This study aims to construct a comprehensive feature system for identifying artificial intelligence–generated content (AIGC) in online Q&A communities, thus uncovering the key factors and mechanisms influencing the identification of AIGC. First, based on the theory of systemic functional linguistics (SFL) and information quality (IQ), this article extracts vocabulary, content, structure, and emotional features from the text, and identifies the AIGC through nine mainstream machine learning algorithms. Subsequently, three widely used resampling strategies are exploited to address the category imbalance problem. The grid search optimisation algorithm fine-tunes different combinations of parameters to improve the performance of the identification classifier. Finally, SHAP values are introduced to evaluate and elucidate the global feature importance and feature influence mechanism. A Chinese corpus from the Zhihu Q&A community is constructed to verify the validity of these methods. The experimental results show that the eXtreme Gradient Boosting (XGBoost) model optimised with hybrid sampling and grid search parameters exhibits excellent performance in identifying AI-generated text, which achieves an F1-score of 0.9935, an improvement of 0.11 percentage points over the original model. In addition, all four dimensions of features constructed in this article contribute to AI-generated text identification, and the results of feature interpretability analysis show the greatest impact of features that focus on content readability. The study facilitates the identification and labelling of AIGC in online Q&A communities, thereby enhancing transparency and accountability of information shared online.
LiuZWuYWangJ.The impact of artificial intelligence generated content technologies on knowledge production and dissemination. J Intell2023; 42(7): 123–130.
2.
ZhangHLiLLiC.ChatGPT performance evaluation on Chinese language and risk measures. Data Anal Knowl Discover2023; 7(3): 16–25.
3.
LuWLiuJMaY.The influence of large language models represented by ChatGPT on information resources management. Doc Inform Knowl2023; 40(2): 6–9.
4.
MoJKangXHuZ, et al. Towards trustworthy digital media in the AIGC era: an introduction to the upcoming IsoJpegTrust standard. IEEE Commun Stand Magazine2023; 7(4): 2–5.
5.
GuoBZhangXWangZ, et al. How close is ChatGPT to human experts? Comparison corpus, evaluation, and detection, 2023, ArXiv preprint, https://arxiv.org/abs/2301.07597
6.
ZhuYChenGLuY, et al. Generative artificial intelligence governance action framework: content analysis based on AIGC incident report texts. Document Inform Knowl2023; 40(4): 41–51.
7.
SunL.Research on artificial intelligence data falsification risk based on GPT model. J Inform Secur Res2023; 9(6): 518–523.
8.
JavieraCáceresBMuñozF.Artificial Intelligence, A new frontier for intellectual property policymaking. NTUT J Intellect Propert Law Manag2020; 9: 108–131.
9.
ZhuJ.To regulate the disinformation brought by generative artificial intelligence as exemplified by the application of ChatGPT: challenges and possible solutions. J Compar Law2023; 5: 34–54.
10.
MarkowitzDMHancockJBailensonJ.Linguistic markers of Inherent AI deception and intentional human deception: evidence from hotel reviews, 2023, https://doi.org/10.31234/osf.io/mnyz8 (accessed 21 September 2023)
11.
SolaimanIBrundageMClarkJ, et al. Release strategies and the social impacts of language models, 2019, arXiv preprint, https://arxiv.org/abs/1908.09203
12.
MitchellELeeYKhazatskyA, et al. DetectGPT: zero-shot machine-generated text detection using probability curvature, 2023, arXiv preprint, https://arxiv.org/abs/2301.11305
13.
JawaharGAbdul-MageedMLakshmananLV. Automatic detection of machine generated text: a critical survey, 2020, arXiv preprint, https://arxiv.org/abs/2011.01314
14.
DongTYangPXuY, et al. Generated text detection based on factual and semantic consistency. J Sichuan Univ2023; 60(4): 65–72.
15.
WangYGuoXLiuZ, et al. Detection and comparative study of differences between AI-generated and scholar-written Chinese abstracts. J Intell2023; 42(9): 127–134.
16.
HayawiKShahriarSMathewSS. The imitation game: detecting human and AI-generated texts in the era of ChatGPT and BARD. J Inform Sci. Epub ahead of print 14February2024. DOI: 10.1177/01655515241227531.
17.
