You are on page 1of 3

https://asistdl.onlinelibrary.wiley.com/doi/full/10.1002/pra2.2015.

14
5052010082

LINGUISTIC APPROACHES

 Bag of words

Words extracted from article

Individual word analyzed

Words are tagged into lexical cues (parts of speech, shallow syntax, location based words)

Disadvantages:

1. Might not include useful contextual information because of n-gram words.


2. ambiguous word sense is absent.
 Deep Syntax Analysis

- implemented through Probability Context Free Grammars (PCFG)

Sentences used in the article

Set of rewrite rules (parse tree)

Describe syntax structure (noun, verbs)

Final rewrite gives a parse tree with a probability of fake news

Disadvantages:

1. Not sufficiently capable of predicting fake news alone. But can be combined with other
techniques.
 Semantic Analysis

Based on
- profile of the author of the news article
- Analysis of the authors description

For example- a writer with no experience is more likely to publish deceptive articles and
omission of facts.

Disadvantages:

1. The ability to determine alignment between attributes and descriptors depends on a


sufficient amount of content for profiles.
2. Challenge of correctly associating authors description with extracted attributes.

 Classfiers:

Support vector machine (SVM)-

Training dataset SVM data format Define SVM parameters for clustering of data

Trained model V- cross validation

Testing dataset SVM prediction

You might also like