Authentication email has already been sent, please check your email box: and activate it as soon as possible.
You can login to My Profile and manage your email alerts.
If you haven’t received the email, please:
|
|
There are 46 papers published in subject: > since this site started. |
Select Subject |
Select/Unselect all | For Selected Papers |
Saved Papers
Please enter a name for this paper to be shown in your personalized Saved Papers list
|
1. Online Chinese Polyphone Disambiguation with Progressive Neural Networks | |||
ZHANG Yi-Fei | |||
Computer Science and Technology 16 March 2023 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:There has been plenty of research on Chinese polyphonic disambiguation (CPD) problems. However, badcases are always found in real-life products. To fix such bad cases without affecting system performance on known cases is a rigid demand. In this paper, continual learning is introduced to CPD problems, and Progressive Neural Networks (PNN) is used to learn new knowledge from bad cases without sacrificing system performance on old datasets. The experimental results show that the proposed method can repair the badcase without forgetting the feature of the original dataset. Compared with the traditional finetune method, the accuracy of the model on the old dataset decreases by nearly 20\%. Our method can ensure that the accuracy of the original dataset just decreases by about 0.3\% after learning the new feature data, and the time consumption is acceptable. Potential improvements like weight pruning are also discussed. | |||
TO cite this article:ZHANG Yi-Fei. Online Chinese Polyphone Disambiguation with Progressive Neural Networks[OL].[16 March 2023] http://en.paper.edu.cn/en_releasepaper/content/4759451 |
2. Using perceptual loss to improve video caption generation | |||
Ye Zhou,Hu Yanzhu | |||
Computer Science and Technology 10 March 2022 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:In the task of video caption generation by deep learning methods, Binary Cross Entropy is often used as the loss function for end-to-end model training, and evaluation metrics such as BLEU, which do not consider semantics, are used as the main judge of model performance, which is quite negative to the actual semantic effect of the generated captions. In contrast, in the image generation task, there is perceptual loss that can be used to learn hidden semantic knowledge for inferencing. This paper is inspired by the perceptual loss in image generation task and proposes to use the perceptual loss function to help optimize statements for video caption generation. On the MSR-VTT dataset, compared with the traditional method using only Binary Cross Entropy, the method proposed in our paper makes it possible to improve the SPICE metrics while ensuring the effect of metrics such as BLEU. | |||
TO cite this article:Ye Zhou,Hu Yanzhu. Using perceptual loss to improve video caption generation[OL].[10 March 2022] http://en.paper.edu.cn/en_releasepaper/content/4756787 |
3. Deep Parallel Neural Network for Event Temporal Relation Classification | |||
Sun Chang,Gao Sheng | |||
Computer Science and Technology 13 April 2021 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:Temporal relation classification is an active natural language understanding research field. Partly because of the lack of high-quality datasets, most of the methods proposed in other papers do not use neural network or they only use simple neural network with external knowledge. In this task, we use Convolution Neural Network (CNN) to discover the complex interaction between events and extract time-related keyword information, and we use Long Short-Term Memory (LSTM) to capture temporal context information in sentences, and then connect the two kinds of information for temporal relation classification. We don\'t use any external knowledge, including dependency trees. In the latest MATRES dataset, the performance of our model is better than the state-of-the-art result. | |||
TO cite this article:Sun Chang,Gao Sheng. Deep Parallel Neural Network for Event Temporal Relation Classification[OL].[13 April 2021] http://en.paper.edu.cn/en_releasepaper/content/4754461 |
4. A Dual-Attentive and Hybrid Word-Character Model for Chinese Short Text Summarization | |||
Li Yufeng,Xu Weiran | |||
