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

Emotion recogonition on text has wide applications. In this study we propose a method of emotion recognition at sentence level based on a relative large emotion annotation corpus (Ren-CECps). From this corpus, we get the emotion lexicons for the eight basic emotions (expect, joy, love, surprise, anxiety, sorrow, angry and hate). Statistics show that the emotion lexicons derived from Ren-CECps are used more often in real use of language for emotional expressions than HOWNET sentimental lexicons. Kernel methods are state-of-the-art for solving machine learning problems. Polynomial kernel (PK) method is used to compute the similarities between sentences and the eight emotion lexicons. Then the experiential knowledge derived from Ren-CECps is used to recognize whether the eight emotion categories are present in a sentence. This method obtain 62.7% F-measure.

Recognizing sentence emotions based on polynomial kernel method using Ren-CECps

To generate natural and less hardwired facial expression for humanoid robots, a facial motion imitation method, which transfers facial geometric characteristics of humans to robots, is proposed Firstly, for mapping servo control space into facial expression space, a forward kinematics model with sequence is built based on recurrent neural network Secondly, the process of expression imitation for humanoid robot is converted to an optimization with the proposed forward kinematics model and servo sequential constraints Finally, we conduct some experiments on the rationality of proposed model as well as the validity of optimization strategy Compared with the art-of-the-state of methods, the proposed method not only has smaller expression deviations with the performer, but also keeps smooth servo movements in dynamic facial imitation

Human-like facial expression imitation for humanoid robot based on recurrent neural network

Extraction of a large number of super-function (SF) is the most important factor in realizing SF based machine translation. This paper presents a method for automatic extraction of SF from a Japanese-English bilingual corpus. The extraction process uses a bilingual dictionary to match Japanese and English nouns in each sentence pair. The experimental results using a Japanese-English bilingual corpus show that this method performs very well in automatically extracting SF for machine translation. In addition, we evaluate the extracted SF in SF based machine translation.

Automatic super-function extraction for translation of spoken dialogue

Most of the previous studies focused on enriching text representation to address text classification (TC) task. However, conventional classification approaches with VSM (vector space model) on Chinese text study intensively only the words and their relationship in some specific corpus/dataset but ignore the basic concept of categories and the general knowledge behind the words learned and used to recognize entities by people. This paper focuses on enriching text representation and proposes a novel approach, which complements information from the online Chinese encyclopedia Baidu Baike for Chinese TC. The similarities between every text and each concept of categories and the most related words from Baidu Baike are added to the feature space. The performance of the proposed approach is measured on the Fudan University TC corpus, which is an imbalanced Chinese dataset. In the experiments, the proposed Baidu Baike-based concept similarity approach obtains promising results when compared with a previous research and the conventional method, with macro-precision of 90.31%, recall of 75.45%, and F1 score 80.32%, which are about 0.02%, 0.15%, 0.12%, respectively, higher than the conventional method, which obviously improves the recall for some small categories while keeping precision at high level and improving the macro F1 score. Moreover, the proposed approach has good expandability, so that many other knowledge bases could be integrated and many other concepts could be referred to improve the effectiveness. © 2016 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Hybrid Chinese text classification approach using general knowledge from Baidu Baike

In Chinese language processing, new words are particularly problematic. It is impossible to get a complete dictionary as new words can always be created. We proposed a unified dual-chain unequal-state CRF model to detect new words together with their part-of-speech in Chinese texts regardless of the word types such as compound words, abbreviation, person names, etc. The dual-chain unequal-state CRF model has two state chains with unequal number of states. The unequal state chains could model flexible hierarchical lexical information for both Chinese new word detection and POS tagging, and also integrate complex context features like the global information. The experimental results show that the proposed method is capable of detecting even low frequency new words and their parts-of-speech synchronously with satisfactory results.

Fuji Ren

Papers

Recognizing sentence emotions based on polynomial kernel method using Ren-CECps

Human-like facial expression imitation for humanoid robot based on recurrent neural network

Automatic super-function extraction for translation of spoken dialogue

Hybrid Chinese text classification approach using general knowledge from Baidu Baike

Dual-chain Unequal-state CRF for Chinese new word detection and POS tagging