Image paragraph generation aims to generate a long description composed of multiple sentences,which is different from traditional image captioning containing only one sentence.Most of previous methods are dedicated to...Image paragraph generation aims to generate a long description composed of multiple sentences,which is different from traditional image captioning containing only one sentence.Most of previous methods are dedicated to extracting rich features from image regions,and ignore modelling the visual relationships.In this paper,we propose a novel method to generate a paragraph by modelling visual relationships comprehensively.First,we parse an image into a scene graph,where each node represents a specific object and each edge denotes the relationship between two objects.Second,we enrich the object features by implicitly encoding visual relationships through a graph convolutional network(GCN).We further explore high-order relations between different relation features using another graph convolutional network.In addition,we obtain the linguistic features by projecting the predicted object labels and their relationships into a semantic embedding space.With these features,we present an attention-based topic generation network to select relevant features and produce a set of topic vectors,which are then utilized to generate multiple sentences.We evaluate the proposed method on the Stanford image-paragraph dataset which is currently the only available dataset for image paragraph generation,and our method achieves competitive performance in comparison with other state-of-the-art(SOTA)methods.展开更多
Document processing in natural language includes retrieval,sentiment analysis,theme extraction,etc.Classical methods for handling these tasks are based on models of probability,semantics and networks for machine learn...Document processing in natural language includes retrieval,sentiment analysis,theme extraction,etc.Classical methods for handling these tasks are based on models of probability,semantics and networks for machine learning.The probability model is loss of semantic information in essential,and it influences the processing accuracy.Machine learning approaches include supervised,unsupervised,and semi-supervised approaches,labeled corpora is necessary for semantics model and supervised learning.The method for achieving a reliably labeled corpus is done manually,it is costly and time-consuming because people have to read each document and annotate the label of each document.Recently,the continuous CBOW model is efficient for learning high-quality distributed vector representations,and it can capture a large number of precise syntactic and semantic word relationships,this model can be easily extended to learn paragraph vector,but it is not precise.Towards these problems,this paper is devoted to developing a new model for learning paragraph vector,we combine the CBOW model and CNNs to establish a new deep learning model.Experimental results show that paragraph vector generated by the new model is better than the paragraph vector generated by CBOW model in semantic relativeness and accuracy.展开更多
Societal risk classification is a fundamental and complex issue for societal risk perception. To conduct societal risk classification, Tianya Forum posts are selected as the data source, and four kinds of representati...Societal risk classification is a fundamental and complex issue for societal risk perception. To conduct societal risk classification, Tianya Forum posts are selected as the data source, and four kinds of representations： string representation, term-frequency representation, TF-IDF representation and the distributed representation of BBS posts are applied. Using edit distance or cosine similarity as distance metric, four k-Nearest Neighbor （kNN） classifiers based on different representations are developed and compared. Owing to the priority of word order and semantic extraction of the neural network model Paragraph Vector, kNN based on the distributed representation generated by Paragraph Vector （kNN-PV） shows effectiveness for societal risk classification. Furthermore, to improve the performance of societal risk classification, through different weights, kNN-PV is combined with other three kNN classifiers as an ensemble model. Through brute force grid search method, the optimal weights are assigned to different kNN classifiers. Compared with kNN-PV, the experimental results reveal that Macro-F of the ensemble method is significantly improved for societal risk classification.展开更多
基金supported in part by National Natural Science Foundation of China(Nos.61721004,61976214,62076078 and 62176246).
文摘Image paragraph generation aims to generate a long description composed of multiple sentences,which is different from traditional image captioning containing only one sentence.Most of previous methods are dedicated to extracting rich features from image regions,and ignore modelling the visual relationships.In this paper,we propose a novel method to generate a paragraph by modelling visual relationships comprehensively.First,we parse an image into a scene graph,where each node represents a specific object and each edge denotes the relationship between two objects.Second,we enrich the object features by implicitly encoding visual relationships through a graph convolutional network(GCN).We further explore high-order relations between different relation features using another graph convolutional network.In addition,we obtain the linguistic features by projecting the predicted object labels and their relationships into a semantic embedding space.With these features,we present an attention-based topic generation network to select relevant features and produce a set of topic vectors,which are then utilized to generate multiple sentences.We evaluate the proposed method on the Stanford image-paragraph dataset which is currently the only available dataset for image paragraph generation,and our method achieves competitive performance in comparison with other state-of-the-art(SOTA)methods.
基金The authors would like to thank all anonymous reviewers for their suggestions and feedback.This work Supported by the National Natural Science,Foundation of China(No.61379052,61379103)the National Key Research and Development Program(2016YFB1000101)+1 种基金The Natural Science Foundation for Distinguished Young Scholars of Hunan Province(Grant No.14JJ1026)Specialized Research Fund for the Doctoral Program of Higher Education(Grant No.20124307110015).
文摘Document processing in natural language includes retrieval,sentiment analysis,theme extraction,etc.Classical methods for handling these tasks are based on models of probability,semantics and networks for machine learning.The probability model is loss of semantic information in essential,and it influences the processing accuracy.Machine learning approaches include supervised,unsupervised,and semi-supervised approaches,labeled corpora is necessary for semantics model and supervised learning.The method for achieving a reliably labeled corpus is done manually,it is costly and time-consuming because people have to read each document and annotate the label of each document.Recently,the continuous CBOW model is efficient for learning high-quality distributed vector representations,and it can capture a large number of precise syntactic and semantic word relationships,this model can be easily extended to learn paragraph vector,but it is not precise.Towards these problems,this paper is devoted to developing a new model for learning paragraph vector,we combine the CBOW model and CNNs to establish a new deep learning model.Experimental results show that paragraph vector generated by the new model is better than the paragraph vector generated by CBOW model in semantic relativeness and accuracy.
基金This study is supported by the National Key Research and Development Program of China under grant No. 2016YFB1000902 and National Natural Science Foundation of China under grant Nos. 61473284, 71601023 and 71371107.
文摘Societal risk classification is a fundamental and complex issue for societal risk perception. To conduct societal risk classification, Tianya Forum posts are selected as the data source, and four kinds of representations： string representation, term-frequency representation, TF-IDF representation and the distributed representation of BBS posts are applied. Using edit distance or cosine similarity as distance metric, four k-Nearest Neighbor （kNN） classifiers based on different representations are developed and compared. Owing to the priority of word order and semantic extraction of the neural network model Paragraph Vector, kNN based on the distributed representation generated by Paragraph Vector （kNN-PV） shows effectiveness for societal risk classification. Furthermore, to improve the performance of societal risk classification, through different weights, kNN-PV is combined with other three kNN classifiers as an ensemble model. Through brute force grid search method, the optimal weights are assigned to different kNN classifiers. Compared with kNN-PV, the experimental results reveal that Macro-F of the ensemble method is significantly improved for societal risk classification.