Computer Science and Technology 24 December 2020 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:Automatic text summarization is an important field for NLP, which includes the extractive and the abstractive method. Among many languages, Chinese has many special properties, such as rich character semantic expressions, flexible abbreviation. Moreover, insufficient training samples are also a problem. In this paper, we propose a dual-attentive and word-character Chinese text summarization model. The hybrid word-character approach (HWC) will preserve the advantages of both word based and character-based representations, which are very suitable for Chinese language. And the extractive and abstractive methods are combined to accurately capture the key information and gain the essence of articles with less supervised samples. We evaluate our model using the ROUGE evaluation on a widely used Chinese Dataset LCSTS2.0. The experimental results show that the model is very effective. | |||
TO cite this article:Li Yufeng,Xu Weiran. A Dual-Attentive and Hybrid Word-Character Model for Chinese Short Text Summarization[OL].[24 December 2020] http://en.paper.edu.cn/en_releasepaper/content/4753279 |
5. Relation Extraction with Domain Adversarial Neural Network and Graphical Model | |||
MA Kuo,ZHANG Xi | |||
Computer Science and Technology 23 July 2020 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:People read a comment on the web to learn about a product or some news, and the adjectives and nouns in these comments express important information. When we extract the adjective and nouns in the comments, If we can determine that there is indeed a relationship between the adjective and the noun, it will be very helpful for us to understand the comment. This thesis is all about extracting these word pairs and using transfer learning to extract them more quickly and accurately. This adjective and noun pair may undergo some changes in their relationship in different domains. This thesis considers the different domains to identify whether they are related or not. In this paper we propose an adversarial neural network approach with the help of a graphical model, DANN-G. This method considers the relationship between the bags well, the relationship within the bags, and thus reduces the noise caused by remote supervision in the common methods of relationship extraction. Our model has improved in the five major data sets of Amazon. | |||
TO cite this article:MA Kuo,ZHANG Xi. Relation Extraction with Domain Adversarial Neural Network and Graphical Model[OL].[23 July 2020] http://en.paper.edu.cn/en_releasepaper/content/4752560 |
6. A Keyword extraction method based on Neural Networks with Joint Training | |||
You Huanying,She Chundong,Liu Shaohua | |||
Computer Science and Technology 07 March 2020 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:Keyword extraction technology has gradually become a hot research problem in Natural Language Processing (NLP) and Information Retrieval. Many language tasks are inseparable from keyword extraction technology, such as long text classification, automatic summary, machine translation, dialogue system, etc. In this paper, we design a keyword extraction algorithm that can combine the benefits of both memorization and generalization. Our model contains a linear model and a deep neural networks. The linear model learns the relationship between statistic features and keywords, which can make full use of the memory capabilities of the shallow model. In the deep component, we feed the projection vector of words on the text to deep neural networks, which can enhance the generalization ability of the model. With the joint training of the linear model and the deep neural networks, our model achieves higher accuracy and scalability. Our method is compared with Frequency, Term Frequency-Inverse Document Frequency (TF-IDF) and TextRank. On the same batch of test dataset, our model is superior to the baseline model in Precision, Recall, and F-score, respectively. | |||
TO cite this article:You Huanying,She Chundong,Liu Shaohua. A Keyword extraction method based on Neural Networks with Joint Training[OL].[ 7 March 2020] http://en.paper.edu.cn/en_releasepaper/content/4751038 |
7. Conversational Recommendation System based on Sentiment Analysis | |||
LI Xinsheng,LI Jian | |||
Computer Science and Technology 25 February 2020 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:The combination of the recommender system and dialogue system which called the conversational recommendation system is a growing interest. Tosolve the problem that it is difficult to obtain users' tastes in conversational recommendation systems. A sentiment analysis method is proposed in our conversational recommendation model to get user preferences. A sentiment analysis dataset is created and the model uses a sentiment analysis approach to obtain a movie seeker\'s preferences and make a recommendation. Experimentresults show that our sentiment analysis model yields a better performance of 0.8362(F1 score) than the baseline(0.7802) and other models. Thus, the movie recommended by our system can meet the needs of users better. | |||
