, state the text index (e.g., text_xx, 文本_xx) and a short remark (e.g., text...

    work page

  30. [30]

    En- sure each dimension ’s note is consistent with the depressive state given

    Read the Weibo text completely and extract relevant evidence from six dimensions. En- sure each dimension ’s note is consistent with the depressive state given. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If relevant evi- dence is detected, provide the text index (e.g., text_xx, 文本_xx) and short description (e.g....

    work page

  31. [31]

    Ensure all notes are con- sistent with the depressive state

    Please read through the following full Weibo text and extract essential evidence across six preset dimensions. Ensure all notes are con- sistent with the depressive state. [Input]User depressive state: {label}, Full Weibo text: {text}. [Output requirements]- If evidence is found, provide the text index (e.g., text_xx,文本_xx) and a short explanation (e.g., ...

    work page

  32. [32]

    This text does not contain any expressions of negative emotions, nor does it involve any dimensions of depression

    "This text does not contain any expressions of negative emotions, nor does it involve any dimensions of depression."

    work page

  33. [33]

    There are no descriptions related to negative emotions in the text, so it does not fit any depression-related dimensions

    "There are no descriptions related to negative emotions in the text, so it does not fit any depression-related dimensions."

    work page

  34. [34]

    This text does not include expressions of neg- ative emotions, and therefore it is not classi- fied under any depression dimensions

    "This text does not include expressions of neg- ative emotions, and therefore it is not classi- fied under any depression dimensions."

    work page

  35. [35]

    The content does not reflect negative emo- tions and does not belong to any depression indicators

    "The content does not reflect negative emo- tions and does not belong to any depression indicators."

    work page

  36. [36]

    This passage contains no expressions of neg- ative emotions and is not applicable to any depression dimensions

    "This passage contains no expressions of neg- ative emotions and is not applicable to any depression dimensions."

    work page

  37. [37]

    There are no descriptions of negative emo- tions in the text, so it is not classified under any depression-related dimensions

    "There are no descriptions of negative emo- tions in the text, so it is not classified under any depression-related dimensions."

    work page

  38. [38]

    This content does not exhibit negative emo- tions and therefore does not involve any de- pression dimensions

    "This content does not exhibit negative emo- tions and therefore does not involve any de- pression dimensions."

    work page

  39. [39]

    The text does not contain any expressions related to negative emotions, nor does it meet the criteria for depression dimensions

    "The text does not contain any expressions related to negative emotions, nor does it meet the criteria for depression dimensions."

    work page

  40. [40]

    There are no signs of negative emotions in this text, so it does not belong to any depres- sion dimension

    "There are no signs of negative emotions in this text, so it does not belong to any depres- sion dimension."

    work page

  41. [41]

    The text does not describe negative emotions and does not cover any depression-related di- mensions

    "The text does not describe negative emotions and does not cover any depression-related di- mensions."

    work page

  42. [42]

    This passage does not contain any expres- sions related to negative emotions and is not within the scope of depression dimensions

    "This passage does not contain any expres- sions related to negative emotions and is not within the scope of depression dimensions."

    work page

  43. [43]

    The content does not include expressions of negative emotions, so it does not fall into any depression dimensions

    "The content does not include expressions of negative emotions, so it does not fall into any depression dimensions."

    work page

  44. [44]

    There are no expressions of negative emo- tions in the text, nor does it meet the criteria for depression dimensions

    "There are no expressions of negative emo- tions in the text, nor does it meet the criteria for depression dimensions."

    work page

  45. [45]

    This text does not exhibit any negative emo- tions and does not involve any depression di- mensions

    "This text does not exhibit any negative emo- tions and does not involve any depression di- mensions."

    work page

  46. [46]

    This content does not include descriptions of negative emotions and therefore is not classi- fied under any dimensions of depression

    "This content does not include descriptions of negative emotions and therefore is not classi- fied under any dimensions of depression."

    work page

  47. [47]

    The text does not show any negative emo- tions and is not applicable to depression- related dimensions

    "The text does not show any negative emo- tions and is not applicable to depression- related dimensions."

    work page

  48. [48]

    No expressions of negative emotions are seen in this passage, so it does not meet any depres- sion dimension standards

    "No expressions of negative emotions are seen in this passage, so it does not meet any depres- sion dimension standards."

    work page

  49. [49]

    There are no expressions of negative emo- tions in the text, so it does not belong to any depression dimension

    "There are no expressions of negative emo- tions in the text, so it does not belong to any depression dimension."

    work page

  50. [50]

    This text does not display any expressions re- lated to negative emotions and does not cover the dimensions of depression

    "This text does not display any expressions re- lated to negative emotions and does not cover the dimensions of depression."

    work page

  51. [51]

    The content lacks descriptions of negative emotions and is therefore not classified under any depression dimensions

    "The content lacks descriptions of negative emotions and is therefore not classified under any depression dimensions." A.4 Instructions for the Fine-Tuning Process • Complete Instruction = Task Instruction + Supplementary Instruction Task Instruction: This task involves the following 6 dimensions of depression: • Potential External Causes of Depression (S...

