This study develops an entropy-based evaluation model to quantify meaning loss in knowledge reproduction through social media paper-reading activities. Using a mixed-methods approach, the research analyzed 420 paper-reading posts across WeChat, Zhihu, Bilibili, and Twitter, examining how academic knowledge transforms during digital dissemination. The entropy model conceptualizes knowledge reproduction as an information transmission process where entropy increases along dissemination chains, with preservation rates varying significantly across platforms. Validation testing confirmed strong correlation between algorithmic and human assessments (r = 0.84, p < 0.001), with the model successfully identifying 92.3% of expert-classified highdistortion cases. Results revealed distinct platform-specific entropy distributions, with structured knowledge environments demonstrating lower entropy values (Zhihu: M = 0.38, SD = 0.13) compared to microblogging formats (Twitter: M = 0.59, SD = 0.16). Cross-disciplinary analysis showed significant entropy variations across knowledge domains (F(3, 416) = 24.38, p < 0.001), with physics/mathematics content exhibiting the lowest entropy (M = 0.32, SD = 0.11) and social sciences the highest (M = 0.53, SD = 0.15). Multiple regression analysis identified content creator expertise as the strongest predictor of low entropy (β = -0.43, p < 0.001), followed by audience domain knowledge (β = -0.37, p < 0.001). Multimodal integration demonstrated a buffering effect against complexity-driven entropy increases, reducing entropy values by 0.14 on average. The study provides both theoretical advancements in understanding digital knowledge transformation mechanisms and practical applications for improving scientific communication in social media environments. The findings suggest that effective knowledge dissemination requires strategic adaptation to platform-specific constraints, with integrated multimodal presentation strategies showing particular promise for preserving meaning across all disciplines and platforms examined.
