http://rdf.ncbi.nlm.nih.gov/pubchem/patent/CN-114611510-A
Outgoing Links
Predicate | Object |
---|---|
assignee | http://rdf.ncbi.nlm.nih.gov/pubchem/patentassignee/MD5_ac31afbea1cbbb03498644721ffb4a62 |
classificationCPCInventive | http://rdf.ncbi.nlm.nih.gov/pubchem/patentcpc/G06F40-216 http://rdf.ncbi.nlm.nih.gov/pubchem/patentcpc/G06F40-289 |
classificationIPCInventive | http://rdf.ncbi.nlm.nih.gov/pubchem/patentipc/G06F40-289 http://rdf.ncbi.nlm.nih.gov/pubchem/patentipc/G06F40-216 |
filingDate | 2022-03-23-04:00^^<http://www.w3.org/2001/XMLSchema#date> |
inventor | http://rdf.ncbi.nlm.nih.gov/pubchem/patentinventor/MD5_fb49077858ddd2def2851f1e7fa76614 http://rdf.ncbi.nlm.nih.gov/pubchem/patentinventor/MD5_6d34a6fea80a135c22020cccb97bd59d http://rdf.ncbi.nlm.nih.gov/pubchem/patentinventor/MD5_9414c5b377326bdbf4382027468438b5 |
publicationDate | 2022-06-10-04:00^^<http://www.w3.org/2001/XMLSchema#date> |
publicationNumber | CN-114611510-A |
titleOfInvention | Implementation method and device for assisting machine reading comprehension based on generative model |
abstract | An implementation method and device for assisting machine reading comprehension based on a generative model, constructing a reading comprehension model for multiple-choice questions, including two workflows, 1) generating a flow, inputting the question into the encoder to obtain the question encoding representation, and then entering the decoder to obtain the answer decoding representation , the teacher-forcing loss is calculated according to the correct options during training; 2) Read the comprehension stream, use the greedy strategy to generate the vector representation of the problem encoding by the decoder alone, and input the problem into the encoder after splicing the options separately, and the corresponding The output question option representation is interactively fused with the extended vector representation, and the logit corresponding to each option is obtained from the obtained fusion result. During training, the cross entropy loss is calculated between these logit and the correct option, and the teacher‑forcing loss and cross entropy loss are combined. It is necessary to train and optimize the reading comprehension model. The invention uses a single data set for training, and improves the reading comprehension accuracy of multiple-choice questions. |
isCitedBy | http://rdf.ncbi.nlm.nih.gov/pubchem/patent/CN-114757154-A |
priorityDate | 2022-03-23-04:00^^<http://www.w3.org/2001/XMLSchema#date> |
type | http://data.epo.org/linked-data/def/patent/Publication |
Incoming Links
Predicate | Subject |
---|---|
isDiscussedBy | http://rdf.ncbi.nlm.nih.gov/pubchem/substance/SID419527388 http://rdf.ncbi.nlm.nih.gov/pubchem/compound/CID518696 |
Total number of triples: 18.