KowalczykPRöderMDürrA, et al. Detecting and understanding textual deepfakes in online reviews. In: Proceedings of the 55th Hawaii international conference on system sciences, Honolulu, HI, 4–7 January 2022, pp. 1480–1489. USA: ScholarSpace.
18.
ZhouLBurgoonJKNunamakerJF, et al. Automating linguistics-based cues for detecting deception in text-based asynchronous computer-mediated communications. Group Decis Negotiat2004; 13: 81–106.
19.
FröhlingLZubiagaA.Feature-based detection of automated language models: tackling GPT-2, GPT-3 and Grover. PeerJ Comput Sci2021; 7: e443.
20.
MindnerLSchlippeTSchaaffK. Classification of human-and AI-generated texts: investigating features for ChatGPT. In: International Conference on Artificial Intelligence in Education Technology, Kobe, Japan, 20–22 March 2023, pp. 152–170. Singapore: Springer.
21.
LeeSCLeeDGSeoYS.Determining the best feature combination through text and probabilistic feature analysis for GPT-2-based mobile app review detection. Appl Intell2024; 54(2): 1219–1246.
22.
ZellersRHoltzmanARashkinH, et al. Defending against neural fake news. Adv Neural Inform Proces Syst2019; 32: 9054–9065.
23.
FagniTFalchiFGambiniM, et al. TweepFake: about detecting deepfake tweets. PLoS ONE2021; 16(5): e0251415.
24.
DengSWangFWangH.Identification methods of artificial intelligence generated content in online communities. Document Inform Knowl2024; 41(2): 28–38.
25.
MatthiessenCMWangBMaY, et al. Systemic functional insights on language and linguistics. Singapore: Springer, 2022.
26.
WangDYangYZhangR.Study on the classification of user standpoints for the rumors of public health emergencies from the perspective of SFL theory. J Modern Inform2021; 41(2): 19–29.
AvaisPRafiMSAnwarN.A meta-functional analysis of Baba Bulleh Shah’s Kafi: Ranja Ranjha Kardi Hun Mein Apey Ranjha Hoi. J Social Sci Humanit2018; 26(1): 105–140.
29.
ZhuJGanJ.Multi-modal discourse analysis of national image construction in the documentary human world. Acad J Manag Social Sci2023; 3(3): 206–212.
30.
ZhangXChengYChenA, et al. How rumors diffuse in the infodemic: evidence from the healthy online social change in China. Technol Forecast Social Change2022; 185: 122089.
31.
MengYYangNQianZ, et al. What makes an online review more helpful: an interpretation framework using XGBoost and SHAP values. J Theoretical Appl Electron Comm Res2020; 16(3): 466–490.
32.
SunXZhaoYZhuQ.The construction of answer quality evaluation index for social search: an empirical study on social Q&A system. J Libr Sci China2015; 41(4): 65–82.
33.
OhSWorrallA.Health answer quality evaluation by librarians, nurses, and users in social Q&A. Libr Inform Sci Res2013; 35(4): 288–298.
34.
GuoSZhangXTaoX, et al. Research on automated evaluation of user generated answer quality in social question and answer community—Taking ‘Zhihu’ as an example. Library and Information Service2019; 63(11): 118–130.
35.
JiaJSongEMSuH.Research on assessment of answer quality in social Q&A platform. J Inform Res Manag2013; 3(2): 19–28.
36.
LuoYCaoQ.Research on answers’ quality of social Q&A platforms based on RIPA. Libr Inform Serv2015; 59(3): 126–133.
37.
WangRYReddyMPKonHB.Toward quality data: an attribute-based approach. Decis Support Syst1995; 13(3–4): 349–372.
38.
WuYZhangXTaoX, et al. Construction of answer quality evaluation index based on user perception of academic question and answer community. Inform Sci2020; 38(10): 141–147.
39.
FichmanP.A comparative assessment of answer quality on four question answering sites. J Inform Sci2011; 37(5): 476–486.
40.
KumarageTGarlandJBhattacharjeeA, et al. Stylometric detection of AI-generated text in Twitter timelines, 2023, arXiv preprint, https://arxiv.org/abs/2303.03697
41.
CrothersEJapkowiczNViktorHL.Machine-generated text: a comprehensive survey of threat models and detection methods. IEEE Access2023; 11: 70977–71002.
42.
LampridisOKaranatsiouDVakaliA.MANIFESTO: a huMAN-centric explaInable approach for FakE news spreaders deTectiOn. Computing2022; 104: 717–739.
