Source-linked AI summary
The Age of Social Sensing
Dong Wang, Boleslaw K. Szymanski, Tarek Abdelzaher, Heng Ji, Lance Kaplan
TL;DR
Social media offers a large-scale lens on physical reality, but its reports can be distorted by human interpretation, bias, and intentional manipulation. This survey organizes social-sensing challenges and research directions, including methods for modeling social channels and assessing reliability. It concludes that reconstructing physical reality requires attention to these distortions, while recognizing that they can also carry information about human preferences and beliefs.
Problem
Method
The paper surveys social-sensing challenges and research directions, including social-channel models, reliability analysis, and linguistic processing.
Source: §2 · pp. 3–4, §3 · p. 6Results
Accounting for bias in polarized communities produced state reconstructions that aligned more closely with physical ground truth than reconstructions that ignored polarization.
Source: §2 · p. 5Takeaways & Limitations
Social-media distortions can indicate human preferences and beliefs, although the article primarily focuses on removing bias to represent physical reality more objectively.
Source: §2 · p. 3The article primarily focuses on removing bias for a more objective representation of physical reality, rather than using distortions as signals of preferences and beliefs.
Source: §2 · p. 3
Abstract
from arXiv · showhide
Online social media, such as Twitter and Instagram, democratized information broadcast, allowing anyone to share information about themselves and their surroundings at an unprecedented scale. The large volume of information thus posted on these media offer a new lens into the physical world through the eyes of the social network. The exploitation of this lens to inspect aspects of world state has recently been termed social sensing. The power of manipulating reality via the use (or intentional misuse) of social media opened concerns with issues ranging from radicalization by terror propaganda to potential manipulation of elections in mature democracies. Many important challenges and open research questions arise in this emerging field that aims to better understand how information can be extracted from the medium and what properties characterize the extracted information and the world it represents. Addressing the above challenges requires multi-disciplinary research at the intersection of computer science and social sciences that combines cyber-physical computing, sociology, sensor networks, social networks, cognition, data mining, estimation theory, data fusion, information theory, linguistics, machine learning, behavioral economics, and possibly others. This paper surveys important directions in social sensing, identifies current research challenges, and outlines avenues for future research.
1 INTRODUCTION
Social media makes observations of the world instantly shareable at unprecedented scale, but also increases the danger of distorted or manipulated perceptions. This section frames the challenge as building a macroscope that can reconstruct physical reality from human interpretations while accounting for social, linguistic, and cyber-physical factors.
1 INTRODUCTION
- Social media’s growing reach makes it harder to find valuable and truthful messages amid a much larger volume of content.The same dissemination power also makes it easier to manipulate perceptions of reality.Source: §1 · p. 1
- A social-media macroscope would use collective posts to observe physical and social phenomena at scale.The paper narrows its focus to reconstructing physical rather than social reality from human observations and interpretations.Source: §1 · pp. 1–2
- Key open questions concern how the macroscope’s properties and distortions can be characterized, and whether observations affect the observed system or later information propagation.The section also raises the possibility of deriving accuracy bounds and inferring human biases and trust relations from reported observations.Source: §1 · p. 1
- Human observers can recognize a broader range of events than physical sensors, but their reports reflect opinions, beliefs, and biases.The challenge is to map differing interpretations of the same event back to a neutral reconstruction of what may have occurred.Source: §1 · p. 2
2 CYBER-PHYSICAL CHALLENGES IN SOCIAL SPACES
Social sensing poses cyber-physical challenges because human reports are shaped by cognition, distortion, social ties, and propagation, unlike calibrated physical measurements. The section surveys models for interpreting signals and estimating reliability, and identifies unresolved challenges in source dependence, polarization, sensor fusion, and control loops.
