A Deep Multiple View Sentence Representation Model for Question Answering
| Autor: | Hongguang Li, Wenfeng Tian, Jun Li |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Matching (statistics)
Word embedding Artificial neural network business.industry Computer science Feature selection 010501 environmental sciences computer.software_genre Semantics 01 natural sciences Convolutional neural network 030507 speech-language pathology & audiology 03 medical and health sciences Knowledge extraction Question answering Artificial intelligence 0305 other medical science business computer Natural language processing Sentence 0105 earth and related environmental sciences |
| Zdroj: | 2018 37th Chinese Control Conference (CCC). |
| DOI: | 10.23919/chicc.2018.8483750 |
| Popis: | Question answering (QA) between humans and computers is regarded as one of the most hardcore problems in computer science, which involves interdisciplinary techniques in natural language processing. Existing deep models rely on a single sentence representation or multiple granularity representations for question answering matching, which cannot capture the semantic information well in the question answering matching process. To solve this problem, we propose a new deep multiple view sentence representation model (DMVSR) to match two question answering semantic sentences. After pre-processed by word embedding, each QA semantic sentence representation is generated by a bidirectional long short term memory (Bi-LSTM) and Convolution neural network (CNN). Through k-Max pooling and a multi-layer perceptron, the final QA matching score is produced by aggregating interactions. Our model has several advantages: (1) Using Bi-LSTM to capture the semantic information; (2) Using CNN to implement feature extraction and feature selection in semantic space; (3) Matching QA sentence representation by aggregate interactions with semantic information. In the experiments, we investigate the effectiveness of the proposed deep neural network structures of all different evidence. We demonstrate significant performance improvement against a series of standard and state-of-art baselines in terms of MAP, nDCG@3 and nDCG@5. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|