Automatic misinformation identification algorithms

Misinformation

Unintentional; inadvertently misleads

Disinformation

Intentionally misleads

Key methods of misinformation identification

Knowledge based methods - is the information provided consistent with the truth

Style based methods - what linguistic strategies are used in the content? Are they usually used when deceiving?

Source based - credibility of the source

Propagation based - how the content has spread online

Steps of misinformation identification

Check worthiness - find what might need verification

Verified claim retrieval - there will be a dataset of labelled verified claims that will be used to help verify the new claims. (style based methods)

Supporting evidence retrieval - We need to see what the source of the information is

Factuality verification - Using all information collected from previous steps, predict the veracity of a claim.

Automatic Misinfo Identification:
Sources of Evidence

• No evidence

• Claims with metadata (media source/speaker profile)

• Information from multiple trusted documents and sentences

• Repositories of previously fact checked claims

Challenges with automatic identification misinformation

• There might not be a labelled or reliable dataset available to use

• It can be hard to trace what source has been used and how it has been spread

• It might not work for complex claims

• It faces the same standard limitations of other AI models