TO cite this article:LI Xinsheng,LI Jian. Conversational Recommendation System based on Sentiment Analysis[OL].[25 February 2020] http://en.paper.edu.cn/en_releasepaper/content/4750892 |
8. Attention based lattice bilstm model for Chinese named entity recognition | |||
CAO Xiaofei,YANG Juan,YANG Juan | |||
Computer Science and Technology 18 December 2019 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:A recently proposed model named Lattice LSTM has focused on integrating segmentation information into the long short-term memory (LSTM) network. However, it can only affect the subsequent character sequence of each character in the sequence from the level of word granularity, which results in insufficient extraction of word segmentation information. Besides, features of characters extracted by LSTM are given the same weight when transferred to the conditional random field (CRF) layer, the key semantic information does not receive much consideration. To solve the above problems, a novel neural network model is proposed in this paper which improves the original lattice model (Att-Lattice BiLSTM) with bidirectional long short-term memory based on the attention mechanism. An information path is added from the end character of word to the start character of word in the back propagation of LSTM, which integrates the word boundary information into both the start and end character of the word during bidirectional transfer of LSTM network, introducing the word information comprehensively. Moreover, this new model allows seamlessly incorporating attention mechanism to capture relatively important semantic feature automatically. Meanwhile, two strategies are provided to aggregate the bidirectional LSTM layers output to integrate semantic features effectively. Experimental results on four data sets show that the proposed model performs better than other most advanced models. | |||
TO cite this article:CAO Xiaofei,YANG Juan,YANG Juan. Attention based lattice bilstm model for Chinese named entity recognition[OL].[18 December 2019] http://en.paper.edu.cn/en_releasepaper/content/4750113 |
9. A Effective Bidirectional Mechanism with Pooling for Universial Sentence Representations | |||
Dian Jiao,Sheng Gao | |||
Computer Science and Technology 11 April 2019 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:BiLSTM with max pooling is adopted as a well-performed supervised universal sentence encoder. Max pooling is a common mechanism to get a fixed-size sentence representation. But we find that the max pooling for sentence encoder discards some useful backward and forward information at each time step and depends on a large number of parameters. In this paper, we propose an improved pooling mechanism based on max pooling for universal sentence encoder. The proposed model uses three kinds of methods to refine the backward and forward information at each time step, and then use a max-pooling layer or attention mechanism to obtain a fixed-size sentence representation from variable-length refined hidden states. Experiments conducted on Stanford Natural Language Inference (SNLI) Corpus, and we use it as a pretrained universal sentence encoder for transfer tasks. Experiments show that our model with less parameters performs better. | |||
TO cite this article:Dian Jiao,Sheng Gao. A Effective Bidirectional Mechanism with Pooling for Universial Sentence Representations[OL].[11 April 2019] http://en.paper.edu.cn/en_releasepaper/content/4748343 |
10. Improving Text Models with Latent Feature Vector Representations | |||
Peng Huaijin,Wang Jing,Shen Qiwei | |||
Computer Science and Technology 26 March 2019 | |||
Show/Hide Abstract | Cite this paper︱Full-text: PDF (0 B) | |||
Abstract:Probabilistic topic models are widely used to discover potential topics in a collection of documents, while latent feature vector representations have been used to achieve high performance in many NLP tasks. In this paper, we first make document topic vector representations by combining LDA and Topic2Vec, and then we perform document representations based on the topic vectors and the document vectors obtained through Doc2Vec training. Experimental results show that our new model has produced significant improvements in topic consistency and document classification tasks. | |||
TO cite this article:Peng Huaijin,Wang Jing,Shen Qiwei. Improving Text Models with Latent Feature Vector Representations[OL].[26 March 2019] http://en.paper.edu.cn/en_releasepaper/content/4747891 |
Select/Unselect all | For Selected Papers |
Saved Papers
Please enter a name for this paper to be shown in your personalized Saved Papers list
|
About Sciencepaper Online | Privacy Policy | Terms & Conditions | Contact Us
© 2003-2012 Sciencepaper Online. unless otherwise stated