Automating social science work around image tagging via various online services.
Project description
Cross-service Laber Agreement Score
Image recognition services provide different labels for the same images, with differences broadly either being different word choices for the same concept or not seeing same objects on both images. Cross-service Laber Agreement Score (COSLAB) is a method to evaluate the semantic similarity on image labels across several image recognition systems. It first produces each image a set of labels, one for each service. Then it uses word embedding models to find the closest match across label sets to see how much cross-service agreement there are for each label.
Conceptual how to use
Our work further highlights that image recognition services differ on what they see on images. Due to this, we suggest approaching them as interprentations instead of objective truths. In response, we propose two different strategies which you can employ to use them in your research:
- Inclusive strategy: to ensure that your work takes different interprentations into account, use more than one image recognition service to label the images. Then merge all labels together and acknowldge that results have inheritent differences emerging from different services.
- Rigorous strategy: to ensure that your work takes different interprentations into account, use more than one image recognition service to label the images. Then use COSLAB to choose only labels above a pre-defined threshold level for further analysis.
Technical how to use
To be added.
References
- Berg, A., & Nelimarkka, M. (2023). Do you see what I see? Measuring the semantic differences in image‐recognition services' outputs. Journal of the Association for Information Science and Technology.
- Nelimarkka, M., & Berg, A. (2023). Is the World Different Depending on Whose AI Is Looking at It? Comparing Image Recognition Services for Social Science Research. Information Matters, 3(8).
Applications
- None yet. Please let us know if you use it.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distribution
Hashes for coslab_core-0.9.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 8fdfb7f9351e199bd50c2bf119178e3810e7754738ab21a195c400552425c15b |
|
MD5 | 73a394d78907f103dc345e68ea099b8b |
|
BLAKE2b-256 | 583c01ce05bc3f5da2b649015cce6282336872a01f6de47d1e0aaf12d4180203 |