Data Mining - Concepts and Techniques.

We will review some of the major results in random graphs and some of the more challenging open problems. We will cover algorithmic and structural questions. We will touch on newer models, including those related to the WWW.

Random graphs

“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Social Network Analysis

Emotion and Adaptation

This paper propose a lightweight convolution neural network based on the depthwise separable convolution and the improved Linear Bottlenecks block. Running time and space complexity of the model are reduced by designing the network structure carefully. The neural network model can run fluently on the device of ARM architecture to complete tasks such as object recognition and object detection. 20 categories of common objects in life from ImageNet are randomly selected to carry out recognition experiments on Raspberry Pi 3B+. The experimental results show that the Top1 accuracy of the model in the testing dataset can reach 91%, and the average recognition speed for 224*224 pixel images can reach 176ms.

Realization and Improvement of Object Recognition System on Raspberry Pi 3B

The growing interest in affective computing AC brings a lot of valuable research topics that can meet different application demands in enterprise systems. The present study explores a sub area of AC techniques – textual emotion recognition for enhancing enterprise computing. Multi-label emotion recognition in text is able to provide a more comprehensive understanding of emotions than single label emotion recognition. A representation of ‘emotion state in text’ is proposed to encompass the multidimensional emotions in text. It ensures the description in a formal way of the configurations of basic emotions as well as of the relations between them. Our method allows recognition of the emotions for the words bear indirect emotions, emotion ambiguity and multiple emotions. We further investigate the effect of word order for emotional expression by comparing the performances of bag-of-words model and sequence model for multi-label sentence emotion recognition. The experiments show that the classification results under sequence model are better than under bag-of-words model. And homogeneous Markov model showed promising results of multi-label sentence emotion recognition. This emotion recognition system is able to provide a convenient way to acquire valuable emotion information and to improve enterprise competitive ability in many aspects.

Textual emotion recognition for enhancing enterprise computing

Emotion recognition based on speech characteristics generally relies on prosodic information. However, utterances with different emotions in speech have similar prosodic features, so it is difficult to recognize emotion by using only prosodic features.

In this paper, we propose a novel approach to emotion recognition that considers both prosodic and linguistic features. First, possible emotions are output by clustering-based emotion recognizer, which only uses prosodic features. Then, subtitles given by the speech recognizer are input for another emotion recognizer based on the "Association Mechanism." It outputs a possible emotion by using only linguistic information. Lastly, the intersection of the two sets of possible emotions is integrated into the final result.

Experimental results showed that the proposed method achieved higher performance than either prosodic- or linguistic-based emotion recognition. In a comparison with manually labeled data, the F-measure was 32.6%. On the other hand, the average of F-measures of labeled data given by other humans was 42.9%. This means that the proposed method performed at 75.9% in relation to human ability.

A novel emotion recognizer from speech using both prosodic and linguistic features

Abstract. Emotion is strongly subjective, and different parts of the image may have a different degree of impact on emotion. The key for solving image emotion recognition is to effectively mine different discriminative local regions. We present a deep architecture to guide the network to extract discriminative and diverse affective semantic information. First, training a full convolutional network with a cross-channel max pooling strategy (CCMP) to extract discriminative feature maps. Second, to ensure that most of the discriminative sentiment regions are located accurately, we add a module consisting of the convolution layer and the CCMP. After obtaining the discriminative regions of the first module, the feature elements corresponding to the discriminant regions are erased, and then the erased features are fed into the second module. Such adversarial erasure operation can force the network to discover different sentiment discriminative regions. Third, an adaptive feature fusion mechanism is proposed to better integrate discriminative and diverse sentiment representations. Sufficient experiments are conducted on the benchmark datasets FI, EmotionROI, Instagram, and Twitter1 to achieve 72.17%, 61.13%, 81.97%, and 85.44% recognition accuracies, respectively. The results of the experiment demonstrate that the proposed network outperforms the state-of-the-art results.

Feature discrimination and diversification for image emotion recognition

Neural networks have received considerable attention in sentence similarity measuring systems due to their efficiency in dealing with semantic composition. However, existing neural network methods are not sufficiently effective in capturing the most significant semantic information buried in an input. To address this problem, a novel weighted-pooling attention layer is proposed to retain the most remarkable attention vector. It has already been established that long short-term memory and a convolution neural network have a strong ability to accumulate enriched patterns of whole sentence semantic representation. First, a sentence representation is generated by employing a siamese structure based on bidirectional long short-term memory and a convolutional neural network. Subsequently, a weighted-pooling attention layer is applied to obtain an attention vector. Finally, the attention vector pair information is leveraged to calculate the score of sentence similarity. An amalgamation of both, bidirectional long short-term memory and a convolutional neural network has resulted in a model that enhances information extracting and learning capacity. Investigations show that the proposed method outperforms the state-of-theart approaches to datasets for two tasks, namely semantic relatedness and Microsoft research paraphrase identification. The new model improves the learning capability and also boosts the similarity accuracy as well. key words: sentence similarity, sentence embedding, deep learning, long short-term memory, convolutional neural network

Fuji Ren

Papers

Realization and Improvement of Object Recognition System on Raspberry Pi 3B

Textual emotion recognition for enhancing enterprise computing

A novel emotion recognizer from speech using both prosodic and linguistic features

Feature discrimination and diversification for image emotion recognition

Sentence-Embedding and Similarity via Hybrid Bidirectional-LSTM and CNN Utilizing Weighted-Pooling Attention