Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-05-10 Cooperative journals: 《计算机应用研究》
Abstract: In Chinese written expression, there is no word segmentation between vocabularies, so the principle of writing (or called lexical features) is what it needs to process the segmentation of Chinese content. Former researches usually mark the lexical features into training content to improve the performance, which increases the manual processing flow and the workload of the algorithm transplantation. Based on Conditional Random Fields (CRF) and the simple tags, this paper improves the recognition performance by concluding the lexical features of Chinese and transforming them to complicated functions which used by CRF. Experiments show that applying complex lexical features in Chinese word segmentation can effectively improve recognition performance and provide a new way to improve the portability of recognition algorithms in applications.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-03 Cooperative journals: 《计算机应用研究》
Abstract: Most of the relation extraction approaches could not learn the long distance dependence information from the long sentences with entity co-occurrence. This paper proposes a new relation extraction model to solve this problem. This model was based on the recurrent convolutional neural network and the sentence-level attention mechanism. It used the Bi-GRU neural network to learn context vectors for words. And it adopted the piecewise maximum pooling method, which could obtain fine grained features. This paper conducted experiments on the NYT dataset. Experimental results demonstrate that the proposed method outperforms the baseline systems.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2019-01-03 Cooperative journals: 《计算机应用研究》
Abstract: Aiming at the problem of automatic identification of existing concepts and discovering new concepts in a specific field, a method based on conditional random field and information entropy is proposed. The conditional random field is used to predict the boundary of conceptual words in text. The candidates of the new concept can be selected with the comparison to the existing concepts in the dictionary and the probably location in text is found. Then the mutual information and the left and right entropy are used to judge the internal degree of integration and the boundary freedom of the concept in the concept window for discovering new professional concepts. Experiments show that the concept discovery using this method has a better effect than the method of using the conditional random field alone. The accuracy of the concept discovery based on word and words model is respectively improved by 20.06% and 46.54%.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-11-29 Cooperative journals: 《计算机应用研究》
Abstract: To improve the correct rate of image classification by convolutional neural network, a multi-model fusion convolutional neural network is proposed after research on the network structure. By extracting the output feature vectors of a single model and then fusion them, the new output feature vectors are obtained, and then a single classifier is set up to classify the images, and the accuracy of the classification is improved. The classification accuracy of single model compare with multi-model fusion, the accuracy of classification of multi-model fusion convolutional neural network is improved. The weight distribution of the last layer of the convolutional neural network is analyzed, and it is found that the weight distribution curve of the same model on different data sets is similar and the weight distribution curve of the network model with better classification effect is more gentle.