专利快速检索-快速检索全球专利，免费商用专利数据库-IPRDB

1. 发明专利

GB2569858A Constructing content based on multi-sentence compression of source content 未知
公开(公告)号：GB2569858A
公开(公告)日：2019-07-03
申请号：GB201817487
申请日：2018-10-26
申请人： ADOBE INC
发明人： NATWAR MODANI , KUNDAN KRISHNA , PRANAV RAVINDRA MANERIKER , BALAJI VASAN SRINIVASAN
IPC分类号： G06F16/31
摘要： Embodiments of the present disclosure provide systems, methods, and computer storage media directed to facilitating corpus-based content generation, in particular, using graph-based multi-sentence compression to generate a final content output. The method involves obtaining source content relevant to an input text snippet (e.g. keywords, phrases or sentences), utilizing a graphical representation of a plurality of sentences form the source content to identify a set of sentences in the snippet having overlapping information, generating a candidate sentence by compressing content in the set of sentences having overlapping information and then generating and providing final content comprising a set of candidate sentences including the candidate sentence. The method may also involve parsing the source content relevant to the input snippet into a plurality of sentences. Each of the plurality of sentences may be mapped to a first graph and assigned a weight, wherein the weight relates to the relevance between the corresponding sentence and the input snippet. At least a portion of the weighted sentences may be parsed into a plurality of word tokens.

2. 发明专利

GB2568571A Generating a targeted summary of textual content tuned to a target audience vocabulary 未知
公开(公告)号：GB2568571A
公开(公告)日：2019-05-22
申请号：GB201814861
申请日：2018-09-13
申请人： ADOBE INC
发明人： SAUMITRA SHARMA , KUNDAN KRISHNA , BALAJI VASAN SRINIVASAN , ANIKET MURHEKAR
IPC分类号： G06F16/34 , G06F40/00
摘要： A method for generating a targeted summary 116 of textual content 112 tuned to a target audience vocabulary 114 wherein a request to summarise textual content tuned to a target audience is received, an attention distribution 304 comprising words 306 from the textual content and audience vocabulary, and selection probability values 308 for each word is generated by a word generation model 120. The next word 312 in the targeted summary is selected based on the attention distribution and a linguistic preference model 122 indicating word preference probabilities 202 for the audience vocabulary and feedback 320 of the selected next word is provided to the word generation model causing it to modify the attention distribution for selection of subsequent words of the summary based on the feedback of the next generated word. Also disclosed is a method of training a linguistic preference model using target audience training data wherein the training data comprises a corpus of textual content and the model is trained using one or more machine learning techniques to learn word preference probabilities of the target audience based on word usage frequencies. Also disclosed is a system for the execution of both methods.

3. 发明专利

GB2573189A Generating a topic-based summary of textual content 未知
公开(公告)号：GB2573189A
公开(公告)日：2019-10-30
申请号：GB201901522
申请日：2019-02-04
申请人： ADOBE INC
发明人： KUNDAN KRISHNA , BALAJI VASAN SRINIVASAN
IPC分类号： G06F16/34 , G06F40/00
摘要： A method for generating a summary of textual content tuned to a specific topic involves using a topic-aware encoding model to encode 804 the text using a topic label (e.g. a one-hot vector) to generate topic-aware encoded text. A word generation model selects a next word 808 for the summary from the encoded text. The word generation model is trained using machine learning and training data including documents with corresponding summaries, each having an associated topic. The selected next word is provided as feedback 810 to the word generation model. Also disclosed is a method for training the encoding word generation models by obtaining an intermediate dataset comprising documents, and a summary of each document with an associated topic. Training data is generated by merging the text of first and second documents to make a new document which is associated with the summary and topic of the first document, merging their text again and associating the resulting new document with the summary and topic of the second document. The original documents are discarded. This process is repeated until the intermediate training set is exhausted. The training data is then used to train the encoding and word generation models.

你已经成功收藏专利！

检索式保存成功!

IPRDB

热门服务

关于我们

友情链接

联系方式