2 CYBER-PHYSICAL CHALLENGES IN SOCIAL SPACES
- Social-sensing systems must model people, algorithms, and physical constraints, while accounting for cognition and human-generated distortion.Human reports reflect perceptions shaped by cognition; distortions can also indicate preferences and beliefs, although this article focuses primarily on removing bias for a more objective view of physical reality.Source: §2 · pp. 2–3
- Social-media reliability is difficult to assess because rumors, misinformation, and bias create correlated reports and errors across connected sources.Social ties, trust, and homophily can reinforce correlated beliefs and misrepresentations.Source: §2 · pp. 3–4
- Social sensing treats social media as a possible instrument for detecting, localizing, and tracking physical events, but its sensing modality remains an open question.Human-generated text, sound, images, and video raise questions about modeling and combining diverse data modalities.Source: §2 · p. 3
- A binary noisy-channel model represents the physical world as the source, human observations as output, and rumors or omissions as distortion.The model supports estimation for objectively observable facts, but needs extension when ground truth is subjective or undefined.Source: §2 · pp. 3–4
- Maximum-likelihood and estimation-theoretic approaches can jointly estimate world state and channel distortion, while error bounds assess reconstruction quality.Reliability analysis must account for correlated rumor errors and vague reports; translating natural-language vagueness into probabilities remains difficult.Source: §2 · p. 4
- Trust and shared views shape propagation, so repeated reports cannot establish reliability without understanding the underlying social network.Propagation patterns can also reveal trust relations and source biases, while complex, changing dependency graphs need further investigation.Source: §2 · pp. 4–5
- Accounting for source bias in polarized communities made reconstructed states align more closely with physical ground truth than when polarization was ignored.Estimated polarity-dependent propagation networks can expose source error dependencies and inform fact-finder design.Source: §2 · p. 5
- Combining social and physical sensing may explain sensor anomalies, but accurate fusion across sources and modalities remains an open challenge.Open questions also include inferring causal relationships between events and assessing the accuracy of those inferences.Source: §2 · p. 5
3 CHALLENGES IN THE LINGUISTIC SPACE
Social-media content challenges information extraction because it is informal, ambiguous, vague, and often depends on context beyond a single sentence. This section calls for adaptable semantic and language-processing methods that can structure information while addressing sentence-level, contextual, and discourse-level ambiguity.
3 CHALLENGES IN THE LINGUISTIC SPACE
- Information extraction is needed to turn social-media content into usable information about the physical world.The section calls for advances in rapid, low-cost information extraction and text mining.Source: §3 · p. 5
- Current language-processing approaches trade costly manual annotation and limited fact types against unsatisfactory performance when language-specific knowledge is excluded.The proposed direction is to unify symbolic and distributional semantics to discover and structure knowledge.Source: §3 · p. 5
- An information-extraction system should discover a domain-rich schema and typed facts without annotated data, then adapt quickly to new scenarios.Structured outputs would be more amenable to integration with sensor-data fusion systems than unstructured human-generated text.Source: §3 · pp. 5–6
- Informal posts often omit background and use short mentions, leaving local context insufficient to disambiguate entities and events.Censorship, sarcasm, and other motivations can also prompt information morphing, complicating co-reference resolution and the discovery of concealed facts.Source: §3 · p. 6
- Discourse-level ambiguity extends beyond sentence boundaries to unresolved pronouns and unclear relations among sentences.Understanding discourse also requires identifying topic, elaboration or contrast, and temporal structure.Source: §3 · p. 6
- Fuzzy and vague human descriptions are difficult to convert into numerical distributions because their use varies across people and societies.The passage describes this conversion as an open and difficult problem.Source: §3 · p. 6
4 FUTURE DIRECTIONS
The section identifies unresolved questions in modeling social media as an imperfect measurement channel and calls for interdisciplinary research. It proposes a field that puts social media analysis on analytic foundations comparable to physical-sensor fusion.
4 FUTURE DIRECTIONS
- The field lacks a unified interdisciplinary formulation of humans as sensors and social media as noisy measuring channels, leaving semantic errors and information-loss effects open.Questions include modeling language ambiguity, semantic errors, lossy summaries, and interpretation errors from mismatched source and recipient backgrounds.Source: §4 · p. 6
- Research must also address how to recognize corroborating evidence, handle irreconcilable observations, model information relaying, and account for social ties and group influence.The passage connects interpretation and processing to connectivity, familiarity, and trust between individuals.Source: §4 · p. 7
- The proposed work spans five disciplines: computational social science, linguistics, information theory, data mining, and cyber-physical systems.Their roles include modeling behavior and communication errors, deriving channel limits, assessing knowledge extraction, and developing robust estimation and control tools.Source: §4 · p. 7
- The authors call for a field combining social and cognitive models, linguistics, estimation and information theory, and reliability analysis to establish analytic foundations for social-media exploitation.They compare this goal to fusing hard data from physical sensors and signals.Source: §4 · p. 7
AUTHOR BIOS
The biographies identify Dong Wang as an assistant professor in computer science and engineering and describe Lance Kaplan's research, education, and professional service.
AUTHOR BIOS
- Dong Wang is an assistant professor in a Department of Computer Science and Engineering.Source: AUTHOR BIOS · p. 7
- Lance Kaplan is a researcher in the Networked Sensing and Fusion branch of the U.S. Army Research Laboratory.Source: AUTHOR BIOS · p. 8
- Kaplan earned electrical-engineering degrees from Duke University and the University of Southern California and has served in IEEE and ISIF leadership roles.The biography lists B.S., M.S., and Ph.D. degrees and service as IEEE Transactions on AES editor-in-chief and ISIF VP of Conferences.Source: AUTHOR BIOS · p. 8
- Kaplan's research interests include information and data fusion, reasoning under uncertainty, network science, resource management, and signal and image processing.Source: AUTHOR BIOS · p. 8