Prediction of social media postings as trusted news or as types of suspicious news
Disclosed are systems, techniques, and non-transitory storage media for predicting social media postings as being trusted news or a type of suspicious news. The systems, techniques, and non-transitory storage media are based on unique neural network architectures that learn from a combined representation including at least representations of social media posting content and a vector representation of communications among connected users.
1. A computer-implemented method of predicting social media posts as trusted news or as types of suspicious news, the method comprising:
providing records from a social media site, each record comprising a social-media posting;
for each record, calculating a text representation based on the record, linguistic markers based on the record, and a user representation representing communications among connected users regarding the record;
inputting the text representations, linguistic markers, and user representations into a neural network having a content sub-network receiving the text representations and a vector representation sub-network receiving the linguistic markers and the user representations;
merging output from the content sub-network and output from the vector representation network according to a fusion operator, wherein the output from each sub-network has a different modality; and
calculating according to the neural network a distribution over probability classes regarding the records.
2. The computer-implemented method of claim 1 , wherein content sub-network comprises an embedding layer and at least one of a convolutional layer or a recurrent layer.
3. The computer-implemented method of claim 2 , wherein the recurrent layer comprises a long-short term memory layer.
4. The computer-implemented method of claim 1 , wherein the linguistic markers are selected from the group consisting of: psycholinguistic cues, style, syntax, biased language markers, subjective language markers and connotations, hedges, implicative, factive, assertive and report verbs, moral foundation theory markers, and combinations thereof.
5. The computer-implemented method of claim 1 , wherein the records comprise one or more foreign languages.
6. The computer-implemented method of claim 1 , further comprising the step of: labeling the probability classes with labels correlated to a set of pre-selected labels based on trusted news or types of suspicious news.
7. The computer-implemented method of claim 1 , further comprising inputting an image, an image representation, or both into the neural network.
8. The computer-implemented method of claim 7 , the neural network having the vector representation sub-network receiving the image, the image representation, or both.