| 近藤 雅芳 | D, 中間発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: Adjoined Convolution Neural Network For Long Text Summarization
abstract: We propose a new neural encoding architecture for abstractive text summarization, called an adjoint convolutional neural network(ACNN). Although existing convolutional neural network (CNN)-based encoders circumvent the vanishing gradient that plagues recurrent BiLSTM encoders, they do not adequately capture word order in input text. ACNN processes the left and right local contexts of input words separately to better encode word order information, after which the encodings of the input word and the contexts are mixed with GRU-like gate functions. The number of parameters can be reduced by weight sharing across layers, and experimental results show that the quality of the generated summaries is not sacrificed. Experimental results also show that ACNN outperforms the state-of-the-art BiLSTM encoder in a standard text summarization task. In a different task with longer input texts, ACNN performs better than BiLSTM and existing non-recurrent encoders, including the gated CNNs and the encoder of the more recent Transformer model. language of the presentation: Japanese | |||
| PHI VAN THUY | D, 中間発表 | 自然言語処理学 | 松本 裕治, 中村 哲, 新保 仁, 進藤 裕之 |
|
title: Ranking-Based Automatic Seed Selection and Noise Reduction for Weakly Supervised Relation Extraction
abstract: This work addresses the tasks of automatic seed selection for bootstrapping relation extraction, and noise reduction for distantly supervised relation extraction. We first point out that these tasks are related. Then, inspired by ranking relation instances and patterns computed by the HITS algorithm, and selecting cluster centroids using the K-means, LSA, or NMF method, we propose methods for selecting the initial seeds from an existing resource, or reducing the level of noise in the distantly labeled data. Experiments show that our proposed methods achieve a better performance than the baseline systems in both tasks. language of the presentation: English | |||