    work page

  52. [52]

    "Please read the following text segment and determine whether it contains any of the above 6 expressions of depressive emotion, and pro- vide a brief explanation; if none is present, please indicate that the text does not belong to any depression dimension."

    work page

  53. [53]

    "Read the following text and analyze whether it demonstrates any one of the aforementioned 6 depressive emotion expression dimensions, and provide a brief explanation; if such ex- pression is absent, please indicate that the text does not correspond to any dimension."

    work page

  54. [54]

    "Please carefully read the following text seg- ment and confirm whether any one of the 6 de- pressive emotion expression dimensions men- tioned above exists, and include a brief expla- nation; if not, please state that the text does not belong to any dimension."

    work page

  55. [55]

    "Please read the following content and de- termine whether it embodies any one of the above 6 depressive emotion expression dimen- sions, and provide a brief explanation; if not, please indicate that the text does not cover any depression dimension."

    work page

  56. [56]

    "Read the following text segment and deter- mine whether it contains any one of the 6 depressive emotion expressions mentioned above, and provide a concise explanation; if not, please state that the text does not belong to any depression dimension."

    work page

  57. [57]

    "Please read the following text and determine whether any one of the above 6 depressive emotion expression dimensions appears, and include a brief explanation; if not, please in- dicate that the text does not meet any dimen- sion."

    work page

  58. [58]

    "Read the following text and check whether any one of the above 6 depressive emotion ex- pression dimensions is presented, and provide a brief explanation; if not, please state that the text does not belong to any dimension."

    work page

  59. [59]

    "Please review the following text segment and determine whether any one of the above 6 depressive emotion expression dimensions ex- ists, and provide a brief explanation; if not, then indicate that the text does not belong to any dimension."

    work page

  60. [60]

    "Read the following text and confirm whether any one of the above 6 depressive emotion expression dimensions is embodied, and pro- vide a brief explanation; if not, please state that the text does not involve any depression dimension."

    work page

  61. [61]

    "Please read the following content and deter- mine whether it contains any one of the 6 de- pressive emotion expression dimensions men- tioned earlier, and provide a brief explanation; if not, please indicate that the text does not belong to any dimension."

    work page

  62. [62]

    "Read the following text segment and verify whether it demonstrates any one of the above 6 depressive emotion expression dimensions, and include a brief explanation; if not, please state that the text does not meet any dimen- sion."

    work page

  63. [63]

    "Please read the following text and check whether it exhibits any one of the above 6 de- pressive emotion expression dimensions, and provide a brief explanation; if not, please in- dicate that the text does not cover any depres- sion dimension."

    work page

  64. [64]

    "Read the following text and confirm whether any one of the 6 depressive emotion expres- sions mentioned above exists, and include a brief explanation; if such a feature is absent, please state that the text does not belong to any dimension."

    work page

  65. [65]

    "Please carefully read the following text seg- ment and determine whether it embodies any one of the above 6 depressive emotion expres- sion dimensions, and provide a concise expla- nation; if not, please indicate that the text does not correspond to any dimension."

    work page

  66. [66]

    "Read the following text segment and confirm whether it contains any one of the above 6 de- pressive emotion expression dimensions, and provide a brief explanation; if not, please state that the text does not involve any depression dimension."

    work page

  67. [67]

    "Please read the following text segment and determine whether any one of the 6 depressive emotion expression dimensions mentioned above appears in the text, and include a brief explanation; if not, please indicate that the text does not belong to any dimension."

    work page

  68. [68]

    "Read the following content and check whether it contains any one of the above 6 de- pressive emotion expression dimensions, and provide a brief explanation; if not, please state that the text does not belong to any depression dimension."

    work page

  69. [69]

    "Please read the following content and deter- mine whether any one of the above 6 depres- sive emotion expression dimensions exists, and provide a brief explanation; if not, please indicate that the text does not involve any di- mension."

    work page

  70. [70]

    "Read the following text segment and confirm whether it demonstrates any one of the above 6 depressive emotion expression dimensions mentioned above, and provide a brief explana- tion; if not, please state that the text does not meet any dimension."

    work page

  71. [71]

    Yes” or “No

    "Please read the following text and determine whether it contains any one of the above 6 de- pressive emotion expression dimensions, and provide a concise explanation; if not, please indicate that the text does not belong to any depression dimension." A.5 Module II Prompt Figure 3: Module II Prompt. A.6 Prompt for Table 3 On Section 5.3.1 Prompt is as fol...

    work page 2019

  72. [72]

    Instead of masking individual characters, it masks entire words during MLM, leading to a 9https://huggingface.co/google-BERT/ BERT-base-chinese richer understanding of context

    leverages RoBERTa (Liu et al., 2019) with a whole word masking (WWM) approach. Instead of masking individual characters, it masks entire words during MLM, leading to a 9https://huggingface.co/google-BERT/ BERT-base-chinese richer understanding of context. The extended pre-training (ext) on large Chinese corpora further refines its effectiveness in various...

    work page 2019