Scribd
Upload
2 views1 page

Risk Token Identification Techniques

The document describes steps to identify risk tokens and groups of risk tokens from text, including converting text to bag-of-words vectors and using WordNet to identify risk keywords. It also discusses extracting paragraph-level features through windowing and applying convolution to the outputs of the windowing approach.

Uploaded by

Bamlaku Amare
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
2 views1 page

Risk Token Identification Techniques

The document describes steps to identify risk tokens and groups of risk tokens from text, including converting text to bag-of-words vectors and using WordNet to identify risk keywords. It also discusses extracting paragraph-level features through windowing and applying convolution to the outputs of the windowing approach.

Uploaded by

Bamlaku Amare
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

1.

Identifying Risk Token (RT)


The first step is to convert the uni-gram, bi-gram, and tri-gram into a Bag-Of-Words vector. The
convolutional technique is used to learn the features at the sentence level. Moreover, WordNet is
used to populate a database with keywords that have unique characteristics that define a risk. The
special words that can describe a variety of risk types are mapped to other words.

2. Identifying Group of Risk Token (GRT)


Noun sequences are among the traditional lexical features. Such pairs have been taken into
account for both bi-grams and tri-grams in risk descriptions. The list of those pairs is regarded as
being a component of the training dictionary.

A. Paragraph Level Features


The representation of the semantic composition of a risk description is limited by the single-word
token and the pair of tokens. The reason for this is that putting together long-distance tokens
within a paragraph may not always suffice to represent the discriminating feature of a specific
risk type. A risk description paragraph contains sentences. Windowing is applied to the entire
paragraph, with no regard for stop words that separate two sentences. This results in the window
feature (WF) and position feature (PF) (PF). Word features (WF) yield a vector of concatenated
token partitions in a sentence/ (s).

B. Feature concatenation by Convolution


A collection of local features computed around words that contribute to the Bag-Of-Words
histogram. Each local feature is the result of the windowing method. The next step is to apply a
convolution function to the window approach's outputs. The dot product of the weight vector m
and each of the windowing approach outputs yields one-dimensional convolution. The type of
convolution used is narrow, which results in a vector that is smaller in size than the original ones.

You might also like