Junsheng Zhou

AMR Parsing with an Incremental Joint Model.

To alleviate the error propagation in the traditional pipelined models for Abstract Meaning Representation (AMR) parsing, we formulate AMR parsing as a joint task that performs the two subtasks: concept identification and relation identification simultaneously. To this end, we first develop a novel componentwise beam search algorithm for relation identification in an incremental fashion, and then incorporate the decoder into a unified framework based on multiple-beam search, which allows for the bi-directional information flow between the two subtasks in a single incremental model. Experiments on the public datasets demonstrate that our joint model significantly outperforms the previous pipelined counterparts, and also achieves better or comparable performance than other approaches to AMR parsing, without utilizing external semantic resources.

An enhanced short text categorization model with deep abundant representation

Short text categorization is a crucial issue to many applications, e.g., Information Retrieval, Question-Answering System, MRI Database Construction and so forth. Many researches focus on data sparsity and ambiguity issues in short text categorization. To tackle these issues, we propose a novel short text categorization strategy based on abundant representation, which utilizes Bi-directional Recurrent Neural Network(Bi-RNN) with Long Short-Term Memory(LSTM) and topic model to catch more contextual and semantic information. Bi-RNN enriches contextual information, and topic model discovers more latent semantic information for abundant text representation of short text. Experimental results demonstrate that the proposed model is comparable to state-of-the-art neural network models and method proposed is effective.

A Search-Based Dynamic Reranking Model for Dependency Parsing

We propose a novel reranking method to extend a deterministic neural dependency parser. Different to conventional k-best reranking, the proposed model integrates search and learning by utilizing a dynamic action revising process, using the reranking model to guide modification for the base outputs and to rerank the candidates. The dynamic reranking model achieves an absolute 1.78% accuracy improvement over the deterministic baseline parser on PTB, which is the highest improvement by neural rerankers in the literature.

A graph-based approach for semantic similar word retrieval

Semantic relatedness or semantic similarity between words is an important basic issue for many Natural Language Processing (NLP) applications, such as sentence retrieval, word sense disambiguation, question answering, and so on. This research issue attracts many researchers, but most of studies focus on improving the effectiveness, i.e., applying kinds of techniques to improve precision (effectiveness) but not efficiency. To tackle the problem, we propose to address the efficiency issue, that how to efficiently find top-k most semantic similar words to the query for a given dataset. This issue is very important for real applications especially for current big data. Efficient graph-based approaches on searching top-k semantic similar words are proposed in this paper. The results demonstrate that the proposed model can perform significantly better than baseline method.

Chinese event descriptive clause splitting with structured SVMs

Chinese event descriptive clause splitting is a novel task in Chinese information processing. Different from English clause splitting problem, Chinese event descriptive clause splitting aims at recognizing the high-level clauses. In this paper, we present a Chinese clause splitting system with a discriminative approach. By formulating the Chinese clause splitting task as a sequence labeling problem, we apply the structured SVMs model to Chinese clause splitting. Compared with other two baseline systems, our approach gives much better performance.

A Multi-view Approach for Relation Extraction

Relation extraction is an important problem in information extraction. In this paper, we explore a multi-view strategy for relation extracting task. Motivated by the fact, as in work of Jiang and Zhais [1], that combining different feature subspaces into a single view does not generate much improvement, we propose a two-stage multi-view learning approach. First, we learn two different classifiers from two different views of relation instances: sequence representation and syntactic parse tree representation, respectively. Then, a meta-learner is trained using the meta data constructed along with other contextual information to achieve a strong predictive performance, as the final classification model. The experimental results conducted on ACE 2005 corpus show that the multi-view approach outperforms each single-view one for relation extraction task.

An effective approach for automatic interpretation of Chinese nominal compounds

Purpose Automatic interpretation of Nominal Compounds is a crucial issue for many applications, for example, sentence understanding, machine translation, question-answering system and so forth. Many automatic interpretation models of Nominal Compounds use the strategies based on verbs or rules to obtain the interpretation of compounds. However, the performances of these models are still limited. The purpose of this paper is to propose an effective approach for automatic interpretation of Chinese nominal compounds. Design/methodology/approach The authors propose a top-down and bottom-up model based on rules and large-scale corpus for automatic interpretation of Nominal Compounds. Findings Experimental results demonstrate that the proposed model outperforms the state-of-the-art automatic interpretation model. Originality/value The paper is an up-to-date study of automatic interpretation for Nominal Compounds. It can help people understand the meaning of Nominal Compounds in reading. With a better understanding of Nominal Compounds, we can discover more hidden knowledge in them.

Syntactic Categorization and Semantic Interpretation of Chinese Nominal Compounds

The development in society and technology generates more Nominal Compounds to represent new concepts in various domains. Earlier literature in linguistic studies has gathered and established several syntactic categories of Nominal Compounds, which can be used for automatic syntactic categorization of these compounds. This paper is focused on Nominal Compounds of head-modifier construction because experiments show that most Nominal Compounds are head-modifier constructions. Based on the combination of templates and word similarity, this paper proposes an algorithm for automatic semantic interpretation which improves the recall ratio while maintaining the precision ratio. The results of syntactic categorization and automatic semantic interpretation of the Nominal Compounds are also applied in dependency parsing and machine translation.

Research on interpretation of nominal compound

Nominal compounds which constituted of two nouns together are very common in reading materials or web pages. The interpretation of these compounds can help us know the meaning of a text or sentences. Traditional approaches utilized the method based on verbs and rules to obtain the interpretation of compounds with low recall. So we investigate an interpretation method based on similarity which makes use of the interpretation templates and similar words to achieve the automatic interpretation. Experimental results show that our method can interpret these nominal compounds with a relatively high precision (84.91%), give an increase of 10.48% in recall than the general method, which contributes to the overall nominal compound recall improvement significantly.

Towards effective web page classification

In order to manage and organize information on the web, we propose a novel web page classification strategy integrating topic model and SVM. We use topic model to harness the implicit information on web pages for feature extraction. Accuracy of the strategy is 84.15%, 2.23% superior to the traditional classification strategy based on CHI.