Browse Prior Art Database

A System and Method to Recognize Paraphrase using Deep Learning

IP.com Disclosure Number: IPCOM000242105D
Publication Date: 2015-Jun-18
Document File: 4 page(s) / 174K

Publishing Venue

The IP.com Prior Art Database

Abstract

A system and method to recognize paraphrasing using deep learning is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 79% of the total text.

Page 01 of 4

A System and Method to Recognize Paraphrase using Deep Learning

Disclosed is a system and method to recognize paraphrasing using deep learning. The disclosed system applies deep learning to the problem of paraphrase recognition. Paraphrase recognition is the task of producing a score to determine if a pair of input texts are paraphrases of each other. Input features are generated automatically. The system learns to recognize the patterns in text and classifies paraphrasing texts uses deep learning techniques. The paraphrase recognition task takes a pair of texts (words, phrases, sentences, documents) and produces a numeric score to indicate how similar they are.

The disclosed system and method applies deep learning to the problem of paraphrase recognition and may include the following capabilities:

Generate input feature representation for texts.


1.

Recognize patterns in input texts through convolutional neural network.

2.

Learn to classify paraphrasing texts using outputs from intermediate layers.


3.

Figure 1

Figure 1 depicts the system input and output.

1


Page 02 of 4

Figure 2

Figure 2 depicts an overall structure of the deep learning based system.

2


Page 03 of 4

Figure 3

Figure 3 depicts example embodiments for feature generation which may comprise a bag-of-words, a sequence of words (w0,w1,…,wn) with embedding, and parse nodes embedding.

3


Page 04 of 4

Figure 4

Figure 4 depicts a deep neural network infrastructure and flow. The system may include support f...