Search

In this paper, we proposed a new method named adjacency based ordering to order sentences for summarization tasks. Given a group of sentences to be organized into the summary, connectivity of each pair of sentences is learned from source documents. Then a top-first strategy is implemented to define the sentence ordering. It provides a solution of ordering texts while other information except the source documents is not available. We compared this method with other existing sentence ordering methods. Experiments and evaluations are made on data collection of DUC04. The results show that this method distinctly outperforms other existing sentence ordering methods. Its low input requirement also makes it capable to most summarization and text generation tasks.

In this paper we propose a clustering-based hybrid approach for multi-document summarization which integrates sentence clustering, local recommendation and global search. For sentence clustering, we adopt a stability-based method which can determine the optimal cluster number automatically. We weight sentences with terms they contain for local sentence recommendation of each cluster. For global selection, we propose a global criterion to evaluate overall performance of a summary. Thus the sentences in the final summary are determined by not only the configuration of individual clusters but also the overall performance. This approach successfully gets top-level performance running on corpus of DUC04.

This paper introduces a Mutual Information Independence Model (MIIM) and proposes a feature relaxation principle to resolve the data sparseness problem in MIIM-based named entity recognition via hierarchical features. In this way, a named entity recognition system with better performance and better portability can be achieved. Evaluation of our system on MUC-6 and MUC-7 English named entity tasks achieves F-measures of 96.1% and 93.7% respectively. It also shows that 20K words of training data would have given the performance of 90 percent with the hierarchical structure in the features compared with 30K words without the hierarchical structure in the features. This suggests that the hierarchical features provide a potential for much better portability.

In this paper, we propose a document re-ranking method for Chinese information retrieval where a query is a short natural language description. The method bases on term distribution where each term is weighted by its local and global distribution, including document frequency, document position and term length. The weight scheme lifts off the worry that very fewer relevant documents appear in top retrieved documents, and allows randomly setting a larger portion of the retrieved documents as relevance feedback. It also helps to improve the performance of MMR model in document re-ranking. The experiments show our method can get significant improvement against standard baselines, and outperforms relevant methods consistently.

In this paper we propose an NP coreference resolution system which does resolution on the entity-level. The framework of the system is presented and different resolution strategies are investigated.

In this paper we propose to use a semi-supervised learning algorithm to deal with word sense disambiguation problem. We evaluated a semi-supervised learning algorithm, local and global consistency algorithm, on widely used benchmark corpus for word sense disambiguation. This algorithm yields encouraging experimental results. It achieves better performance than orthodox supervised learning algorithm, such as kNN, and its performance on monolingual benchmark corpus is comparable to a state of the art bootstrapping algorithm (bilingual bootstrapping) for word sense disambiguation.

Analyzing the semantic representations of 10000 Chinese sentences and describing a new sentence analysis method that evaluates semantic preference knowledge, we create a model of semantic representation analysis based on the correspondence between lexical meanings and conceptual structures, and relations that underlie those lexical meanings. We also propose a semantical argument-head relation that combines ‘basic conceptual structure’ and ‘Head-Driven Principle’. With this framework which is different from Fillmore’s case theory (1968) and HPSG among other, we can successfully disambiguate some troublesome sentences, and minimize the redundancy in language knowledge description for natural language processing.

In this paper, we propose a document reranking method for Chinese information retrieval. The method is based on a term weighting scheme, which integrates local and global distribution of terms as well as document frequency, document positions and term length. The weight scheme allows randomly setting a larger portion of the retrieved documents as relevance feedback, and lifts off the worry that very fewer relevant documents appear in top retrieved documents. It also helps to improve the performance of maximal marginal relevance (MMR) in document reranking. The method was evaluated by MAP (mean average precision), a recall-oriented measure. Significance tests showed that our method can get significant improvement against standard baselines, and outperform relevant methods consistently.