43.
YiMZhangTLiZ.Research on the ranking of social Q&A community answer based on multidimensional features. Libr Inform Serv2020; 64(17): 103–113.
44.
SeeAPappuASaxenaR, et al. Do massively pretrained language models make better storytellers?2019, arXiv preprint, https://arxiv.org/abs/1909.10705
45.
GehrmannSStrobeltHRushAM. GLTR: statistical detection and visualization of generated text, 2019, arXiv preprint, https://arxiv.org/abs/1906.04043
46.
ChengYXuDDongJ.Automatic grading of Chinese text reading difficulty based on multiple linguistic feature. J Chin Inform Proces2020; 34(4): 101–110.
47.
MoZPanDLiuH, et al. Analysis on AIGC false information problem and root cause from the perspective of information. Document Inform Knowl2023; 40(4): 32–40.
GuoSBuH.Research on the sorting method of answer usefulness in social Q&A community integrating Word2vec and WGRA – Taking ctrip Q&A as an example. Libr Inform Serv2021; 65(23): 126–135.
50.
FengWTangJ.A ranking model for answer selection with deep matching features. J Chin Inform Proces2019; 33(1): 118–124.
51.
ReimersNGurevychI.Sentence-bert: sentence embeddings using siamese bert-networks, 2019, arXiv preprint, https://arxiv.org/abs/1908.10084
52.
BaoMLiJZhangJ, et al. Learning semantic coherence for machine generated spam text detection. In: 2019 International joint conference on neural networks (IJCNN), Budapest, 14–19 July 2019, pp. 1–8. New York: IEEE.
53.
MitrovićSAndreolettiDAyoubO. ChatGPT or human? Detect and explain. Explaining decisions of machine learning model for detecting short ChatGPT-generated text, 2023, arXiv preprint, https://arxiv.org/abs/2301.13852
54.
MieczkowskiHHancockJTNaamanM, et al. AI-mediated communication: language use and interpersonal effects in a referential communication task. Proc ACM Human-Comput Interact2021; 5: 1–14.
55.
XuLLinHPanY, et al. The construction of Chinese emotion word ontology. J China Soc Sci Tech Inform2008; 27(2): 180–185.
56.
LiuTWangLZhuQ.Research on user churn prediction of smart senior care service platforms based on SHAP interpretation method. Data Anal Knowl Disc2024; 8(1): 40–54.
57.
ZaitsuWJinM.Distinguishing ChatGPT (-3.5,-4)-generated and human-written papers through Japanese stylometric analysis, 2023, arXiv preprint, https://arxiv.org/abs/2304.05534
ZhangZWuCQuS, et al. An explainable artificial intelligence approach for financial distress prediction. Informa Proces Manag2022; 59(4): 102988.
60.
ZengZLiTSunS, et al. A novel semi-supervised self-training method based on resampling for Twitter fake account identification. Data Technol Appl2022; 56(3): 409–428.
61.
HartmannJHuppertzJSchampC, et al. Comparing automated text classification methods. Int J Res Market2019; 36(1): 20–38.
62.
GuoHXLiYJShangJ, et al. Learning from class-imbalanced data: review of methods and applications. Expert Syst Appl2017; 73: 220–239.
63.
FonsecaABMartinsDCJrWicikZ, et al. Addressing classification on highly imbalanced clinical datasets. In: International conference on computational advances in bio and medical sciences, Virtual Event, 16 December 2021, pp. 103–114. Cham: Springer.
64.
JohnsonJMKhoshgoftaarTM.Survey on deep learning with class imbalance. Jo Big Data2019; 6(1): 1–54.
65.
FengHDanTWangW, et al. A combination of resampling method and machine learning for text classification on imbalanced data. In: International conference on AI and mobile services, Honolulu, HI, 10–14 December 2021, pp. 3–17. Cham: Springer.
66.
SunYDingSZhangZ, et al. An improved grid search algorithm to optimize SVR for prediction. Soft Comput2021; 25: 5633–5644.
67.
LundbergSMLeeSI.A unified approach to interpreting model predictions. Adv Neural Inform Proces Syst2017; 2: 4765–4774.
68.
TianYLiGMaoJ.Predicting the evolution of scientific communities by interpretable machine learning approaches. J Inform2023; 17(2): 101399.
