
NSF Org: |
IIS Division of Information & Intelligent Systems |
Recipient: |
|
Initial Amendment Date: | July 27, 2016 |
Latest Amendment Date: | April 25, 2017 |
Award Number: | 1618244 |
Award Instrument: | Standard Grant |
Program Manager: |
Sylvia Spengler
sspengle@nsf.gov (703)292-7347 IIS Division of Information & Intelligent Systems CSE Directorate for Computer and Information Science and Engineering |
Start Date: | September 1, 2016 |
End Date: | May 31, 2020 (Estimated) |
Total Intended Award Amount: | $471,992.00 |
Total Awarded Amount to Date: | $471,992.00 |
Funds Obligated to Date: |
|
History of Investigator: |
|
Recipient Sponsored Research Office: |
10 W 35TH ST CHICAGO IL US 60616-3717 (312)567-3035 |
Sponsor Congressional District: |
|
Primary Place of Performance: |
Chicago IL US 60616-3717 |
Primary Place of
Performance Congressional District: |
|
Unique Entity Identifier (UEI): |
|
Parent UEI: |
|
NSF Program(s): | Info Integration & Informatics |
Primary Program Source: |
|
Program Reference Code(s): |
|
Program Element Code(s): |
|
Award Agency Code: | 4900 |
Fund Agency Code: | 4900 |
Assistance Listing Number(s): | 47.070 |
ABSTRACT
Measuring public perceptions and how they change over time is a central problem in marketing, public health, and politics. Traditional measurement methods rely on surveys and focus groups, which can be costly and time-consuming. Online social networks offer an attractive alternative: real-time perceptions can be estimated from public, online activity and compared with an entity's communications to quantify how public messaging affects perception. While prior algorithmic approaches rely purely on text-based sentiment analysis, this project will develop novel methods based on the insight that an entity's online social connections are indicative of how they are perceived (e.g., "birds of a feather flock together"). Thus, rather than typical one-dimensional measures of sentiment, the project will instead investigate public perception with respect to multiple characteristics of an entity (e.g., is it seen as pro-environment, pro-health, etc.). A multi-faceted evaluation will be performed to study the phenomenon of "greenwashing," a deceptive marketing practice in which firms market their products or policies as more environmentally friendly than they truly are. This project has the potential to enhance consumer protection by exposing deceptive marketing practices.
The project will develop social network analysis algorithms to assess perception of an entity and also language processing algorithms to quantify the communications of an entity with respect to a perceptual attribute. The approaches to both problems rely on innovative algorithms to measure the strengths of the social and linguistic relations between public entities and exemplar accounts that typify the perceptual attribute of interest. A key advantage of the approach is its minimal requirement of human input, e.g., given only a single keyword like "environment," the approach identifies suitable exemplars and fits linguistic and perceptual models. The project will develop novel machine learning methods for domain adaptation, positive-unlabeled learning, and learning from label proportions in order to fit such models and ensure they are robust to omitted variable bias. The models will be evaluated using public Twitter and Facebook data to quantify the relationship between the perceptions and online communications of brands and other public entities, with a particular focus on identifying cases of greenwashing.
PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH
Note:
When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external
site maintained by the publisher. Some full text articles may not yet be available without a
charge during the embargo (administrative interval).
Some links on this page may take you to non-federal websites. Their policies may differ from
this site.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
Measuring public perceptions and how they change over time is a central problem in marketing, public health, and politics. Traditional measurement methods rely on surveys and focus groups, which can be costly and time-consuming. Online social networks offer an attractive alternative: real-time perceptions can be estimated from public, online activity and compared with an entity's communications to quantify how public messaging affects perception. While prior algorithmic approaches rely purely on text-based sentiment analysis, the project developed novel methods based on the insight that an entity's online social connections are indicative of how they are perceived (e.g., "birds of a feather flock together"). Thus, rather than typical one-dimensional measures of sentiment, the project investigated public perception with respect to multiple characteristics of an entity (e.g., is it seen as pro-environment, pro-health, etc.). As a use case, this project examined "greenwashing," a deceptive marketing practice in which firms market their products or policies as more environmentally friendly than they truly are.
The project developed social network analysis algorithms to assess perception of an entity and also language processing algorithms to quantify the communications of an entity with respect to a perceptual attribute. The approaches to both problems rely on innovative algorithms to measure the strengths of the social and linguistic relations between public entities and exemplar accounts that typify the perceptual attribute of interest. A key advantage of the approach is its minimal requirement of human input --- e.g., given only a single keyword like "environment," the approach identifies suitable exemplars and fits linguistic and perceptual models.
There are a number of broader, technical contributions of this project that have general applicability to other areas of machine learning and text classification:
1. Classifier Robustness: While machine learning has made great strides recently, automated classification algorithms still make seemingly simple mistakes. Drawing on a long line of research in causal inference, this project has developed a number of novel classification methods that have increased the robustness of such methods on instances that differ somewhat from the original training instances. For example, if a human makes a minor edit to a sentence, the classifier should still be able to classify it correctly. The resulting methods are less susceptible to spurious correlations than existing approaches, not only increasing robustness but also improving trust in autonomous systems.
2. Alternative training methods for machine learning: Traditional machine learning requires many thousands, if not millions, of training examples, which can be expensive to maintain and can quickly become outdated. A key part of this project has been the development of weakly supervised learning methods that operate using cheaper, more readily available types of supervision. For example, the fact that 80% of a set of instances are positive is often easier to collect than labeling every single instance individually. The methods developed have been shown to achieve comparable accuracy using this sort of supervision as methods that require more expensive supervision, which expands the types of problems that can be solved with machine learning.
3. Combining social networks and text classification: Understanding human language is difficult to do without understanding the social context in which the communications occur. With online social networks, we can observe not only the language used, but also how two users are connected socially. By combining these perspectives, we have found that we can construct more accurate models of the intent and semantics of online communications.
Last Modified: 09/29/2020
Modified by: Aron Culotta
Please report errors in award information by writing to: awardsearch@nsf.gov.