Method for identifying verifiable statements in text
A method, system and computer-usable medium are disclosed for identifying verifiable statements in a corpus of text. A training corpus of text containing manually annotated instances of verifiable and non-verifiable statements is processed to parse the text into segmented statements, which are in turn processed to extract features. The extracted features and the annotated statements are then processed with a machine learning algorithm to generate a verifiable statement classification model. In turn, the verifiable statement classification model is referenced by a verifiable statement classification system to distinguish verifiable and non-verifiable statements contained within an input corpus of text.
1. A computer-implemented method for identifying verifiable statements, comprising:
receiving a text containing a plurality of statements, the text received by a system configured to parse text input;
processing the text to parse the plurality of statements into segmented statements; and
processing the segmented statements to identify individual segmented statements that are verifiable, the verifiable statement being a subset of factual utterances that can be checked and verified or falsified via an authoritative source, each individual segmented statement that is verifiable being identified as a verifiable statement, the verifiable statement comprising an objective claim that is amenable to external verification, the identification performed by the system;
processing the segmented statements to extract features associated with the statements via a feature extraction, the feature extraction simplifying an amount of resources required to accurately describe a large set of data, the feature extraction performed by the system, the features being extracted via the feature extraction comprising semantic features, the semantic features being represented in a notation form to express an existence or non-existence of pre-established semantic properties, the feature extraction being performed using a plurality of named entities, each of the plurality of named entities referring to a place holder for a particular feature or piece of information, the feature extraction receiving information regarding language knowledge, the information regarding language knowledge including information regarding magnitude, information regarding velocity and information regarding quantified items, the information regarding language knowledge being used when performing the feature extraction to classify the segmented statements as verifiable or not verifiable;
receiving a training text comprising annotated verifiable statements, the annotated verifiable statements being verified by a statement verification service; and
processing the extracted features and the annotated verifiable statements to generate a verifiable statement classification model, the generation performed by the system, the verifiable statement classification model enabling automation of identification and classification of statements that contain objective claims that are not amenable to external verification; and wherein
named entity phrases are used to construct grammars, regular expressions and the verifiable statement classification model, each named entity phrase being associated with a list of possible phrases.
2. The method of claim 1 , wherein individual extracted features correspond to at least one member of the set of:
sentiment;
verbs;
verb tense;
nouns;
proper nouns;
magnitude;
velocity;
importance;
quantified items;
quantitative comparison operators; and
reference.
3. The method of claim 1 , wherein:
a machine learning algorithm is used by the system to perform the generation of the verifiable statement classification model.
4. The method of claim 3 , wherein:
the system performs the identification by referencing the verifiable statement classification model.