Browse Prior Art Database

Method and System for Detecting Hate Speech by Utilizing Low-Dimensional, Distributed Representations of User Comments

IP.com Disclosure Number: IPCOM000241363D
Publication Date: 2015-Apr-21
Document File: 2 page(s) / 24K

Publishing Venue

The IP.com Prior Art Database

Related People

Nemanja Djuric: INVENTOR [+6]

Abstract

A method and system is disclosed for detecting hate speech by utilizing low-dimensional, distributed representations of user comments. A two-step method for hate speech detection is utilized wherein, joint modeling of and words comments is performed and corresponding distributed representations are learned in a joint space that result in a low-dimensional text embedding. The low-dimensional text embedding is used to train a binary classifier to distinguish between hateful and clean comments.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 53% of the total text.

Method and System for Detecting Hate Speech by Utilizing Low-Dimensional, Distributed Representations of User Comments

Abstract

A method and system is disclosed for detecting hate speech by utilizing low-dimensional, distributed representations of user comments.  A two-step method for hate speech detection is utilized wherein, joint modeling of and words comments is performed and corresponding distributed representations are learned in a joint space that result in a low-dimensional text embedding.  The low-dimensional text embedding is used to train a binary classifier to distinguish between hateful and clean comments.

Description

In the age of ever-increasing volume and complexity of internet, millions of users are provided unrestricted access to vast amounts of content that allows for unimaginable privileges.  However, due to internet's non-restrictive nature and legal protection of free speech in certain countries, users misuse the internet to promote offensive and hateful language, which mars experience of regular users, affects business of online companies, and cause severe real-life consequences.  To mitigate these detrimental effects, many companies strictly prohibit hate speech on websites by implementing algorithmic solutions to discern hateful content.  However, due to scale and nature of task, hate speech still remains a problem in online user comments.

Disclosed is a method and system for detecting hate speech by utilizing low-dimensional, distributed representations of user comments.

In accordance with the method and system, a two-step method for hate speech detection is utilized.  First, paragraph2vec is used for joint modeling of comments and words.  Thereafter, corresponding distributed representations are learned in a joint space using a Continuous Bag of Words (CBOW) neural language model.

The neural language model takes advantage of word order and states same assumption of n-gram language models i.e. words that are close in a sentence are a...