Yu Lun Hsieh

Linguistic Template Extraction for Recognizing Reader-Emotion and Emotional Resonance Writing Assistance

In this paper, we propose a flexible principle-based approach (PBA) for reader-emotion classification and writing assistance. PBA is a highly automated process that learns emotion templates from raw texts to characterize an emotion and is comprehensible for humans. These templates are adopted to predict reader-emotion, and may further assist in emotional resonance writing. Results demonstrate that PBA can effectively detect reader-emotions by exploiting the syntactic structures and semantic associations in the context, thus outperforming wellknown statistical text classification methods and the state-of-the-art reader-emotion classification method. Moreover, writers are able to create more emotional resonance in articles under the assistance of the generated emotion templates. These templates have been proven to be highly interpretable, which is an attribute that is difficult to accomplish in traditional statistical methods.

A semantic frame-based intelligent agent for topic detection

Detecting the topic of documents can help readers construct the background of the topic and facilitate document comprehension. In this paper, we propose a semantic frame-based topic detection (SFTD) that simulates such process in human perception. We take advantage of multiple knowledge sources and extracted discriminative patterns from documents through a highly automated, knowledge-supported frame generation and matching mechanisms. Using a Chinese news corpus containing over 111,000 news articles, we provide a comprehensive performance evaluation which demonstrates that our novel approach can effectively detect the topic of a document by exploiting the syntactic structures, semantic association, and the context within the text. Experimental results show that SFTD is comparable to other well-known topic detection methods.

Effective pseudo-relevance feedback for language modeling in extractive speech summarization

Extractive speech summarization, aiming to automatically select an indicative set of sentences from a spoken document so as to concisely represent the most important aspects of the document, has become an active area for research and experimentation. An emerging stream of work is to employ the language modeling (LM) framework along with the Kullback-Leibler divergence measure for extractive speech summarization, which can perform important sentence selection in an unsupervised manner and has shown preliminary success. This paper presents a continuation of such a general line of research and its main contribution is two-fold. First, by virtue of pseudo-relevance feedback, we explore several effective sentence modeling formulations to enhance the sentence models involved in the LM-based summarization framework. Second, the utilities of our summarization methods and several widely-used methods are analyzed and compared extensively, which demonstrates the effectiveness of our methods.

Sentiment analysis on Chinese movie review with distributed keyword vector representation

In the area of national language processing, performing machine learning technique on customer or movie review for sentiment analysis has been? frequently tried. While methods such as? support vector machine (SVM) were much favored in the 2000s, recently there is a steadily rising percentage of implementation with vector representation and artificial neural network. In this article we present an approach to implement word embedding method to conduct sentiment analysis on movie review from a renowned bulletin board system forum in Taiwan. After performing log-likelihood ratio (LLR) on the corpus and selecting the top 10000 most related keywords as representative vectors for different sentiments, we use these vectors as the sentiment classifier for the testing set. We achieved results that are not only comparable to traditional methods like Naïve Bayes and SVM, but also outperform Latent Dirichlet Allocation, TF-IDF and its variant. It also tops the original LLR with a substantial margin.

Distributed keyword vector representation for document categorization

In the age of information explosion, efficiently categorizing the topic of a document can assist our organization and comprehension of the vast amount of text. In this paper, we propose a novel approach, named DKV, for document categorization using distributed real-valued vector representation of keywords learned from neural networks. Such a representation can project rich context information (or embedding) into the vector space, and subsequently be used to infer similarity measures among words, sentences, and even documents. Using a Chinese news corpus containing over 100,000 articles and five topics, we provide a comprehensive performance evaluation to demonstrate that by exploiting the keyword embeddings, DKV paired with support vector machines can effectively categorize a document into the predefined topics. Results demonstrate that our method can achieve the best performances compared to several other approaches.

A frame-based approach for reference metadata extraction

In this paper, we propose a novel frame-based approach (FBA) and use reference metadata extraction as a case study to demonstrate its advantages. The main contributions of this research are three-fold. First, the new frame matching algorithm, based on sequence alignment, can compensate for the shortcomings of traditional rule-based approach, in which rule matching lacks flexibility and generality. Second, an approximate matching is adopted for capturing reasonable abbreviations or errors in the input reference string to further increase the coverage of the frames. Third, experiments conducted on extensive datasets show that the same knowledge framework performed equally well on various untrained domains. Comparing to a widely-used machine learning method, Conditional Random Fields (CRFs), the FBA can drastically reduce the average field error rate across all four independent test sets by 70% (2.24% vs. 7.54%).

Semantic Frame-Based Natural Language Understanding for Intelligent Topic Detection Agent

Detecting the topic of documents can help readers construct the background of the topic and facilitate document comprehension. In this paper, we proposed a semantic frame-based method for topic detection that simulates such process in human perception. We took advantage of multiple knowledge sources and identified discriminative patterns from documents through frame generation and matching mechanisms. Results demonstrated that our novel approach can effectively detect the topic of a document by exploiting the syntactic structures, semantic association, and the context within the text. Moreover, it also outperforms well-known topic detection methods.

A Position-Aware Language Modeling Framework for Extractive Broadcast News Speech Summarization

Extractive summarization, a process that automatically picks exemplary sentences from a text (or spoken) document with the goal of concisely conveying key information therein, has seen a surge of attention from scholars and practitioners recently. Using a language modeling (LM) approach for sentence selection has been proven effective for performing unsupervised extractive summarization. However, one of the major difficulties facing the LM approach is to model sentences and estimate their parameters more accurately for each text (or spoken) document. We extend this line of research and make the following contributions in this work. First, we propose a position-aware language modeling framework using various granularities of position-specific information to better estimate the sentence models involved in the summarization process. Second, we explore disparate ways to integrate the positional cues into relevance models through a pseudo-relevance feedback procedure. Third, we extensively evaluate various models originated from our proposed framework and several well-established unsupervised methods. Empirical evaluation conducted on a broadcast news summarization task further demonstrates performance merits of the proposed summarization methods.

IASL valence-arousal analysis system at IALP 2016 shared task: Dimensional sentiment analysis for Chinese words

Sentiment lexicons with valence-arousal ratings are useful resources for the development of dimensional sentiment applications. In order to solve the significant lack of Chinese valence and arousal lexicons, the objective of the DSAW is to automatically acquire the valence-arousal ratings of Chinese affective words. In this task, we develop a novel approach that integrate word embeddings into a graph-based model with K-Nearest Neighbor to identify both valence and arousal dimensions. We also propose to use character embeddings to represent unseen words, which is a major challenge in collecting large corpora. The evaluation results demonstrate that our system is effective in dimensional sentiment analysis for Chinese words with 0.847 and 1.281 mean absolute error (MAE) for valence and arousal respectively.

Principle-Based Approach for Semi-Automatic Construction of a Restaurant Question Answering System from Limited Datasets

Question answering (QA) is an important research issue in natural language processing, and most state-of the-art question answering systems are based on statistical models. After witnessing recent achievements in Artificial Intelligent (AI), many businesses wish to apply those techniques to an automatic QA system that is capable of providing 24-hour customer services for their clients. However, one imminent problem is the lack of labeled training data for the specific domain. To address this issue, we propose to combine a knowledge-based approach and an automatic principle generation process to build a QA system from limited resources. Experiments conducted on a Mandarin Restaurant dataset show that our system achieves an average accuracy of 44% for 10 question types. It demonstrates that our approach can provide an effective tool when creating a QA system.